A computing enthusiast has shared images of a fat stack of DDR4 RAM sticks that they found at home. VastOption8705 told other tech fans on Reddit that they “found 10k worth of RAM sitting in my garage.” We count 30 sticks of DDR4-2666, and three sticks shown separately are all 64GB modules. If they are all 64GB modules, that number of sticks would add up to 1,920GB of RAM…

How did this huge amount of RAM end up in VastOption8705’s garage? The story goes that the Redditor was allowed to keep an old server from their workplace. Such 'generosity' isn’t unheard of, and it might have actually saved the company some recycling fees.

Is server DDR4 useful?

Numerous comments on the Reddit post jokingly pitch for RAM stick donations. But they might be barking up the wrong tree. Looking more closely at the sticks, where the labels are clearly readable, reveals that they are Samsung 64GB DDR4 ECC registered DIMMs (RDIMMs).

These kinds of RAM sticks are commonly used in servers built around Intel Xeon, AMD Epyc, or AMD Threadripper platforms. Sadly for their market value and the folks on Reddit hoping for some spare RAM, they aren’t consumer PC friendly. Even if you have a DDR4-supporting motherboard, these ECC RDIMMs will be both physically and electrically incompatible with your home creator/gaming PC.

Pricing and the Australia question

A quick poke around eBay U.S. listings shows that a single 64GB DDR4-2666 stick like those shown can be bought for between $350 and $425. So, we could roughly value the 30 sticks at $12,000 if a price of $400 per stick were achieved. The OP’s touted $10K valuation wasn’t far off. But if VastOption8705 lives in Australia, as they seem to, and guesstimated the value in Australian dollars, AU$10,000 currently equates to ~US$7,200 – quite low.

If you have spare RAM in your garage, don't share your address on public forums. The best I could find is a pair of 8GB DDR4 SO-DIMMs in my drawer. I also had some 30-pin SIMMs, SDRAM, PowerMac ROMs, and DDR3 modules probably worth just pennies, despite the RAMpocalypse.

Earlier this year, V-Color announced its 1+1 DDR5 filler memory kits to help suspend disbelief and offer psychological salvation against rising memory prices. These kits mix real DDR5 sticks with dummy modules that don't actually have any RAM inside — just working RGB — to give you the illusion of a fully-slotted motherboard. Today, the kit is finally available to buy in the UK and Europe, with just the filler sticks on their own also listed for sale in North America.

V-Color has two different DDR5 products: Manta XSky and Manta XFinity, the latter being the higher-end model. The 1+1 kit is only available for the Manta XSky variant and only in the UK and Germany; not yet in the US. The 0GB kits (where both sticks are dummies) are available in both the Manta XSky and XFinity models across multiple regions. Pricing starts from $43 and goes up to almost $300, so let's break it all down.

In the UK, Overclockers currently has the 16GB 6,000 MT/s CL30 kit listed for 210 GBP, or $285, in black and white colorways. You get a single 16GB stick with those specs and a filler module to occupy two slots on your motherboard. There's a slightly more expensive 16GB 6,000 MT/s CL28 kit as well, selling for 219 GBP, or $297. Once again, you can choose either the black or white color, and you get 1 real stick of RAM and one filler.

• V-Color (1+1) Manta XSky RGB 16GB C30 6,000MHz - Black
• V-Color (1+1) Manta XSky RGB 16GB C30 6,000MHz - White
• V-Color (1+1) Manta XSky RGB 16GB C28 6,000MHz - Black
• V-Color (1+1) Manta XSky RGB 16GB C28 6,000MHz - White

In Europe, using Geizhals, we were only able to find one in-stock listing of the 16GB 6,000 MT/s CL28 kit for 330 Euros, or $384, in the black colorway. The same kit with CL30 speeds can be had for 220 Euros, or $256, but it's out of stock on CaseKing, the one Germany-based store in the entire EU with these products. We'll remind readers that there are no North America listings for V-Color's 1+1 DDR5 kits as of now.

But we did spot plenty of 0GB kits on Newegg and Amazon. These kits don't offer any real RAM whatsoever and are purely meant for decorative purposes. You can only use these if you already have working RAM in your system and two extra slots left to populate on your motherboard. Amazon has the Manta XFinity variant listed for $57 in either black or white, while Newegg is selling the Manta XSky model for $44 in white.

• V-Color Manta XFinity DDR5 RGB Filler Solution - Black (Amazon)
• V-Color Manta XFinity DDR5 RGB Filler Solution - White (Amazon)
• V-Color Manta XSky DDR5 RGB Filler Solution - White (Newegg)

CaseKing in Germany is selling the Manta XSky filler kit for 50 Euros, or $58, but we weren't able to find any other listing across Europe. The prices for these kits across the world are not much higher than their regular, non-filler counterparts, so you might as well spring for these until the RAMpocalypse settles down. Just don't forget that you'd still be limited to single-channel memory with the 1+1 kits, which isn't that big of a deal when it comes to DDR5, especially if you have an X3D CPU.

The saga involving Chinese DRAM maker CXMT's alleged theft of Samsung's trade secrets is going strong. The South Korean court case already includes multiple convictions, two of which carry prison sentences for ex-Samsung engineers. The latest chapter is a doozy, though. Korean publication NoCut News spilled the chips on Project Hefei, a purported CXMT roadmap outlining long-term planning about said technology "acquisitions," personnel poaching, and production tape-out — all key pieces that may have directly led to CXMT's ascension to 10% of the global DRAM market.

According to leaked court documents, the prosecution says that Project Hefei was CXMT's entire DRAM development plan and was spearheaded by the firm's head of development (formerly Samsung's DRAM development lead), around August 2016 — not much longer after CXMT itself was created in June 2016.

In brief, the purported plan was to nab Samsung's Process Recipe Plan (PRP) by September 2016, poach key Samsung engineers by October 2016, have R&D complete in July 2017, and start making DRAM wafers by August 2018 at a rate of 10,000 a month. NoCut says the PRP dataset comprises 620 steps in DRAM manufacturing and includes data on equipment, consumables, and production methods.

The report states that in August 2016, CXMT first attempted to make wafers of 18nm chips by relying on the collective memories of the Samsung engineers it had hired away. Those recollections apparently proved insufficient, so after allegedly gaining illicit access to Samsung's PRP, CXMT prepared its own document in September 2016. The leaked data even included specific equipment suppliers and model numbers.

CXMT's "new" PRP was then handed out to key specialists, many of them ex-Samsung engineers, whom the prosecution says ought to have immediately recognized the data as originating from the Korean firm. The document apparently included notation and notes on process developments that were all unique to Samsung. A convicted ex-Samsung researcher with the surname Jeon, previously sentenced to 7 years in prison for manually copying parts of the PRP before leaving for CXMT, testified in this case as a witness.

The whole CXMT debacle has been playing out in South Korean courtrooms since January 2024 and is arguably far bigger than just "a company stole some tech from another." A decade ago, most of the DRAM market was taken by the Big Three: Samsung, Micron, and SK hynix. CXMT was created in June 2016 in Hefei (hence the project name), with a modest government investment of around $1.9 billion USD, allegedly with no R&D facilities whatsoever or any plan for research.

Going from zero facilities and institutional expertise to DRAM wafer production in little over two years would be an unprecedented feat, and presumably nigh impossible without the alleged IP theft; getting just the memory fab up and running usually takes two to four years, let alone any time for research. The firm claimed in 2019 that it designed its then-new 8 Gb DDR4 chips entirely in-house as a "leapfrog, independent" technology and began selling DRAM chips locally.

Then the AI locusts charged in and ate every chip on the planet. This demand resulted in explosive growth for CXMT, from an estimated 4% of the global DRAM market in Q2 2025 to around 10% in Q2 2016, marking the first time in over a decade that the Big Three held less than 90% of the pie.

Production capacity expanded to three 12" wafer fabrication facilities, expected to churn out a collective 350,000 wafers until the year is done — close to the same figure that Micron produces, of 375,000 to 385,000. CXMT is also planning to open a second fab in Beijing, aiming to produce over 600,000 wafers per month once it's online.

Furthermore, CXMT's gains aren't coming just from its DRAM market share. Revenue grew 716% in Q2 2026 alone, and its IPO on the Shanghai STAR Market in July 2026 saw its stock climb 466%, netting the firm a cool $8.6 billion USD and making it the most valuable chipmaker in the Chinese stock market.

About 70% of that money is reportedly going towards further expansion of DRAM production, rather than pricier, more complicated HBM. However, the firm has supplied HBM3E samples to Alibaba's T-Head and Cambricon, even as Samsung and SK hynix enter HBM4 production.

As of the South Korean prosecution's last tally in December 2025, Samsung's damages due to CXMT's alleged machinations ascended to "at least tens of trillions of won." A back-of-the-envelope extrapolation, considering quite a while has passed, might suggest a figure in the range of ₩40 trillion, or $29.5 billion, and rising exponentially.

The lasting impact on the memory market and technology in general is going well beyond plain number descriptors, though. All things considered, if the allegations are true, then CXMT's machinations resulted in a geopolitical-scale economic event. Dr. Evil would be proud.

The goal of next-generation HBM5 memory specification is to double performance and increase power efficiency compared to HBM4E, Samsung announced at the 'Memory Executive Summit,' an event that precedes 'Semicon Taiwan 2026.' Considering that it is barely realistic to double the data transfer rate of HBM4E in just one generation, the comment made by Samsung may imply that HBM5 will double the number of interface pins to boost bandwidth.

The goal of HBM5, which is currently under development, is to double the performance and improve the performance per watt by 20% compared to the HBM4E generation, according to Choi Jang-seok, the head of the product planning team of the memory business department of Samsung Electronics DS division. Previously, Samsung announced that its HBM5 memory stacks will feature a heat path block (HPB), which will reduce thermal resistance by 20% and simplify cooling of HBM5 modules.

Doubling HBM5’s per-stack memory bandwidth would result in peak bandwidth of around 4 TB/s per stack already in 2028 – 2029. TSMC expects AI accelerators to get to 20 – 24 HBM5/HBM5E per package configurations by the end of the decade, which means that these systems-in-packages will get a whopping 80 TB/s – 96 TB of memory bandwidth just several years down the road.

Although both DRAM makers like Micron, Samsung, or SK hynix as well as developers of HBM controllers and PHYs like Cadence, Rambus, or Synopsys already offer HBM4/HBM4E controllers and interfaces rated for 16 GT/s data transfer rates, the official JEDEC transfer rate for HBM4E will be around 12 GT/s.

If Samsung expects HBM5 to deliver twice the bandwidth of HBM4E, the HBM5 specification must either double the per-pin data transfer rate from 12 GT/s to 24 GT/s, double the interface width from 2,048 to 4,096 bits, or combine a wider interface with a higher data transfer rate. Increasing HBM5 memory stack interface width to 4,096 bits has so far been envisioned by KAIST and Marvell; however, this is certainly not an even semi-official confirmation of the specification's target.

From a performance-per-watt perspective*, widening the interface is generally easier than doubling the per-pin signaling rate. Higher per-pin speeds require faster drivers, receivers, clocks, equalization, tighter timing margins, and a more sophisticated PHY since maintaining signaling integrity at speeds well beyond 20 GT/s is not easy. But while a wider HBM interface moves more bits in parallel at a lower speed per wire, going from 2,048 to 4,096 pins means twice as many TSV/I/O paths, drivers, receivers, bumps, more complicated routing, and extremely complicated base die. So while a 4,096 I/O at moderate speed is probably best for pJ/bit, the interface/package complexity gets so extreme that it may offset the gains. Furthermore, such a wide interface Perhaps a 3,072-bit interface combined with a moderate increase in data transfer rate would be a reasonable engineering compromise, though engineers involved in JEDEC’s decision-making process may think otherwise. In any case, for now, any discussion of HBM5 specifications remains speculative.

Nonetheless, the target to double the bandwidth of HBM4E (2 TB/s per stack) within one generation is clearly a very aggressive one.

*It should be noted that achieving a 20% higher energy efficiency can be achieved not only by increasing performance, but by improving DRAM process technology, optimizing base die, lowering TSV I/O voltages, architectural changes, or power management.

China's DRAM champion ChangXin Memory Technologies (CXMT) has started risk production of HBM3E memory, The Information reports. Although CXMT remains a generation behind the big three memory manufacturers, which are mass producing HBM4, it goes without saying that reaching the HBM3E milestone highlights the company's rapid technological progress.

Specifications of CXMT's HBM3E products are unknown, though keeping in mind that we are dealing with risk production, specifications of actual HBM3E products from CXMT may differ from risk production samples. Meanwhile, general JEDEC specifications are well known: HBM3E modules feature a 1,024-bit-wide memory interface, supports data transfer rates up to 9.6 GT/s per pin, and can stack 8 or 12 memory devices. Depending on the exact speed bins, HBM3E can provide up to 1.228 TB/s of memory bandwidth.

Capacity of actual memory stacks depends on the number of DRAM dies used in that stack: HBM3E products can offer 24GB of capacity using an eight-die stack or 36GB using a 12-die stack when built with 24Gb DRAM devices.

Several Chinese developers of advanced processors are already evaluating CXMT's HBM3E with their processors, including Alibaba Group's T-Head and Cambricon Technologies, according to the report. If testing and qualification proceed as planned, these companies could begin using CXMT's HBM3E in commercial products as early as next year.

Since every HBM package requires multiple large DRAM dies, growing HBM output could consume considerable DRAM manufacturing capacity, which is when CXMT's aggressive capacity expansion will be useful.

CXMT's manufacturing of HBM3E is important for multiple reasons and arguably the HBM3E generation itself matters less than CXMT's ability to manufacture usable HBM at all.

Firstly, HBM is one of the critical components of modern AI accelerators, so if CXMT's HBM3E qualifies with processors from Cambricon, T-Head, and other Chinese designers, China becomes less dependent on Micron, Samsung, and SK hynix for building high-performance AI hardware.

Secondly, despite being a generation behind HBM4, HBM3E is still very capable memory and can provide several TB/s of bandwidth, sufficient for powerful AI accelerators. In fact, many Chinese developers of AI accelerators do not necessarily need HBM4/HBM4E to build useful AI systems.

Thirdly, HBM is substantially harder than making ordinary DRAM. CXMT needs not only competitive DRAM dies, but also high-yield stacking, through silicon vias (TSVs), very fine interconnects, thermal management, packaging, and testing. Reaching HBM3E risk production proves that China's memory ecosystem is advancing beyond simply manufacturing commodity DRAM.

Finally, there is also an important geopolitical angle. Export restrictions constrain China's access to advanced AI processors and HBM, so a combination of Chinese-designed accelerators, CXMT HBM3E, and advanced domestic packaging indicates that China becomes one step closer to semiconductor self-sufficiency.

G.Skill, which makes some of the best RAM on the market, has announced the Flare X5X series of DDR5 memory kits. Like the Trident Z5 NeoX, the new Flare X5X leverages AMD's EXPO ULL (Ultra Low Latency) technology, which features optimized memory subtimings to maximize performance on Ryzen processors, especially those without 3D V-Cache.

G.Skill offers a plethora of different DDR5-6000 memory kits across its Trident, Ripjaws, and Flare lineups. It also offers variations that specifically cater to Intel and AMD processors. Historically, the Flare series targets AMD platforms.

The Flare X5X will initially launch at DDR5-6000, since many consider it the sweet spot for AMD's Ryzen chips. Like the other offerings, the Flare X5X debuts with a 32GB memory kit, made up of two 16GB memory modules.

As with any AMD EXPO ULL memory kit, it is difficult to tell the Flare X5X apart from the vanilla Flare X5 since we are not accustomed to listing specifications beyond the main memory timings. In the case of the Flare X5X, the memory kit has its timings configured to 36-36-36-96 at 1.35V, the same as the normal Flare X5 memory kit. The memory subtimings are tighter on the Flare X5X because of AMD EXPO ULL.

G.Skill Flare X5X Specifications

With memory timings of 36-36-36-96, the Flare X5X is not the best AMD EXPO ULL memory kit G.Skill offers. The company sells more optimized variants, including 26-36-36-32, 28-36-36-32, and 30-38-38-32 at 1.45V, 1.40V, and 1.35V, respectively. It is plausible that G.Skill will offer similar configurations for the Flare X5X down the line.

When AMD first announced it, the chipmaker confirmed that EXPO ULL memory kits would be priced similarly to regular EXPO memory kits. However, because of the memory shortage and leftover stock, the price difference was substantial. Pricing for AMD EXPO ULL memory kits has somewhat stabilized over the last few months, so the price tags are falling in line with what AMD anticipated.

G.Skill has not shared the pricing for the Flare X5X DDR5-6000 C36 32GB (2x16GB) memory kit. However, we should have a pretty good idea of what it will retail for by looking at G.Skill's other memory kits on the market. For instance, the Trident Z5 NeoX RGB DDR5-6000 C36, a less optimized version of the C30 version we have reviewed, sells for $10 more than the normal Trident Z5 RGB DDR5-6000 C36. In a perfect world, the Flare X5X DDR5-6000 C36 may sell for around $509.99, $10 above the standard Flare X5 DDR5-6000 C36.

G.Skill will roll out the Flare X5X series in phases so that availability will vary by region. It should not be long before we find the Flare X5X DDR5-6000 C36 memory kit at all major U.S. retailers.

ChangXin Memory Technologies (CXMT) has announced in a social media post that it has started mass production of LPDDR6 memory for a Xiaomi smartphone, Reuters reports. This is not entirely unexpected given the general LPDDR6 mass-production timeline, but the announcement is still surprising because two China-based companies are ahead of industry announcements. Yet, there are caveats.

LPDDR6 is perhaps the most substantial upgrade to the GDDR-class memory ever. Firstly, it uses PAM3 signaling, which encodes 3 bits of information across two symbols, compared with the NRZ signaling used by LPDDR5/LPDDR5X. Second, the data transfer rate increases to 43.2 GT/s. Third, the reduced I/O width from 32 bits to 24 bits requires rework at both the hardware and software levels.

Given the differences between LPDDR5 and LPDDR6, the most important factor here is not exactly LPDDR6 manufacturing, but a broader view.

On the one hand, CXMT reached mass production of LPDDR6 roughly at the same time as the established DRAM leaders, meaning it is on par with them, if not ahead. This marks a substantial change from earlier generations, when Chinese DRAM technology generally lagged considerably behind Samsung, SK hynix, and Micron.

On the other hand, reaching mass production does not necessarily mean that CXMT has achieved full technological or manufacturing parity with its larger rivals. The official announcement covers only one smartphone maker and one model, meaning CXMT's LPDD6 still has to pass validation for other handsets and other makers.

Still, the significance of CXMT's announcement is hard to understate. CXMT is no longer merely catching up by introducing a new memory generation years after its international competitors. With LPDDR6, CXMT is entering the market during the standard's initial commercialization, which means the technology gap between China's leading DRAM producer and the industry's established players has narrowed considerably.

Which brings us to the question of whether CXMT can maintain this pace. Developing a competitive LPDDR6 device is one thing, producing it at high yields and in sufficient volumes to supply major customers is another. For now, we are talking about a niche smartphone with memory operating at undisclosed performance.

If CXMT can mass-produce LPDDR6 for the whole market and manufacturers, we may well have seen the change of the industry leader. Yet, given one niche device, we certainly cannot say so for now.

China's leading DRAM producer ChangXin Memory Technologies (CXMT) has filed a lawsuit against the U.S. Department of Defense in an attempt to overturn its classification as a company linked to China's military, Reuters reports. CXMT claims that it has no connection to the Chinese People's Liberation Army, and argues that the Pentagon's decision lacks supporting evidence, violated due-process requirements, and has damaged both its business and reputation.

The Pentagon initially classified CXMT as a Chinese military company in January 2025, when Joe Biden was the president. The Trump administration subsequently maintained the designation when the Defense Department updated the list in June. CXMT says it spent more than a year supplying information to the Defense Department in an attempt to overturn the classification and remove itself from the list.

The company claims the DoD published a notice in February indicating that CXMT would be taken off the list but withdrew that notice later the same day without any explanation. In June, the Pentagon classified CXMT as 'directly affiliated with MIIT and indirectly affiliated with SASAC and MIIT.' MIIT is China's Ministry of Industry and Information Technology, whereas SASAC is the State-owned Assets Supervision and Administration Commission of the State Council, an organization that formally owns the government's stakes in non-financial state-owned enterprises.

While CXMT memory can certainly be found in systems used by the PLA or China's secret services, this does not automatically make CXMT a direct supplier to these entities and therefore does not prove that CXMT is a Chinese military company.

CXMT points out the Pentagon's original motivation to include the DRAM maker in its list was a remarkably weak-looking 'military grade' argument. According to the complaint, DoD's 2024 report said CXMT contributed to China's defense industrial base partly because its DRAM is dual-use and cited an advertisement describing CXMT products as 'military grade.' CXMT says that advertisement came from an unauthorized distributor, which later admitted the characterization was erroneous.

In addition, CXMT claims that its DRAM products conform to JEDEC commercial standards and lack the requirements imposed on Chinese military-grade semiconductor components involving areas such as temperature, packaging, testing, and military oversight. Also, it says that should its products be made in the U.S., they would be classified as for 'purely civilian uses' and that it 'does not hold any licenses to manufacture products for military use.'

Inclusion on the Pentagon's list of companies linked to the Chinese military can result in restrictions involving U.S. government contracts, something that CXMT barely seeks. But more importantly, it can affect a company's commercial relationships (as many companies do not want to be affiliated with companies linked to the Chinese military) and reputation. Indeed, CXMT claims that since January 2025 the designation has continuously hurt it commercially and reputationally, and that the lawsuit is intended to protect its business interests.

The ongoing DRAM crisis has led to a sharp increase in laptop prices, with manufacturers facing exponentially high component costs, particularly memory and SSDs. The situation has created a growing gap in the affordable gaming laptop segment, with most entry-level models becoming increasingly expensive. In its response, MSI has introduced the new Katana 15 HX C14 that features Intel’s 14th-gen Raptor Lake Refresh processors along with support for both DDR5 and DDR4-3200 memory.

Depending on the region, MSI will be offering the value-focused gaming laptop in two variants, featuring dual SO-DIMM slots with either DDR4 or DDR5 compatibility. For context, a 32GB (2x16GB) DDR4-3200 SO-DIMM memory kit can be purchased for around $200, while a DDR5-6000 kit goes for over $400. MSI did not specify the DDR4 capacity as standard, only that the laptop supports up to 96GB, which seems unlikely given no 48GB DDR4 SO-DIMMs exist.

Configurable with up to a 24-core Intel Core i9-14900HX processor, the laptop should offer plenty of CPU performance for demanding tasks besides gaming. It will be available with up to an Nvidia RTX 5070 laptop GPU offering support for the latest AI features, GDDR7 memory, DLSS 4.5, and real-time ray tracing. MSI claims that the Katana 15 HX C14 is tuned for up to 170W of combined CPU and GPU power, while the cooling system with dual fans, three heat pipes, and four exhaust outlets helps in sustained performance. There are also dual-PCIe Gen 4x4 M.2 slots, allowing gamers to add a separate SSD dedicated to their gaming library.

Display options include a 15.6-inch IPS panel available with either QHD (2560x1440) resolution that supports up to a 165 Hz refresh rate or FHD (1920x1080) resolution with up to a 144 Hz refresh rate. In terms of design, the laptop comes in a stealthy black finish with textured slash patterns inspired by traditional Japanese blades along with bright red logos. The chassis measures 22.9mm in terms of thickness and weighs 4.62 pounds, making it fairly compact for a 15-inch gaming notebook.

Other notable features include support for Wi-Fi 6E and Bluetooth 5.4, a 4-Zone RGB gaming keyboard with 1.7mm key travel, a 240W power adapter, USB Type-C with Power Delivery 3.0, a 60Whr battery, and DTS-enhanced dual 2W speakers. According to MSI, the new Katana 15 HX C14 will be available globally, although there's no pricing information at the moment.

Cloudflare says that it has freed up roughly 100TB of RAM across its global fleet without reconfiguring any physical RAM modules in its servers, and that it did so by redesigning how each DNS cache entry is laid out in memory. In a technical blog post, systems engineer Sebastiaan Neuteboom gave us an in-depth look at five Rust-level changes to Big Pineapple, the platform behind the 1.1.1.1 resolver, that shrank each cached entry from 953 bytes to 420 bytes and made the cache faster in the process, with insert throughput up 43% and lookup latency down 19%. The recovered memory matches the combined DDR5 in 130 of Cloudflare's 768GB Gen 13 servers, reclaimed in the face of server-grade DDR5 prices that are on track to double year over year.

Just one wasted byte per entry costs Cloudflare more than 250GB of fleet memory, which is a consequence of Big Pineapple holding more than “250 billion cached DNS entries at any given time.” The first of the five changes, replacing Rust's growable Vec and String containers with fixed-size boxed slices, saved over 15TB on its own by dropping capacity fields the data never uses once cached.

Further changes collapsed the three record lists in every DNS response into one buffer indexed by 2-byte offsets, and dropped owner names that duplicate the queried domain, rebuilding them at read time instead. The last change stores record data as length-prefixed raw wire-format bytes, ending an arrangement in which a 4-byte A record took up the same 144 bytes as the largest record type Cloudflare caches, the rarely seen NAPTR.

Packing record data contiguously also improved CPU cache locality, meaning that most record types now copy straight from the stored buffer into outgoing responses instead of being re-serialized field by field.

According to Cloudflare’s benchmarks in the technical blog, insert throughput is now 893,000 entries per second, up from 625,000, with lookup latency down from 828 to 670 nanoseconds. Over the rollout, which ran from mid-May to early July, p99 resident memory per instance fell from 9.3GB to 5.3GB.

The Gen 13 servers, which Cloudflare calls its most powerful yet, each carry 768GB of DDR5-6400. The company settled on that config back in March after pricing out a 1,152GB option and rejecting it partly because of high memory prices.

This DNS cache work is the second large memory-reclamation project Cloudflare has completed in the last year, following the FL2 rewrite of its request-handling layer in Rust last September. Efficiency has been driving the company’s hardware picks for years, including the switch to AMD's 96-core EPYC 9684X for its Gen 12 servers.

None of the liberated RAM will be cashed out as smaller memory configurations, though; Cloudflare plans to pour it back into larger DNS caches, which raise hit rates and cut the query traffic it sends upstream to authoritative servers.

High-bandwidth memory (HBM) is one of the key components of AI systems that is not currently assembled in the U.S., but this is going to change in 2029, when SK hynix initiates packaging of HBM memory modules at its facility in Indiana. This week, the company held a groundbreaking ceremony for its HBM production base in America and intends to start construction of the plant shortly.

SK hynix's HBM production site in West Lafayette, Indiana, will specialize in advanced packaging, testing, and R&D of HBM. The facility is set to cost the company around $4 billion, which highlights its advanced technology capabilities as well as vast capacity. The DRAM maker expects to open the cleanroom by October 2028 and begin volume production of next-generation HBM in the second half of 2029, which is about a year later than originally envisioned. Once commercial operations kick off, SK hynix expects the Indiana site to employ around 1,000 people.

Despite being described as an HBM production base, the Indiana site will not manufacture the DRAM wafers, but will package HBM stacks using DRAM wafers produced in South Korea and base dies produced in South Korea, Taiwan, or the U.S. Nonetheless, SK hynix intends to market the resulting devices as its first U.S.-made next-generation HBM products. Meanwhile, as custom HBM-class memory gains traction in the second half of the decade, assembling such stacks close to American customers may be particularly important both for SK hynix and its clients, as it enables SK hynix to shrink the feedback loop significantly.

In addition to building the advanced packaging and testing plant, SK hynix will also establish an Advanced Packaging R&D Testbed next to the manufacturing lines. The facility will enable SK hynix to work with customers, universities, and commercial partners on future packaging technologies, build prototypes, and quickly validate their performance and manufacturability in an actual production environment. In the era of customized HBM, such a facility could become a significant competitive advantage for SK hynix.

SK hynix also signed a memorandum of understanding with Purdue University to jointly research system integration and advanced packaging technologies, including HBM.

"The United States is the epicenter of AI innovation, bringing together premier customers, top-tier R&D capabilities, and partners and Indiana fab will become a gateway to deliver first Made in USA HBM products," said Kwak Noh-Jung, CEO of SK hynix. "We will expand our investments and collaboration in the U.S., grow together, and become the most trusted partner in shaping the future of AI in America."

Nvidia's NVLink Fusion program gives the company's partners the building blocks necessary to connect custom chips with the NVLink scale-up domain used to join many separate processors into a single coherent system like the Vera Rubin NVL72 rack-scale accelerator. Today, Nvidia is adding a new building block to that toolkit: NVHBM, a custom implementation of the high-bandwidth memory that underpins practically every AI accelerator in use today.

As Nvidia describes it, NVHBM is a custom HBM base die that promises higher bandwidth, lower power usage, and a smaller on-die footprint than traditional HBM4e. Nvidia says it's designed and validated with "leading memory vendors," so it promises custom silicon developers faster time-to-market than implementing commodity HBM from the ground up. But it's worth re-emphasizing that this isn't an HBM replacement. Instead, it's a new building block that Nvidia is only offering to its custom silicon partners.

Memory bandwidth is everything for AI accelerators, and NVHBM promises up to 30% higher bandwidth per stack than standard HBM4e. For memory-bandwidth-bound AI workloads, that higher bandwidth translates into higher throughput, such as a higher tokens-per-second rate for AI inference.

The custom NVHBM base die also reduces the footprint of memory-related circuitry on the main custom accelerator die. Traditionally, the HBM memory controller has been incorporated into the primary silicon die on the package. NVHBM instead moves the memory controller into the base die of the HBM stack and provides a smaller custom PHY that NVLink Fusion customers can then integrate into their designs.

Nvidia says this approach frees up precious package real estate that can then be used for additional compute die area — up to 30% more compute on the primary silicon die. NVHBM further promises to simplify the interposer routing used to join multiple chips together for designs using advanced packaging techniques.

NVHBM also provides power savings versus off-the-shelf HBM4e stacks. As Nvidia has continuously hammered home in the Vera Rubin roll-out, every watt that isn't going into token production is a watt wasted. Nvidia says NVHBM uses 15% less power than commodity HBM4e, and that power savings can be banked for performance-per-watt improvements, reallocated into more functional units for a custom accelerator design, or translated into higher sustained performance within the same power budget.

Higher bandwidth at lower power is a huge win for AI accelerators that are moving massive data structures like model weights and KV caches around, especially when those savings are multiplied across many thousands of chips. The energy saved on data movement can be plowed back into higher performance from the accelerator itself or reallocated to support larger numbers of accelerators within the same fixed power envelope.

But these are, as of now, reasons for Nvidia's prospective partners to consider incorporating NVLink Fusion and NVHBM into their custom designs, not benefits that will materialize in the Rubin rack-scale systems already in production.

Along with NVHBM itself, Nvidia announced that Amazon's Annapurna Labs will be its first partner on NVHBM. , and Annapurna VP Nafea Bshara says: “We look forward to this technology collaboration to benefit future AWS infrastructure designs.” Annapurna's next-generation Trainium 4 AI chips will already support the NVLink Fusion scale-up interface, so it seems likely that follow-on chips will support NVHBM, too.

Looking back in time is always interesting, especially when visiting any topic that's technology-adjacent — and doubly so when looking at the evolution of pricing. The price of memory continues to be a hot-button topic, and an eagle-eyed Redditor came across a 1995 press release from Advantage Memory Corporation announcing prices for its freshly released DIMMs, revealing that fitting a high-end machine with 1 GB of RAM would set you back as much as a cheap house.

The figures on display are rather shocking when converted to today's dollars, with 8 MB of RAM going for $399, or $874 in today's money, and a 64 MB stick ringing in at $3,900, or $8,540 in 2026. The Apple Power Macintosh mentioned in the PR had 12 slots, so it could fit in as much as 1.5 GB of RAM, and although the pricing for 128 MB DIMMs wasn't mentioned, those would be around $8,000 a pop, so "only" 1 GB of RAM would come in at $64,000 — enough for a house back then in the more affordable areas of the U.S. By the way, that amount would be $140,000 today.

In 1995, memory technology was transitioning from SIMMs to the spankin' new DIMMs, and Command and Conquer, MechWarrior 2, and the original Need for Speed were coming out. 16 MB of RAM was generally accepted as a decent amount of memory for a build, roughly matching the 16 GB of today. Some quick math says those 16 MB of the new technology were about $800, or in today's money, the princely sum of $1,751.

As of today, 16 GB of DDR5 will set you back $200 to $250, but the price of memory has been extremely volatile, to say the least. Plus, not only is supply rumored to drop, server RAM contract prices are expected to rise, and the few crumbs left for the consumer market are still rising in price.

SK hynix chairman Chey Tae-won even suggested that DRAM wafer shortages could last until 2030 at least. The fact that all the major RAM makers are shifting their production lines to HBM for AI accelerators is only making the problem worse. Yes, $200 is still a long way from $1,750, but with things keeping in this direction, there's a non-zero chance this very article will be linked to in the near future.

At any rate, despite the poor outlook, these figures reveal just how much technology really became democratized with time. Any form of PC was really expensive back in 1995. Around that era, Bill Gates said that there would be "a PC at every home, and at every desk." That statement sounded a little insane back then, but in retrospect wasn't crazy enough, as today we have what would be sci-fi hardware... in everyone's pocket or purse.

When it comes to building a gaming PC, DDR5 RAM clocked at speeds of 6,000 MT/s is often touted as the sweet spot for specs. While slower and faster speeds are available, it's very much the benchmark for both price and performance, which is why it's noteworthy that this particular configuration now costs $400, a more than 400% increase from the low prices we've seen in the past.

• $401 for Corsair Vengeance 32GB DDR5 600 ($215 off)

In addition to our ongoing RAM price tracking, we perform daily spot checks on pricing for discounts or deals for 32GB (2x16GB) of DDR5 RAM, given its aforementioned significance as the "benchmark" when it comes to capacity. In recent weeks, the price of 32GB DDR5 6000 RAM has been creeping ever closer to that unfortunate $400 milestone, and in the last few hours we've just crossed the threshold. That's not to say that prices might fluctuate back down in a day or two, but generally the traffic is only moving in one direction.

You can buy 32GB of DDR5 RAM at slower speeds for less — this Corsair Vengeance DDR5 4800 is "just" $359 — but good RAM is now harder (and more expensive) to come by than ever before.

According to data from PCPartPicker, this particular 32GB DDR5 6000 kit from Corsair has been as little as $80 in years gone by at Newegg, marking a full 400% price increase on the lowest we've ever seen on this model.

RAM prices elsewhere have climbed by as much as 500%, with 128GB of DDR5 now costing $3,399 in stores. The causes of the PC component crisis are numerous and well-documented. AI buildouts are consuming much of the world's NAND and flash manufacturing capacity, which is having a knock-on effect for consumer production. Famously, Micron even wound up its consumer brand Crucial in favor of serving more enterprise customers.

All signs point to the memory shortage, and thereby pricing, getting worse in the coming months. As such, bundles and pre-built PCs remain one of the only ways to insulate yourself from cost increases if you're in the market for new hardware. As mentioned, slower memory speeds are available at lower prices, but compromising for 4,800 MT/s RAM to save $40 when building a PC is an extremely tough state of affairs.

Nvidia has told some of its largest customers that the prices of servers containing its AI chips will rise by more than 15% in many cases, Bloomberg reported on Saturday. The increases will take effect on Grace Blackwell and Vera Rubin systems shipping early next year, according to people familiar with the matter, who commented on communications that have not yet been made public. The size of each increase will depend on the chip generation and the memory configuration involved. Companies that build servers under contract for large data center operators, including Microsoft, Google, and Oracle, have recently notified their customers of the forthcoming increases, the people told Bloomberg.

This is another example of the so-called "RAMageddon" that's gripping the DRAM market, with contract prices having risen at record rates this year. Analysts projected that conventional DRAM contract prices would climb 58% to 63% quarter-over-quarter in Q2 2026, following a Q1 surge of 90% to 95%, as suppliers reallocated capacity toward HBM and server products. SK hynix said in October last year that it had already sold out its entire 2026 memory production capacity, and Samsung and SK hynix raised 2026 HBM3E supply prices by close to 20% before the year began.

AI systems carry enormous memory loadouts, with Nvidia's Rubin GPU shipping with up to 288GB of HBM4 per package, and the NVL72 rack-scale system combines 72 of those GPUs, putting more than 20TB of HBM in a single rack before accounting for the LPDDR attached to its Vera CPUs. With HBM production consuming roughly four times the wafer area of equivalent conventional DRAM, memory has become one of the largest line items in an AI server's bill of materials, and it's continuing to rise at a stratospheric pace.

Ironically, the supply crunch that's now inflating Nvidia's systems is the same one its demand helped to create. The three major memory makers spent this and last year shifting advanced nodes and new capacity toward HBM and high-capacity server DRAM, starving commodity markets in the process. Consumer DDR5 pricing has more than doubled since late 2025 as a result, with a mainstream 32GB DDR5-6000 kit selling for around $392 in August against $110 to $140 a year earlier, according to our RAM price tracker.

Nvidia has already passed rising costs through to consumers, raising prices on GeForce graphics cards earlier this month. The Bloomberg report indicates the same unrelenting pressure has now reached the top of the Nvidia stack, where hyperscalers as well as PC builders will be absorbing the increase. A 15% rise on rack-scale systems that sell for several million dollars each adds hundreds of thousands of dollars per rack across deployments that run to thousands of racks.

Nvidia runs a gross margin of roughly 75% non-GAAP, among the highest in the semiconductor industry, and the reported hikes indicate the company intends to pass memory cost inflation on to customers rather than absorb it, which it can more than afford to do. Meanwhile, supply of its accelerators from TSMC still can't meet demand, which limits buyers' immediate leverage.

Whether the increases push hyperscalers further toward AMD's accelerators or their own custom silicon will depend on how quickly those alternatives can absorb displaced demand, and all of them draw HBM from the same three constrained suppliers.

Bad bots account for 91% of the traffic reaching one retailer's DDR5 memory product pages, roughly 10 automated requests for every legitimate visit, according to a LinkedIn post this week from Jérôme Segura, VP of threat research at bot-mitigation firm DataDome. The figure updates research DataDome published in March, which measured the ratio at about 6:1 across several e-commerce sites.

Over the same period, the cheapest 128GB DDR5-6400 kit tracked by Tom's Hardware has reached $3,399, ten times its record low pricing. The Thales 2026 Bad Bot Report puts bad bots at 40% of all web traffic, less than half the share Segura reports on DDR5 listings.

DataDome's March report documented a single operation that generated more than 10 million blocked requests, with a one-hour sample showing 91 DDR5 listings each polled around 551 times, or once every 6.5 seconds. The requests carried cache-busting parameters to defeat cached stock data and targeted industrial module makers such as Micron and Apacer, as well as DIMM socket suppliers Amphenol and TE Connectivity, alongside consumer brands.

Segura wrote on LinkedIn that retailers carrying memory should assume their "traffic analytics are probably understating the problem significantly."

Keep in mind that the 91% figure comes from one unnamed retailer, and DataDome hasn't published a sample window or its methodology. DataDome also sells the product that blocks this traffic, and page requests ultimately aren't purchases, so the ratio measures monitoring pressure on listings rather than how much stock bots actually end up securing, which can’t easily be quantified.

Apacer CEO C.K. Chang said in July that DRAM supply to independent module makers could fall to 30% of 2026 levels next year, and TrendForce expects conventional DRAM contract prices to rise another 13% to 18% in Q3 after roughly 60% in Q2. Hyperscale buyers have reportedly placed deposits against most of 2027's DRAM output.

A 32GB DDR5-6000 kit that averaged $72 this time last year now averages $392 on PCPartPicker. While graphics card and console scalping in 2020 and 2021 ran against scheduled restocks that eventually outpaced resale demand, no equivalent restock date exists for consumer DDR5 before 2027 at the earliest.

Framework stopped selling standalone RAM in November last year, citing scalpers, and Micro Center sells its CPU, motherboard, and 32GB DDR5 bundles in-store only with a one-per-household limit. Newegg and Amazon have run repeated CPU-plus-memory bundles that price a 32GB kit well below its standalone cost, which ties each kit to a processor a reseller then has to move as well. On eBay, we found G.Skill's Trident Z5 Neo 32GB kit listed at $836.54 in December against a $429.99 retail price, roughly double, and about seven times the $117.99 it sold for before the shortage.

ChangXin Memory Technologies (CXMT), China's largest maker of dynamic random access memory (DRAM), planned to obtain Samsung's IP related to its memory process technologies in a bid to develop its own production nodes, according to testimony given in a South Korean criminal case involving leaked Samsung technology, reports Maeil Business. The company's managers allegedly had no intention to develop an in-house process technology from scratch, so obtaining IP from Samsung was a strategic decision.

The intention was a key part of CXMT's business plan rather than an opportunistic attempt to acquire Samsung's technology after the start of its own DRAM process development, according to the testimony of a former Samsung engineer identified as Jeon, who worked at Samsung for 28 years before jumping ship to CXMT sometimes around 2016. This April, Jeon was sentenced to seven years in prison for illegally obtaining a 600-step Process Recipe Plan (PRP) information concerning Samsung's DRAM fabrication processes in general and the 18nm-class node in particular.

"From the beginning, CXMT CEO and CTO tried to develop DRAM by stealing Samsung Electronics' Process Recipe Plan (PRP) information," Jeon told the court this week, according to Maeil Business. "CXMT had no research institutes or facilities at the time of its establishment, and had no intention of developing technology through its own research."

Copying the recipe does not automatically give you a working process. Nonetheless, a detailed process recipe can dramatically shorten development of any production node because it reveals things like the sequence of steps, deposition or etch conditions, anneals, cleans, implant parameters, temperatures, pressures, tool types, and so on. For a memory process, a ready process recipe can remove a huge amount of trial-and-error and reveal how the process integration is supposed to work. Meanwhile, even to ensure that the recipe works as intended, one needs to obtain the same or compatible tools as well as adjust their characteristics accordingly.

However, while a process recipe can potentially help to ramp any fabrication technology, a process recipe itself does not necessarily reveal the complete transistor architecture or physical design. For that, perpetrators also want things like cross-sectional structures, masks/layout information, design rules, device dimensions, capacitor geometry, process window, and so on. Such details are not easy to get for a brand-new fabrication technology, but they can be obtained by reverse engineering of shipping products. By combining process recipe with reverse engineering, experienced DRAM engineers can reconstruct much of the technology. Meanwhile, process integration engineers can save years of trial-and-error with actual fab tools.

Normally, it takes DRAM companies around 3 – 5 years to design a process node from device development to yield ramp. For a new player, development time from scratch will probably be around five years rather than three unless pre-competitive IP is licensed or certain heavy-lifting is done elsewhere. In fact, the Maeil Business report claims it took Samsung five years to develop its 18nm-class node.

Meanwhile, CXMT began mass production of DDR4 SDRAM in 2019 using a 22nm-class node. Mass production of memory using a 18nm-class process technology began in 2023. Such a timeline raises a question of whether CXMT's 22nm-class G1 was derived/adapted from the knowledge of Samsung's process technologies and/or their process recipes, the more advanced 18nm-class node, or did the allegedly obtained Samsung IP primarily help CXMT develop the later 18nm generation? In any case, there is a court ruling that CXMT stole IP from Samsung to boost its own memory production capability.

After posting record profits amid high memory prices, SK hynix ran into a dispute with employees over profit sharing and wage increases, which threatened to escalate into collective action by the company's union. This week, SK hynix management and the labor union reached a preliminary 2026 wage and collective bargaining agreement that significantly changes how the company's profit-sharing bonuses will be distributed, reports Maeil Business citing Yonhap. The good news is that no strikes are planned at the memory maker.

The tentative labor package reportedly calls for a 6.3% wage increase and an increase in welfare points provided by the company. In addition, under the proposed arrangement, employees will receive 40% of their performance-based profit distribution in cash, while the remaining 60% will come in the form of SK hynix shares. The proposal changes SK hynix's 2025 profit-sharing scheme, which allocated 10% of the previous year's operating profit to employee profit-sharing program, but the bonus was to be paid complete in cash: employees received 80% in cash immediately and remaining 20% over the next two years.

The union argued that stock-based bonuses expose employees to potential losses if SK hynix shares decline and could create tax complications, as income tax may be calculated using the share price when the stock is awarded. Specifically, the workers argued that performance bonuses normally paid in cash should not be converted into assets that carry investment risk. Yet, the SK hynix management and union have agreed to remove the 10% ceiling, potentially providing more money to employees, maintain the mechanism for 10 years, and pay the bonuses in cash and stock.

If SK hynix meets current market expectations and earns 25 trillion won ($17.897 billion) in operating profit this year, the profit-sharing pool would amount to 2.5 trillion won ($1.79 billion) based on the 10% formula. SK hynix reportedly has approximately 35,000 employees, so on average every employee will get 70 million won, or $50,120 ($20,048 in cash and $30,072 in stock). Yet, actual compensation will differ depending on an employee's position and individual performance assessment.

On August 4, before the fifth round of negotiations, an SK hynix union told members that if management did not present a revised proposal, it would be difficult to settle the dispute through negotiations alone and the union would have to take necessary action, which was obviously a way to threaten the management with a strike. Yet, the parties have come to terms.

The dispute comes as Samsung Electronics has faced similar labor action over compensations.

G.Skill, maker of some of the best RAM for PC gaming, has begun sending payments to claimants in the $2.4 million class action lawsuit over claims that it falsely advertised the speeds of its memory, misleading customers because the very fastest speeds could only be obtained by overclocking. According to screenshots from different users, reimbursements range from a paltry $20 to $25.

According to the lawsuit filed in 2022, G.Skill reportedly did not make it clear to consumers that overclocking was necessary to reach the advertised memory speeds on its DDR4 and DDR5 offerings from the Trident Z and Ripjaw series. The plaintiffs claimed that G.Skill’s packaging did not warn consumers about the potential risks associated with overclocking, such as system instability, or even possible hardware damage.

While G.Skill denied the allegations and asserted that its products were properly labeled, the company eventually reached a $2.4 million settlement with the plaintiffs. As part of the settlement, G.Skill would send cash payouts to eligible U.S. consumers who purchased certain DDR4 or DDR5 memory kits during the period mentioned in the lawsuit.

The exact payout amount for each claimant depends on the total number of claims submitted and how many qualifying G.Skill memory kits each person purchased. For example, one X user shared that they received $19.68, while separate reports on Facebook indicated that two users received $24.60 each.

In addition to the cash payout, G.Skill agreed to update the packaging for its memory kits and online product listings with clear disclaimers that consumers must overclock the memory kit, whether through Intel XMP or AMD EXPO enablement, to reach the rated memory speeds.

Memory speeds are a complex topic for the average computer user. Not everyone understands that memory kits default to the JEDEC baseline, whether DDR4-2133 or DDR5-4800, for maximum compatibility. However, to reach advertised speeds above the baseline, you need to overclock. Multiple hardware variables affect whether you can successfully use a memory kit that exceeds the manufacturer's specification.

For starters, the processor’s integrated memory controller (IMC) determines whether it can run memory above the manufacturer's specification, for example, DDR5-5600 for AMD Ryzen 9000 or DDR5-6400 for Intel Core Ultra 200S processors. That is the memory speed that the chipmaker guarantees will work on the chip. Furthermore, the motherboard must support the desired memory speeds. The QVL is a good place to start since it includes memory kits that the vendor has tested and confirmed to work on the motherboard. However, the QVL is not foolproof because the motherboard is only one of two factors at play, and sometimes the processor's IMC plays a bigger role.

Less experienced consumers may blame a memory kit for not working in a system. However, they fail to consider that the processor and motherboard are equally important. Memory speeds above the processor's default specification are not always plug-and-play, and it is good that manufacturers communicate this clearly to avoid consumer confusion and disappointment if a memory kit does not perform as advertised.

Pine64 has stopped producing Linux devices entirely, announcing this week on its official Telegram news channel that "there isn't [sic] any plans to continue producing more Linux devices in the near future" because of the ongoing DRAM and eMMC shortage. The stoppage covers the company's full lineup of Arm and RISC-V single-board computers, the PinePhone family, the PineTab2 tablet, and the PineNote e-reader, with remaining PineNote and PineTab2 stock estimated to last around three months.

Microcontroller-based products, such as the PineTime smartwatch, PineVoice smart speaker, and Pinecil soldering iron, won't be affected, and production will continue as usual. Development on the upcoming PineTime Pro will also continue: Pine64 said its demo firmware now has basic drivers for the display, touch, motion sensing, GPS, and audio, though some issues need fixing before a pilot batch enters production. The company also hasn't ended software support for existing Linux hardware.

The AI-driven memory crunch has raised prices on everything from DDR5 kits — which have sold on eBay for more than $2,000 — to prebuilt PCs from Dell, CyberPowerPC, and Maingear. Unfortunately, Pine64 buys the single worst-supplied part in the entire NAND market. TrendForce reported in April that the eMMC/UFS segment faces the tightest supply gap of any NAND category, because its production capacity overlaps with far higher-margin enterprise SSDs, making it the last in line for supplier allocation.

Phison CEO Khein-Seng Pua said in February that 8GB eMMC modules had risen from $1.50 to $20 in a year, a 13-fold increase, and warned the shortage could shut down entire consumer electronics companies in 2026. Pine64 hasn't shut down, but a whole product category has, and it's the first named vendor to kill a lineup outright instead of repricing it.

Other small-board makers have taken the opposite route. Raspberry Pi has raised prices repeatedly, pushing the Pi 5 16GB to $205 and the 500+ to $410, while launching a 3GB Pi 4 and leaning on a stockpile of older LPDDR2 to hold prices on legacy boards. Framework followed in January, raising its desktop pricing to cover LPDDR5x costs, which, since Strix Halo boards use soldered RAM, it couldn't pass to users any other way.

Pine64 sells its hardware near cost as a community service, so it has no margin to absorb higher component prices and, evidently, believes its buyers won't, either. Its mid-2027 checkpoint aligns with supply forecasts that don't expect new memory fab capacity to arrive in volume before late 2027 or 2028.

Pine64 said any decision on resuming production depends on component pricing after mid-2027, meaning its Linux hardware could be off the market for two years — or more.

There was a time when PC RAM and storage products were sold with lifetime warranties. Replacement of these components was an inconvenience at worst, with parts exhibiting great reliability and sold at trifling prices. In 2026, the tide has fully turned, and we see even the largest businesses wriggling to avoid warranty repairs and replacements with little shame. Yet another anti-consumer case in point is shared by GRINZ_DOCTOR on the Mildly Infuriating Subreddit. It appears to be a bizarre episode where Best Buy ‘repaired’ an under-warranty laptop by simply removing one of its SODIMMs, testing it worked reliably, and then sent it back to the customer with half the memory it originally shipped with installed, confusion possibly caused by a purchase made from a third-party seller.

You can see at the top of the letter from Best Buy/Geek Squad, dated August this year, the device sent in for warranty repair was a “Lenovo – Upgraded – Legion 5 15.1-inch WXGA Gaming Laptop, Intel i9-14900HX, RTX 5070, 64GB DDR5, 2TB SSD, +1TB dock station, Win 11 Pro” laptop in black, available to buy at Best Buy in the exact configuration. One possible reason for the mishap is that the linked product here is sold by a third-party retailer, PC Heaven, which offers "professionally customized and upgraded configurations." Upgraded in the product listing may be a giveaway that this is, in fact, one of PC Heaven's souped-up options.

GRINZ_DOCTOR sent the eight-month-old 64 GB laptop in for attention after beginning to experience repeated blue screen crashes. The notes mention that underlying software issues were basically ruled out ahead of the warranty return, as reinstalling Windows didn’t help.

Amputation was the chosen 'cure'

The technician accepted the Legion 5 in good cosmetic condition with “no visible wear or damage.” After investigations, the blue screen issues were clearly pinned on RAM, as the Best Buy tech closed the case with repair notes as follows: “Unit has two 32GB sticks for a total of 64GB; this unit only accepts 32GB RAM maximum. One stick of 32GB has been removed and sent back to client with unit.” Before ending the note with the punchline “Unit passed all tests.”

Of course, we don’t know exactly what was done by the repair tech during their fault-finding and resolution process. While GRINZ_DOCTOR insists they bought the unit as described on the upper section of the document with 64GB pre-installed, a product you can still get on Best Buy right now, it seems the most likely explanation for the snafu is that the customer bought an upgraded model from a third-party reseller. Best Buy's insistence that the laptop only supports 32GB of RAM does clash with the user's report that the laptop worked for eight months. Moreover, this certainly doesn't change the fact that they simply took the RAM from the machine without the customer's input. Best Buy also continues to list a configuration of a laptop on its website that the company admits is unsupported by the hardware.

Plenty of commenters have advised escalating the issue with Best Buy. They are advised to go to the store in person and keep escalating until a proper remedy - meaning a new working second RAM stick, or a like-for-like laptop replacement - is done. However, if they did indeed buy a third-party offering, their options may be limited.

Earlier this month, we reported on Micron offering pennies on the dollar for Crucial RAM returns. Meanwhile, in NAND-land, SK hynix reportedly ran out of replacement SSDs and is instead insisting on original purchase price refunds.

We're officially in dire straits. There's almost no way, if you're reading this site, that you aren't aware that memory prices have become entirely divorced from reality. Some are calling it the RAMpocalypse; I prefer "RAMageddon." Whatever you want to call it, though, the reality for PC builders right now is just grim. To put this in perspective, just take a look at our RAM price tracking post:

That's right: 128GB DDR5 kits are fully ten times more expensive than the lowest price we've ever seen. Things improve as you step down the memory capacities and speed tiers, but not as much as we'd like. You're still looking at $392 for a memory kit that was just $72 last year.

To reinforce the point, I pulled the latest average price data from PCPartPicker, comparing where we are today (August 2026) to exactly one year ago. This data is somewhat approximate, but it should be broadly accurate.

The numbers speak for themselves; in just 12 months, the cost of DDR5 memory has essentially exploded. We are looking at year-over-year increases nearing 500% for high-capacity kits. A standard 64GB (2x32GB) DDR5-5600 kit that would have cost you under $200 last summer is now demanding over $1,100. It is a 5x multiplier on a component that used to be a fairly boring and predictable line item in a PC build budget.

If you plan to just wait it out on an older AM4 or LGA1700 motherboard with DDR4, you'd better hope your memory holds out too, because DDR4 isn't safe from the fallout. With DDR5 entirely out of reach for most builders, the resulting scramble for older platforms running DDR4 memory has created a massive knock-on effect, and as a result, DDR4 kits are up anywhere from 120% to nearly 180% across the board. Nowhere near as bad as DDR5 pricing, but it still stings when a kit that was $105 last year is $281 this year.

This phenomenon is by no means exclusive to the US, either. German tech site ComputerBase have also been tracking this global trend, reporting just this week that average RAM prices in Europe have skyrocketed by 345% compared to September 2025. Their data shows the squeeze is bleeding into other components too, with hard drive and SSD prices both climbing over 125% in that same timeframe.

In fact, the situation is so severe that hyperscale buyers have reportedly already locked in almost all of the global DRAM production capacity for 2027, handing over advance deposits to guarantee their supply of precious DRAM, which is now among the highest-value commodities in the world by weight; mainstream DRAM chips are worth over half as much per kilogram as solid gold. PC and smartphone makers are simply fighting over the scraps, as low-priority markets compared to the extremely lucrative AI datacenter buildouts. And lucrative they are indeed; all four memory vendors (SK hynix, Samsung, Micron, and China's CXMT) have increased revenues by double, triple, or even more, all in just the space of one year.

For consumer prices to drop, we would essentially need a major contraction in the AI market, which would mean a massive correction in financial markets worldwide. Absent that, the people actually making the chips don't see an end in sight. SK Hynix CEO Kwak Noh-jung recently warned that 2027 will be the worst year for memory supply in the industry's history, forecasting that demand will outstrip their ability to produce it well into 2030. ADATA's Chairman, Simon Chen, was even more pessimistic, suggesting this DRAM crisis could last another 10 years and dismissing the idea of an "AI bubble" bursting anytime soon.

The era of cheap, plentiful memory is over, at least for now. If you need RAM today, you're either going to have to bite the bullet and pay the premium, or learn to get by with less. Make sure you check out our RAM price tracking post for the best deals we found on memory, as well as some tips on how to secure solid deals yourself.

What could possibly go wrong when you try to steal trade secrets from a restricted facility using a 360-degree action camera hidden in snacks? Forgetting to turn off the device’s Bluetooth and Wi-Fi transmitters, as it turns out.

A senior engineer at Nanya Technology has been indicted in Taiwan for allegedly copying 32 files containing information about the company's DRAM manufacturing process, reports Freedom Times. The engineer was reportedly in talks with Chinese recruitment agencies and allegedly planned to pass Nanya's trade secrets to its next employer in the People's Republic in exchange for higher compensation.

Yeh Yen-wei, a senior engineer in the Advanced Process Development Division at Nanya, joined the company in 2022 and began to communicate with Chinese recruiters in 2025 to explore employment opportunities in the People's Republic. Sometime in mid-February 2026, he submitted his resignation. However, before leaving the company in mid-April, on March 28, April 4, and April 5 — during weekends and either late at night or early in the morning — he visited the company's restricted office areas with his Insta360 X4 Air action camera hidden among snacks to get it past security checks.

Once inside, Yeh used his legitimate account credentials to access Nanya's virtual machines and opened various internal research and development documents. He then used the camera to capture photographs and video of information associated with Nanya's DRAM process technologies and production methods. It goes without saying that he retained the information when he officially resigned from Nanya in mid-April.

However, his actions did not go unnoticed. Nanya's IT security team detected 'abnormal' wireless signals inside the company's restricted facilities (i.e., identified the device using its Bluetooth and/or Wi-Fi signals) and tied them to the employee's access records and activity logs. Eventually, Nanya reported the suspected theft to Taiwanese authorities, which led to an indictment.

The New Taipei District Prosecutors Office charged Yeh under Taiwan's Trade Secrets Act with unauthorized reproduction of trade secrets intended for use in China, an offense for which he faces up to 10 years in prison, as well as with breach of trust under the Criminal Code. Meanwhile, forensic examination of the seized equipment uncovered no evidence that Yeh had successfully transmitted Nanya's confidential information to a Chinese company or anyone outside the company, so he could potentially get off easy.

Since late 2025, RAM, storage, and graphics card prices have shot through the roof, making items that were once affordable out of reach for many people. Anyone still in the market for those parts has to rely on combo deals for any reasonable pricing. Newegg has an excellent 3-item combo running where you essentially get the 16GB of DDR5 RAM for FREE with a huge $274 discount. For only $658.99, you get the powerful Ryzen 9 9900X processor, a quality Asus X870E-Plus Wi-Fi 7 motherboard, and 2x8GB Team Group DDR5-5200 RAM, along with a free Cooler Master 240mm AIO.

● Check out this deal at Newegg

The star of the combo has to be the ‘free’ RAM. The combo offers 16GB (2x8GB) of TeamGroup’s T-Force Vulkan DDR5-5200. And while the RAM speed is outside of AMD’s sweet spot, you’d be hard-pressed to notice a performance difference without looking at a frame counter. These black, non-RGB sticks will also look great with the included motherboard. If you apply the ~$274 discount to the RAM, you’re getting it for free. And if 16GB isn’t enough, buy the same kit again net 32GB for only $240.

Although we didn’t review AMD’s Ryzen 9 9900X, we know the non-X3D chips, like its big brother, the Ryzen 9 9950X, were generally well received by our reviewer and the public alike. Its strong multicore and single-threaded performance, native AVX-512 support, and backward compatibility with AM5 motherboards make this CPU an excellent choice for someone who uses their PC for more than just gaming and can make use of the additional cores and threads, as you can see from our benchmark data:

The Ryzen 9 9900X is a 12-core, 24-thread processor built on TSMC’s 4nm FinFET process (6nm for the I/O die) with a base clock of 4.4 GHz and boost up to 5.6 GHz. It has a TDP of 120W and is relatively easy to manage the thermals, even when overclocking the fully unlocked processor. While overclocking isn’t what it used to be, you can still squeeze out a couple of hundred more GHz, and sometimes you can get there, even when lowering the voltage. In short, there’s a lot of performance out of the box, and even a bit more headroom left if you want to overclock.

Last, the Asus TUF Gaming X870E-Plus Wifi 7 is a solid all-around motherboard for the AM5 platform. It delivers sufficient power to support flagship-class processors and easily handles the 9850X3D in the combo, even with some overclocking. It comes with four M.2 sockets (one PCIe 5.0) and two SATA ports for storage, fast networking with integrated Wi-Fi 7 and 2,5 GbE, ample USB ports on the rear IO (13, including two USB4 40 Gbps Type-C), and a quality audio solution based on the last-gen Realtek flagship (ALC1220P). Asus’ AI solutions (AI Cache Boost, Overclocking, Cooling II, and Networking II) and the DIY-friendly design make it easy to build and get the most out of your system.

If you’re in the market for an AM5-based system that isn’t based on an X3D chip, this $658.99 combo (an incredible $274 savings) is one of the better deals available at Newegg right now. And you even get a free 240mm Cooler Master AIO. Getting RAM for what essentially amounts to ‘free’ is not something we see frequently. You can even upgrade to 32GB and still find yourself well ahead versus buying alone. Get this deal while it lasts.

Intel CEO Lip-Bu Tan has revealed that new memory architectures — once thought a commodity business — are now strategically interesting and one of his pet projects, while speaking about the comeback of the American chip industry. He further noted that the memory industry is ripe for innovation, and also hinted at exploring ways to stack memory on top of a CPU.

After years of being a commodity, memory became a strategic asset in recent quarters and will likely remain one for a while, so suddenly, leading makers of 3D NAND and DRAM became highly profitable companies, something the Intel CEO is keenly aware of:  "I used to be, 'do not invest in memory because it is a kind of commodity business,' but now it has become different," Tan said. […] "There is a lot of new technology coming out. So, we are kind of looking at one of my pet projects, some of the new memory architecture. I think you just saw the news: I hired my good friend, Seok-Hee Lee, who used to run SK Hynix. So, you kind of know something that I am thinking about. We are not ready to unfold it."

Tan did not reveal anything about the pet project and did not even specify whether the project is one of his personally favored strategic initiatives at Intel, or an initiative in one of the companies that he has invested in. He did mention that stacking memory on top of a CPU could make a lot of sense, though did not elaborate. Tan's remarks also follow the revelation of an Intel XBM patent that does away with the silicon interposer of HBM.

" I think CPU and memory, I think there are a lot of ways we can really do stacking together," Tan said. "And also try to find some new architecture for memory. I think in some way the memory, a lot of innovation are not there. So, there is some really good area."

Not everyone in the industry remembers, but Intel started as a memory company in 1968 and was quite a successful memory maker until the eighties, when Japanese companies took the lead, and Intel had to exit the market completely after suffering severe losses. Since then, the company has attempted either to return to the memory market with NAND and Optane, or at least to capitalize on a new type of memory with RDRAM. In all three cases, the company abandoned its memory initiatives without incurring significant losses.

Given the current profitability of 3D NAND and DRAM makers, producing memory is certainly a good business again and will remain profitable for some time. However, to re-enter it, companies like Intel would need capital to build at least one fab, R&D to develop competitive process technologies, and time. While licensing a technology — assuming that a company has capital — is an option, we strongly doubt that at this point Intel may be inclined to invest capital in memory and not in its core products and foundry businesses. Furthermore, it is unclear how investors react to such investments given the fact that the company struggles to become a strong competitor in the foundry market and has exited 3D NAND and 3DXPoint/Optane businesses after failing to achieve strategic targets.

Micron used to produce some of the best RAM available to mainstream users through its Crucial brand. Despite the company's exit from the consumer market through its now-defunct Crucial brand, Micron had pledged to continue honoring warranties on existing Crucial products. A recent case on Reddit, however, highlights ongoing challenges with Micron's warranty process, where it reportedly offered a customer a ludicrous payout for an RMA. Only after further communication did Micron provide an improved offer. Another user also chimed in with a similar story about an SSD return with Micron. We have reached out to Micron for comment.

Redditor Key_Tailor6948, who shared the email exchange with Tom's Hardware, recently shared their RMA experience with Micron's warranty process after encountering a failing memory module from a 48GB (2x24GB) DDR5 memory kit. While the Redditor did not specify the exact model of the memory kit, Crucial used to sell two such kits in this capacity: the Crucial Pro DDR5-6000 C48 and the Crucial Pro DDR5-5600 C46. The last known pricing for the memory kits was $489.99 and $459.99, respectively, during the memory shortage.

According to the Redditor, Micron allegedly said in an email that the company is no longer offering replacement products for warranty claims. Instead, Micron purportedly stated it would provide a refund based on the original purchase price, including taxes where applicable. However, the proposed refund amount was $241.86.

The major issue at hand is that the reimbursement Micron offered was based on the cost of the 48GB memory kit purchased before the recent memory shortage and the resulting price hikes. Nowadays, the most affordable 48GB DDR5 memory kit available on the market, such as a DDR5-6400 C32 one, which is faster than Micron’s discontinued offerings, starts at a staggering $649.99. The refund Micron proposed amounts to just 37% of the current market value for a somewhat comparable product. The Redditor's math about being 17% of market value and estimating similar-capacity memory kits going for “roughly $1,400" is a bit off based on verifiable pricing, but the proposed payment was still well short of current market value.

Beyond the disappointment over the low reimbursement, the Redditor expressed frustration that they had to send back the entire 48GB memory kit, even though only one of the two modules was faulty. This meant parting with a memory module that was still fully functional. Unfortunately, this is a common industry practice. Most manufacturers require the return of the full memory kit for warranty claims. It is exceedingly rare for companies to allow returns or replacements of only a single memory module from a multi-module kit, regardless of whether the other module is working perfectly. (EDIT 8/11/2026 : Micron's warranty states that it will either replace a product under warranty or refund the current market value of the product or the amount of the original purchase cost, whichever is lower. Therefore, Micron's offer is within the bounds of its warranty.)

The Redditor was also not happy about covering the shipping costs to return the defective memory module. This is another policy that varies widely among brands: some manufacturers cover all shipping fees, others split the cost, so you cannot really single out Micron in that aspect.

It would seem that Micron subsequently offered a more favorable resolution. The company agreed to send the user three Crucial Pro DDR5-6400 C32 16GB (CP16G64C32U5B) memory modules as a replacement. This solution matches the original capacity of 48GB and also delivers better performance than the user's original memory kit. The user accepted the offer.

Meanwhile, Crucial SSD owners are seemingly experiencing similar RMA practices from Micron. A Reddit user named LeanaIsTheBest highlighted in the same thread that Micron is allegedly refusing to provide direct replacements for a failing 4TB SSD. Instead, the company ostensibly offered a gift card equivalent to the original purchase price of the 4TB SSD as compensation. However, the company provided an alternate option to receive the same amount in cash through PayPal or Venmo, with the condition that the user pays the 6% transaction fee.

Micron wound down its consumer business back in December 2025. It is reasonable to think the chipmaker has reshaped its manufacturing factories to cater to more profitable markets, such as data centers and the AI sector. Micron probably is not maintaining inventory or production lines for consumer products, so direct replacements are unlikely. For a company that reported revenue of $41.46 billion in its last earnings call, it isn't a great look for Micron to not reimburse the replacement cost for loyal customers who have purchased its Crucial products.

The best RAM on the market generally utilizes integrated circuits (ICs) from industry giants such as SK hynix, Samsung, or Micron. Recently, however, Chinese chipmaker ChangXin Memory Technologies (CXMT) has emerged as a strong competitor amid the ongoing global memory shortage. In Colorful's latest overclocking exhibition, CXMT's ICs achieved speeds up to DDR5-8800 on AMD's X870E platform, keeping pace with the best overclocking ICs, such as SK hynix's A-die and M-die.

Many often regard CXMT as an underdog in the memory industry, particularly when compared to the "Big Three." Founded only a decade ago, the ambitious Chinese chipmaker eventually became China's largest domestic memory manufacturer and is now the fourth-largest memory maker globally.

The recent memory shortage has accelerated CXMT’s rise to fame. The chipmaker has been silently gaining ground in the memory market. Its momentum has been especially evident in the DDR5 segment despite the company having only unveiled its first DDR5 ICs in November 2025. Many top Chinese memory brands, including Asgard, Colorful, Gloway, and KingBank, are widely using CXMT ICs, and even Corsair has done so to a certain extent.

Colorful showed off an incredible overclock with its iGame Shadow II DDR5-6000 C34 memory kits, pushing both the 32GB (2x16GB) and 48GB (2x24GB) configurations far beyond their rated speeds. These kits reached DDR5-8812 and DDR5-8817, respectively, on Colorful's iGame X870E Vulcan W OC motherboard. Unfortunately, the screenshots provided by Colorful were too blurry to discern detailed specifications such as memory timings or DRAM voltage. For context, some high-end DDR5-8800 memory kits typically operate at timings around 42-55-55-140 with a voltage of 1.45V.

The noteworthy aspect is that Colorful achieved the overclocking feat on 16GB memory modules. Most, if not all, DDR5-8600 and faster memory kits currently on the market come with 24GB memory modules. The reason is that many use SK hynix’s 24Gb (3GB) M-die ICs, produced on the chipmaker's advanced 1a-nm (~14nm) process node with Extreme Ultraviolet (EUV) lithography. These M-die ICs are the go-to choice for memory vendors thanks to their superior density, power efficiency, and ability to scale to higher frequencies.

In comparison, CXMT’s accomplishments are notable when considering the technological challenges. Due to U.S. export restrictions on China, CXMT does not have access to cutting-edge EUV equipment, so its 16nm node, which it uses for DDR5 ICs, leverages Deep Ultraviolet (DUV) machines instead. CXMT’s 16nm process is roughly in the same alley as the 1z nm (~14–15nm) nodes previously used by the Big Three, so CXMT is easily a few years behind the top-tier process nodes.

The reality is that CXMT will have to make do with what it has. The chipmaker is already developing a 1a-nm node but on DUV. It is far from prime time, though.

In addition to seeing CXMT ICs in mainstream memory kits, companies such as HP, Asus, and Acer have already started selling consumer devices, such as laptops or desktops with CXMT memory. Even Apple reportedly wants to buy memory chips from CXMT.

CXMT recently went public via IPO and saw its stock price skyrocket by 466%, raising over $8.6 billion. Without access to EUV machinery, CXMT will focus primarily on memory production and is rapidly building factories across China. The ambitious Chinese chipmaker hopes to steal 30% of the market share from the Big Three by 2030. It is just a matter of time before we start seeing CXMT ICs in all the prominent consumer memory brands.

Software performance expert, scientist, and GitHub developer Daniel Lemire compared historical computer memory prices and showed that decades of falling prices were undone in just a matter of months because of the massive demand for HBM. According to his X post, this event is a historical anomaly, and he cannot think of an instance where technology hardware prices reverted to levels from decades before.

“RAM on a per unit basis is about as expensive as it was in 2007,” Lemire said. He also added, “I do not think anyone can predict what will happen, but I am guessing that we either find a way to build AI systems without so much memory, or we find really clever ways to make much more memory much faster.”

We also checked third-party data compiled and maintained by David Shim for the Stanford DAM Project. The latest price for DDR5 memory sits between $13.28 and $11.41 per GB — the last time RAM hit this price was in 2008, when the price for DDR2 hovered by $15 and 11.00 per GB. These are based on nominal USD values, though. If we take 2024 inflation into account, the current amount for DD5 is between $12.74 and $10.94 per GB, which was last achieved in 2011, when DDR3 prices were at $11.85 per GB.

Technological advances, especially in manufacturing and miniaturization, meant that tech became cheaper over time. This has been true ever since the first general-purpose digital computer was created 80 years ago. ENIAC cost the government $400,000 in 1946, meaning it would probably cost around $6,850,297.44 today. However, its computing power equaled roughly 5,000 additions per second — by comparison, one of the cheapest smartphones you can get today, the Moto G Play, is on sale for $99.99. This smartphone is powered by a Snapdragon 680 processor with AI performance of 3.3 TOPS — that is 3.3 trillion operations per second. We can’t really compare the ENIAC to a modern processor, but it at least shows us the scale by which technology has advanced, and prices have dropped.

The massive increase in memory prices in recent times wasn’t caused by technological regression or a bottleneck in supply — instead, it’s the massive demand for HBM driven by the AI race. “The memory output is increasing by around 20% per year. Now, normally, that would be fantastically fast and amazing for any large, mature industry, but ask yourself, ‘Is the demand increasing by 20% a year?” Elon Musk said during the last SpaceX earnings call. “No, the demand is increasing by 200% a year, maybe higher.”

Other experts and industry leaders agree with Musk. The chairman of the SK Group, the holding company that owns memory chip manufacturer SK hynix, said that RAM prices are “abnormally high,” and that the industry must take steps to increase supply and lower prices. Even RAM that uses chips from Chinese manufacturer CXMT, which some people look to as a cheaper alternative, generally tracked the prices of modules that use chips from the big three (Micron, Samsung, and SK hynix).

Aside from computers, other industries are already being affected by the shortage, including graphics cards, smartphones, gaming consoles, and even automobiles. We do not know how long the current memory crisis will last, but the situation is unsustainable and “something has to give.” The scary thing is that nobody knows what would happen if and when that time comes.

Apple reportedly has around $1 billion in "processor wafers on the shelves awaiting packaging" amid ongoing memory shortages, putting pressure on the firm as it nears the launch of the iPhone 18. Tim Culpan of Culpium says Apple is "scrambling" for memory supply ahead of release, working with suppliers to get memory ready so it can finish packaging the A20 Pro and C2 chips that are reportedly inside the iPhone 18 lineup.

It's an interesting peek behind the curtain of the current supply chain limitations, which reach beyond simply DRAM chips. With fully-integrated SoCs across both desktop and mobile, Apple is uniquely vulnerable to supply chain limitations, not only in DRAM supply, but also in packaging capacity to bring together DRAM chips with processor wafers.

According to Culpan, Apple primarily sources DRAM from Micron, with smaller engagements from SK Hynix and Samsung. Micron (along with the other major memory players) has reportedly sold capacity for DRAM and HBM through 2027. A flurry of long-term agreements (LTAs) have cropped up over the past several months, guaranteeing supply to certain companies for several years.

The A20 Pro is the first chip to use TSMC's N2 process, and it has been progressing well, with TSMC reportedly sitting on around $1 billion worth of Apple wafers awaiting packaging with DRAM. Packaging happens with TSMC's new Wafer-Level Multi-Chip Module (WMCM) process, according to Culpan, which is similar to the CoWoS packaging but for mobile parts.

Culpan reports that "Apple and its assemblers remain confident that they can meet initial demand," but DRAM shortages could put pressure on shipments past the initial launch. Apple reportedly plans to ship around 200 million iPhone 18 units across the lineup throughout the device's lifespan. The iPhone 17 range saw record shipments last year, surpassing 245 million devices shipped in 2025.

Apple is one of several companies reportedly looking to alternative suppliers for DRAM, namely China's CXMT. The Chinese memory supplier has become a hot topic in the halls of the U.S. government over the past several weeks, with Apple reportedly lobbying for clearance as the government weighs the option to ban CXMT.

Pushback against Micron's planned manufacturing complex in Clay, New York, which comprises four separate fabs, is once again at the center of an environmental lawsuit. Filed by Neighbors for a Better Micron and Jobs to Move America — the same groups that filed a separate lawsuit against Micron in January — filed a new complaint Friday, claiming that approved wastewater and air permits aren't valid and will allow Micron to leak "forever chemicals" into the Oneida River.

"The Permits were issued against eh backdrop of the significant environmental consequences already acknowledged through the course of the Micro Project's State Environmental Quality Review Act (SEQRA) review. including the release of PFAS and other highly persistent and hazardous chemicals," the complaint reads. "Rather than fully and meaningfully evaluate and address those consequences and impose the necessary protections before granting final authorization, [the Department of Environmental Conservation] issued permits that defer essential analyses, controls and limitations, and mitigation decisions until after issuance or until future regulations and studies have been conducted."

The lawsuit names the New York State Department of Environmental Conservation (DEC), Onondaga County (where Clay is located), and Micron. Its claim is that, despite Micron receiving wastewater and air permits from the DEC, those permits aren't adequate to regulate the plant's emissions. In one part of the lawsuit, the petitioners claim the permits are "not routine operating approvals," and in another, claim the loose permit guidelines "are not collateral or technical omissions."

Per- and polyfluoroalkyl substances, otherwise known as PFAS or "forever chemicals," are synthetic chemicals that are essential to semiconductor manufacturing. The chemicals are present throughout nearly the entirety of the manufacturing chain, necessitating strict water and air containment management to align with environmental regulations. "The SPDES permit authorizes the acceptance and discharge of Micron's industrial wastewater containing identified emerging contaminants, including PFAS compounds, but it imposes no substantive controls on those harmful contaminants beyond monitoring their presence and concentration."

In January, the two petitioners named in this complaint filed another lawsuit against Micron, claiming that the project failed to comply with the state's environmental review process. Last month, Micron asked a judge to dismiss the lawsuit. It's still set for oral arguments on August 25, however.

This lawsuit currently doesn't have a date set for oral arguments, if it ever reaches that point. The petitioners are seeking the nullification of both wastewater and air permits, which would force Micron back under the environmental review process, and costs associated with the suit.

Storage and memory manufacturer Lexar has listed a new 32GB (2x16GB) DDR5-7200 CL38 memory kit in China with a price tag of 3,999 yuan (roughly $592). Part of the second-generation Thor RGB series, the kit features a sandblasted aluminum heat spreader, thermal pads, and eight LEDs that support a variety of RGB lighting effects. It is rated for 38-48-48-86 timings at 1.40V and supports both Intel XMP 3.0 and AMD EXPO profiles.

Lexar says the modules use carefully selected domestically produced DRAM, although it does not specify the actual manufacturer. While there is no confirmation, the company may be using DRAM chips made by CXMT (ChangXin Memory Technologies). Lexar has also published a compatibility list covering 43 motherboards, including 30 Asus models and 13 MSI models across Intel Z890, Intel B860, and AMD B850 platforms. The compatibility list also includes supported boards for a faster 32GB (2x16GB) DDR5-7600 CL38 memory kit.

Interestingly, none of AMD's X870 or previous-generation motherboards appear on the list. That said, the kit should work on most DDR5 boards that Lexar hasn't tested, but the absence of AMD's flagship consumer chipset from the compatibility list is a bit odd. Notably, major motherboard manufacturers, including MSI and Gigabyte, have already started optimizing their motherboards to support CXMT memory ICs.

A few months ago, Corsair began adopting CXMT DDR5 chips in select memory kits, including its popular Vengeance lineup. Likewise, major PC manufacturers such as HP and Dell have started qualifying CXMT DRAM, while Acer and Asus have reportedly asked their Chinese manufacturing partners to source locally made memory chips.

The arrival of CXMT-made DDR5 chips earlier this year raised hopes that a new DRAM supplier could ease supply constraints and eventually lower memory prices. So far, there seems to be little evidence of that happening. At around $592, Lexar's new DDR5-7200 kit is priced in line or only slightly below many comparable 32GB DDR5-7200 kits currently available in the U.S. Similar kits from brands such as G.Skill, Corsair, and Kingston are currently selling anywhere from $585 to $700.

There’s been a shift toward using a single DIMM of DDR5 to skirt the worst of the RAM pricing crisis. AMD has suggested to us previously, at least, that it’s an option PC builders are exploring, and it’s a wrinkle that’s shown up in dozens of prebuilts; one that we’ve been careful to avoid when recommending products, particularly during peak shopping events. But just how much gaming performance are you giving up with a single DIMM when using one of the best CPUs for gaming?

The conventional wisdom is that you’re giving up quite a bit of performance, resting on the ideas of previous DDR generations, where going down a single stick will cut your bandwidth in half. The same is true with DDR5, but things are a bit messier, which we’ll dig into later.

At a high level, using a single DDR5 DIMM doesn’t compromise gaming performance by nearly as much as you might expect. We saw drops of around 8% to 10% across Intel and AMD CPUs, and less than a 5% difference with AMD’s Ryzen 7 9800X3D.

A proper dual-channel configuration is still the way to go for optimal performance. However, our testing shows that even a single DDR5 DIMM offers superior performance to dual-channel DDR4 on CPUs that support both memory generations.

Using a single DIMM isn’t ideal, and it brings up compatibility issues if you plan on adding another DIMM down the line (another area we’ll touch more on in-depth later). But as a stopgap measure amid surging DRAM prices, a single DIMM of DDR5 holds up surprisingly well in gaming scenarios, and it represents a nice on-ramp into a DDR5 platform while prices are out of control.

Testing single-DIMM DDR5 in games

The geomean below speaks for itself. Moving down to a single DIMM drops some performance, but it’s not nearly as stark as I expected. There are some important notes about how we tested before dissecting the results, however.

As usual, we tested at 1080p with a mixture of High and Ultra settings, using Nvidia’s RTX 5090 Founder’s Edition to minimize the influence of the GPU. That’s even more important here, as we’re not evaluating CPUs but the memory that feeds the CPU. We're looking at a separate GPU and CPU here, as well; dual-channel memory is vital if you're using an iGPU.

We also wanted to create a realistic testing scenario. Assuming you’re going for a 2x16GB configuration, that means starting with a single 16GB DIMM, not a 32GB one. That’s what we did here. Our single-DIMM configurations run with just 16GB of memory, while dual-DIMM configurations run with 32GB. Capacity isn’t a major limitation for our testing here, though it will pose performance issues if you’re running several applications alongside your game.

The Ryzen 5 7600X was our biggest loser, dropping 10.9% of its average performance across our 13-game geomean with a single 16GB DIMM. The Core Ultra 7 270K Plus, meanwhile, dropped 8.7% of its average performance, while the Core i5-14600K similarly dropped 8.3% of its performance. The 1% lows are especially notable here, scaling with average performance rather than falling off a cliff.

Unsurprisingly, the Ryzen 7 9800X3D fared the best, dropping just 2.8% of its performance on average. As you’ll see through our individual game tests, the Ryzen 7 9800X3D shows identical performance with two DIMMs and a single DIMM in multiple games. The large and speedy L3 minimizes the impact of using a single DIMM, giving AMD’s 3D V-Cache CPUs another notch in their belt (as if they needed another).

Our other geomeans don’t say much. A single-DIMM configuration consumes a few watts less, leading to slightly lower efficiency when balanced against the lowered performance.

007 First Light, the newest game in our test suite, shows good scaling across our test pool, so it’s a good place to start. The Core Ultra 7 270K Plus dropped 8.9% of its performance with a single DIMM, while the 9800X3D dropped 6.5%. The 14600K is worth highlighting here. A single DIMM lost us 7.9%, but that’s still better than dual-channel DDR4, and by a decent 6.8% margin.

This trend is fairly consistent, too. Starfield is punishing for single-DIMM configurations, with the 14600K losing 10.6% of its average performance. Even then, a single DDR5 DIMM is 6.9% faster than dual-channel DDR4. Intel platforms can scale up to high data rates, and we use 7,200 MT/s for Intel chips. With slower speeds, the performance will equalize. Still, I’d strongly consider a single DDR5 DIMM if you’ve been eyeing an LGA 1700 platform with DDR4, assuming you’re primarily focused on gaming.

Starfield shows the benefits of 3D V-Cache in this single-DIMM configuration clearly. While the performance losses with other CPUs are brutal in this title, the 9800X3D dropped just a few frames, a 1.6% decrease in average performance.

Doom: The Dark Ages is similarly harsh on single-DIMM configurations. The Ryzen 5 7600X dropped 17.3% of its average performance, and the 270K Plus fell by 14%. The 9800X3D holds up well, however, with just a 4.2% average performance loss. Marvel’s Spider-Man 2 (in the gallery below) shows a similar dynamic.

However, the addition of 3D V-Cache mainly helps in these games where we see bigger losses with other CPUs. There are several games in our suite where the performance loss between a single DIMM and dual-channel configuration was minor, particularly in GPU-heavy titles. The Last of Us Part One shows that in action.

Similarly, the performance loss is minor in Counter-Strike 2, just a handful of frames even north of 600 FPS. Even the Core i5-14600K, which was hit the hardest by far, dropped just 5.9% of its average performance (and it’s still faster than a dual-channel DDR4 configuration).

You can browse the results of the other games we tested in the gallery above, though they tell much the same story as the titles we highlighted. The two biggest learnings from this testing are that a single, fast DDR5 DIMM often outpaces dual-channel DDR4, and that AMD’s 3D V-Cache can smooth over the rough edges of using a single DIMM, particularly in those memory-sensitive titles.

How we tested

We tested on our normal test bed that we use for CPU reviews, with the only difference being dropping one of our DIMMs. As usual, we enabled XMP/EXPO for testing, both on single- and dual-DIMM configurations, opting for higher speeds on Intel platforms. AMD’s sweet spot is around 6,000 MT/s.

The data rate is noteworthy for the DDR4 comparison, in particular. With slower DDR5, we expect the delta between dual-channel DDR4 and single-DIMM DDR5 to decrease.

In addition to enabling XMP/EXPO, we disable Virtualization-Based Security (VBS), enable Resizeable BAR, and disable any automatic overclocking features that aren’t covered by warranty, including AMD’s Precision Boost Overdrive and Intel’s Extreme performance profile.

Complexities of DDR5

DDR5 has much higher data rates compared to DDR4 (2x or more), and that’s the result of significantly more complexity on DDR5 DIMMs compared to DDR4. Notably for this testing are the two 32-bit subchannels of DDR5, as well as enlarged prefetch and bank groups. You are getting half the total bandwidth with a single DDR5 DIMM, absolutely, but it’s a little more nuanced than simply running your memory in a single-channel configuration.

DDR4 uses a single 72-bit channel, and 64 of those bits transfer data (the additional 8 are for ECC). With DDR5, you get two 40-bit subchannels, 32 of which are used for data transfer. Critically, these subchannels have their own Command/Address (CA) interface, allowing them to operate independently of each other.

In addition, DDR5 doubles the prefetch size to 16N, doubles the number of banks and bank groups, and increases burst length from 8 bytes to 16 bytes. The math here is straightforward enough. With DDR4, you can transfer 64 bits of data over the interface at a burst of 8 bytes, giving you 64 bytes of data and aligning with cache line sizes. Each of the subchannels in DDR5 can transfer 64 bytes of data, as well, with a 32-bit width and burst length of 16 bytes.

It’s pseudo-dual-channel memory on a single DIMM. There are two independent memory operations that can happen each cycle, both of which transfer 64 bytes of data. That is a big reason why even a single DDR5 DIMM (particularly one with a higher data rate) can outpace even a proper dual-channel DDR4 configuration.

The common wisdom about using a single DIMM comes mainly from previous DDR generations, where you can only get that single 64-byte transfer each cycle, necessitating a second DIMM to have two independent operations.

Let me be clear: two DDR5 DIMMs are still better. Instead of two operations per cycle, you can perform four. The main difference is that, for modern CPUs and games, just those two operations give you the vast majority of your performance, and you don’t run into a massive performance bottleneck as you do with just a single operation via DDR4. After all, there are still a lot of DDR4 systems out there.

On the issue of compatibility

The elephant in the room with upgrading to a DDR5 platform with a single DIMM is compatibility. If you get a single DDR5 DIMM now, you should plan to add another in the future. Unfortunately, there aren’t a lot of easy answers when it comes to combining two DIMMs that aren’t in a kit together. It should work, but mixing and matching can lead to stability issues and/or lowered performance.

Generally, combining two DIMMs of the same spec should work. For instance, if you bought this Patriot Viper Venom DDR5-6000 CL30 DIMM to start your new build, and then bought a second one later, it should work. “Should” carries a lot of weight on its shoulders, however. Maybe Patriot sources DRAM from multiple places, or your second DIMM’s timings don’t completely align with your first. That could cause stability or performance issues.

There have been some attempts to get past this hurdle, such as the ROG Certified program with select Asus motherboards, but it’s impossible to know for sure if two separate DIMMs will work together or not without testing. The best you can do is to make sure you buy the same brand of memory, with the same spec, if you add a second DIMM. Make sure to keep your receipt handy in the event you need to return it, as well.

Once you have your second DIMM, make sure to test it first at stock JEDEC speeds/timings (no XMP/EXPO). Run games or applications you normally run, and then use a utility like MemTest64 to check for stability. I’d recommend keeping your PC running in this state for a few days to see if any BSODs, application crashes, or poor performance issues pop up. After that’s done, repeat the process with XMP/EXPO enabled, assuming you’re using matching DIMMs (incompatible XMP/EXPO profiles won’t work together).

Assuming DRAM prices continue at an elevated rate before you upgrade, you may want to consider investing in a dual-channel kit down the line and selling your single DIMM. Even kits can run into compatibility issues, though the likelihood is far, far lower.

Memory supply from major DRAM manufacturers to independent module makers could drop to just 30% of the amount supplied in 2026 next year, according to C.K. Chang, CEO of Taiwanese memory vendor Apacer, further tightening an already squeezed supply chain.

Chang made the comments during the company's 1H 2026 investor conference on July 24, and in subsequent media interviews after the conference. Although memory price increases are expected to slow during the second half of 2026 — reportedly due to consumers deciding that RAM is too expensive — he believes severe shortages will persist into at least the middle of 2027, and that DRAM will remain the most constrained segment.

According to Chang, the greatest risk facing Apacer, which buys memory wafers from manufacturers such as SK Hynix and Samsung and turns them into finished memory products, is no longer overpaying for the chips, but failing to obtain any supply at all.

In preparation for growing shortages and to prevent a wider shortfall next year, the company says it grew its inventory to NT$12.4 billion ($383.2 million USD) at the end of June, up from NT$8.38 billion ($259 million USD) one quarter earlier, representing an increase of approximately 48%. The company is also arranging a five-year syndicated loan of up to NT$4 billion ($123.6 million USD) to purchase additional chips whenever manufacturers make them available, among other needs.

Chang’s projection does not mean worldwide DRAM production will fall by more than 70%. His statement specifically refers to the volume major chip manufacturers may allocate to downstream module companies such as Apacer, which purchases memory chips and packages them into DIMMs, SSDs and embedded storage products. DRAM manufacturers are increasingly reserving their output for high-bandwidth memory (HBM), server memory, and other products purchased directly by large AI and cloud customers.

Samsung, SK Hynix and Micron — the world's biggest memory chip makers — are prioritizing higher-margin HBM and advanced server memory as AI infrastructure spending consumes an increasing share of available manufacturing capacity. Chang estimates that approximately 60% of DRAM capacity is now directed toward server-related applications, leaving conventional DDR4 and DDR5 products competing for a shrinking portion of the market.

DDR5 RDIMM server modules have received some of the most aggressive price increases and retain the greatest potential for future mark-ups, according to Chang. Demand from AI servers, enterprise storage, industrial computers and edge AI systems remains strong, while consumer PC and smartphone demand is considerably weaker and more sensitive to rising component costs. That weaker demand is, of course, mostly in comparison to the AI industry. Consumer demand remains strong enough to mop up the shrinking supply. The RAM shortage is now reportedly affecting the price of cars, far beyond consumer electronics.

NAND flash is also being pulled into the AI boom. High-capacity enterprise SSDs are increasingly being used for model storage, data staging and key-value cache offloading, allowing colder portions of AI workloads to spill out of expensive DRAM. Flash cannot replace DRAM outright because it offers considerably lower bandwidth, but it can still serve as a tier within AI memory systems, increasing demand for enterprise SSDs and NAND alongside server memory.

Chang expects DRAM contract prices to rise by approximately 30% during the third quarter of 2026 and NAND flash to increase by more than 20%. He expects price growth to moderate again during the fourth quarter.

Elsewhere, analysts project a 40% increase in DRAM prices in Q3 2026, even as demand extends to the oldest standards still in production. Samsung and SK Hynix, both South Korean companies — who together with US-based Micron Technologies control over 90% of the global DRAM market — had earlier warned that shortages could last beyond 2027. And the CEO of Adata projects that the global DRAM shortage will run for another 10 years.

Chinese manufacturers are not expected to provide immediate relief. Chang said Chinese memory maker CXMT’s DDR5 products had become competitive and that both CXMT and Chinese NAND producer YMTC had narrowed their pricing gaps with established international suppliers. However, domestic demand in China already exceeds available supply, while capacity expansion, manufacturing yields, product validation and platform compatibility continue to limit their ability to change the global balance. Furthermore, the prices of new DRAM modules from CXMT are reportedly similar to those of the big three.

Stockpiling memory inventory at historically high prices carries the risk of substantial losses if the market suddenly reverses. Apacer, however, says it has seen no evidence of an approaching collapse. For now, the company is betting that possessing expensive memory in 2027 will be considerably better than having no memory to sell at all.

On the eve of August, 2026, the DIY PC building market looks bleaker than ever. Over the last 12 months, enthusiasts have endured spiralling prices on DDR4 and DDR5, as well as hard drives and SSDs, driven by shortages induced by massive AI buildouts. But just how bad is the RAM market in 2026, and is it at all possible to buy cheap RAM these days? The short answers are “bad” and “not really, no.” In this brief 2026 ‘State of the Union’ on the RAM market, we observe just how much prices have changed in the last 12 months, and examine whether there really is any hope for PC builders at all.

It was around July last year that the alarm bells really started ringing when it came to RAM pricing, with TrendForce warning of 45% increases in the second half of the year. Until then, both RAM and SSDs had been something of an afterthought when it came to PC building. Not that those components weren’t important, but rather that you knew you could always find a cheap 32GB DDR5 kit and a decent SSD for whichever build you were planning. The tough decisions were choosing between AMD and Intel’s respective platforms, and trying to find a decent graphics card.

So what has happened since?

What a difference a year makes

To see the impact of the AI crunch on RAM pricing, I had a bit of a dig through the Tom’s Hardware archives, but I should warn you: It’s not pleasant reading. This time last year, we had just come through the throes of Amazon’s 2025 Prime Day event, held in July as opposed to this year’s June offering. As is customary, we offered our thoughts on a great PC build using Prime Day discount parts featuring a Crucial Pro 32GB DDR5 6000 memory kit, reduced from $119 to just $90. One year later, that very same kit now costs $459. That’s if you can even buy one, of course, as there was only one left in stock when I wrote this. Oh, also, Crucial no longer exists. Feeding the AI machine is proving so lucrative that Micron literally wound up its consumer memory and SSD offering so that it could put all of its efforts into HBM and enterprise.

The story is similar wherever you look. Our pick for the best RAM for gaming in 2026 remains G.Skill’s Trident Z5 Neo RGB DDR5-6000. When we reviewed this kit back in 2023, we bemoaned that users were paying the early adopter DDR5 tax. The price? A paltry $179. You now can’t buy 32GB of DDR5 for less than $379. G.SKill’s kit subsequently sank in price to an even $100; in August 2026 it’ll cost you $599.99. I guess you could call that a late adopter tax.

Even 16GB of DDR5 6000 will set you back more than $200 these days, which is double the price for half the capacity you could buy a year ago. Large kits, like Corsair’s Vengeance 192GB (4x48GB) offering, now cost upwards of $2,600, more than most pre-built PCs.

The death of Crucial

As mentioned, one of the most pertinent death knells of the RAM market in the last 12 months was the end of Crucial. In December, Micron announced that it was winding up its entire consumer offering, reallocating all of its resources to enterprise DRAM and SSDs. Crucial products have slowly died off since, and the very few kits and drives that remain are extremely expensive to buy. The company is continuing to honor warranty and technical support obligations, at least, but that will be of little comfort to PC builders left out in the cold. “Micron has made the difficult decision to exit the Crucial consumer business in order to improve supply and support for our larger, strategic customers in faster-growing segments,” Micron admitted bluntly at the time.

The return of DDR4

Not totally oblivious to the plight of gamers, a few vendors are at least trying to assuage gamers and the DIY market, notably with the return of DDR4. Admittedly, this isn’t super comforting — I can’t imagine many console gamers would be thrilled about Sony solving PlayStation 5 shortages by bringing back the PS4 — but the gulf between DDR4 and DDR5 isn’t so bad that it’s not a viable option for some in 2026. We’ve heard rumblings that Intel is preparing a surprise Raptor Lake Next return for 2027, and AMD recently rolled out its 10th anniversary edition of its 5800X3D gaming CPU. Still, the best gaming CPU for DDR4 that money can buy still costs $350, and the gulf between DDR4 and DDR5 is widening, in our testing. As our resident CPU analyst Jake Roach noted at the time, buying a Ryzen 5 7600X3D and 16GB of DDR5-6000 is only $80 more expensive than buying a DDR4-based 5800X3D.

The problem lies therein: DDR4 memory prices have also skyrocketed. A 32GB Corsair Vengeance DDR4 3200 kit from 2019 would have set you back 48 bucks last year. Now, you’ll be lucky to find it on sale for $200.

Prebuilt your way out of this mess

Pre-built PCs from OEM vendors like HP, as well as bespoke offerings from CyberPower and the like, are not immune to rising prices, nor are gaming laptops. To return to our one-year flashback, an Intel Core 7 Alienware Aurora 16 we recommended to users at $1,099 during July 2025 will now set you back $1,999, and we’ve seen price increases on Alienware desktops of up to $2,000 in recent months. In fact, Maingear started offering “Bring Your Own RAM” builds late last year, rather than absorbing the cost of price rises.

With that being said, if you have given up hope of getting RAM and doing a DIY build in 2026, there are still pre-built gaming PCs on sale that get you a lot of PC for your money. Perhaps more money than you might have paid last year, but in reality it's a simple math problem. RAM prices have risen by 400, 500, or even 600% in some cases in recent months, while PC builds as a whole have increased, but not by nearly as much. When you factor in the increased cost of SSDs and GPUs, pre-builts in 2026 tend to represent much better value than they generally would have a year ago.

If you’re set on buying RAM?

If you are set on building a PC right now and buying RAM to put inside it, what’s the state of play? The going rate for 16GB of DDR5 RAM is $209, the lowest amount of money you can expect to pay for a 16GB kit. For 32GB, the going rate is $379. Per GB, 32GB remains better value, at around $12 rather than $13. 64GB of DDR5 from Corsair is $699, ironically the best price per GB, but ludicrously expensive.

We’ve been touting Newegg RAM bundles at Tom’s Hardware for months, and they remain a stalwart way to save at least something on RAM if you’re building a PC. Generally, you can expect to find a bundle that will get you the RAM for around $200 if you buy it alongside a motherboard and a CPU. Remember, the value in a RAM bundle is pairing RAM with other parts, the prices of which are not inflated in the current market. With CPU, motherboard, and cooling prices holding steady or falling, buying components together gives you a clear picture of the actual value proposition. The RAM in this Newegg bundle, for instance, costs $236, thanks to the price of the 9800X3D and the X870E.

The future of the RAM market

Sadly, if you’re in the market for RAM, even at these exorbitant prices, we’re still firmly in the bite-the-bullet phase. Manufacturers like SK hynix are working to add capacity to their roster, and the advent of China’s CXMT could provide some relief. But industry figures warn that 2027 could actually be even worse for memory shortages than 2026. By extension, that presumably means pricing will also get worse. In truth, the best solution right now is probably not to buy RAM, but if you’re in desperate need, you’re going to have to pay high prices in some way, shape, or form.

Waiting it out is always an option if you can do so, but Adata’s chairman recently said you might be waiting a decade or more. The latter prognostication sounds a little extreme, but one year into the RAMapocalypse, more than a few indications point to things getting worse rather than better.

Unless you already have a kit of DDR5, or deep pockets for PC hardware, you’re stuck right now. Despite seeing some of the best CPUs for gaming released just in the past year, the DDR5 tax represents a major hurdle to building or upgrading your PC. Intel’s LGA 1700 CPUs hold an unintended solution, offering support for both DDR4 and DDR5. We wanted to see if this on-ramp to a next-gen platform was really as smooth as it looks at first glance.

The idea is simple. If you already have a DDR4 kit, you can upgrade to a relatively new CPU while keeping the door open for a transition to DDR5 without buying a new chip. It’s inexpensive, and a bit more of a practical solution than upgrading to DDR5 with a single DIMM and worrying about pairing it with another later.

Our testing shows how much of a performance difference there is with DDR4 in games. We’re focused on games here for a few reasons. First, in non-gaming applications, memory speed is highly dependent on the specific workload you’re running. Games, despite using different engines, are largely similar workloads that stress the hardware in similar ways. Further, those workloads are continually evolving as new games, new technologies, and new hardware targets are released.

We kept our testing focused on Intel’s LGA 1700 CPUs because they’re the only options that give us a true apples-to-apples comparison between DDR4 and DDR5. We don’t have any AMD CPUs in our test pool because there aren’t any chips that support both DDR4 and DDR5. This isn’t intended to be a competitive analysis between AMD and Intel. That conversation opens up a much broader conversation about total platform cost.

Testing DDR4 vs. DDR5 on Intel’s LGA 1700 CPUs

To test DDR4 and DDR5’s gaming performance, we ran Intel’s main LGA 1700 stack through a 15-game gauntlet of benchmarks. Notably, we don’t have results for the Core i9-12900K, as our sample has a janky memory controller that won’t boot on DDR5 systems. You can read a full breakdown of how (and why) we tested in the way we did in the section below.

We ran all of the games through our 16-game benchmark suite, though we cut Doom: The Dark Ages from the results here. An update for the game rolled out mid-way through testing and skewed our results, so we’re omitting it. The 15 remaining games still paint a clear picture of the performance difference between DDR4 and DDR5.

As usual, we tested with the RTX 5090 Founder’s Edition at 1080p with a mixture of High and Ultra settings. We test at 1080p to maximize the differences between CPUs, and it remains the most widely-used resolution for gaming. Deltas between CPUs will decrease as the resolution climbs and the GPU becomes a larger influence in performance.

Below is our geomean from testing across the suite. Given that we’re looking to compare a single CPU’s performance with DDR4 and DDR5, we’ve highlighted some CPUs with the same color (13700K is always dark red, for instance, while 14700K is always dark blue).

The chart above is a bit difficult to parse, mainly because we’re often jumping several entries to get to a comparison point. For the stack of chips we tested, here’s a clearer overview of the performance difference with DDR4:

As we’ll get into with the individual game tests below, the drop in performance between DDR4 and DDR5 can be much larger than the comparison in our geomean (and, in turn, smaller in other games). What’s important for the overall performance drop is the trend as we move to newer chips. You can see a 9% drop grow to 14%, as DDR4 becomes a more significant bottleneck when the processor it’s feeding becomes more powerful.

That trend is one of the main driving forces behind our look at DDR4 and DDR5 gaming performance today. We looked at DDR4 and DDR5 performance extensively near the Alder Lake launch, and even more recently last year, evaluating performance across a wide range of applications. The testing here is focused solely on gaming to answer a critical question of building a PC today: Are you really giving up that much gaming performance with DDR4?

The answer is, yes, you’re giving up around 11% to 14% of your gaming performance overall when the memory is the only variable that changes. That’s a consistent drop, as well. Older titles care see less of a benefit from DDR5 generally, but we saw a steady decrease of around 11-14% across the games we tested, with some titles pushing above a 20% drop.

007 First Light is the newest game we tested, and it outpaces our averages. The 14900K sees a 16.5% drop with DDR4, while the 13600K drops 12.7% of its performance. Especially among the most powerful CPUs here, you can see DDR4 level off performance in a way that’s undesirable. The 14900K, for instance, is just 1.7% faster than the 13900K with DDR4, but it’s 4.6% faster with DDR5.

Similarly, in Crimson Desert, you can see the 13700K lose 13.7% of its performance with DDR4, and the 14600K drop 13.2% of its performance. It’s worth noting that neither of these games are particularly sensitive to memory speed. You can see in our recent Ryzen 7 5800X3D re-review that even a boost from 3D V-Cache doesn’t give CPUs with a robust memory chain an automatic advantage.

Perhaps the most consequential of the newer games we tested is Marvel Rivals, not only because it shows a much larger gap between DDR4 and DDR5, but also because it’s the most-played game in our test suite by a long shot (short of Counter-Strike 2). For weaker chips like the 12600K, the gap is 13.9%, just slightly above our geomean. For the more powerful 14700K, however, it shows a gap of 25.3%. Even the 14600K drops 17.4% of its performance with DDR4.

Counter-Strike 2 is less extreme, perhaps as a consequence of having such a high floor for average frame rates. You can see that the Alder Lake chips don’t care much about DDR4, with the Core i7-12700K performing identically across both memory standards. That gap grows with more powerful CPUs, however, with the 14700K creeping up toward a 10% delta.

In Spider-Man 2, we can see more consistent drops across the stack of chips we tested, with most CPUs dropping around 18% of their average performance with DDR4. That’s true even on the weaker 12700K, which lost 15% of its performance. The 14900K, meanwhile, was 18.7% slower with DDR4.

Even in an older, GPU-heavy title like Cyberpunk 2077, we can see a consistent drop. You can see in this game that we become completely bound by the GPU at the top of the rankings, but that’s a level these chips can’t hit when paired with DDR4. The 14700K drops about 14% of its performance, and the 13600K, about 13% of its performance.

You can browse through our remaining benchmarks using the gallery above, but most games tell a similar story. The gap between DDR4 and DDR5 persists, of course, but there’s another angle to look at the data. DDR4 flattens off performance, creating a bottleneck with faster CPUs. That could be justification for buying a weaker chip if you can only afford DDR4 memory right now.

Outside of performance, one important aspect of using a DDR4 system compared to DDR5 is power consumption. With DDR4, power consumption of the CPU increases, as you can see in the chart above. The jump is generally only a few watts, but it’s worth noting nonetheless. In games, we’re well below the operating temperature of all the chips in our test pool, but when pushing toward maximum power draw, DDR4 will post yet another limitation.

How (and why) we tested DDR4 vs. DDR5 now

DDR5 is way too expensive, and Intel’s LGA 1700 CPUs represent a unique value proposition given the unprecedented increases in memory prices over the past several months. The idea is that you can upgrade to a newer platform while sticking with DDR4, giving you a stopgap for an eventual DDR5 upgrade. It seems Intel recognizes this value proposition, as well, as it’s reportedly slated to launch Raptor Lake Next CPUs on the LGA 1700 socket.

We tested the main stack of chips, short of the Core i9-12900K, as described above. We didn’t test variants like KS releases to keep our testing focused, as well as lower-end SKUs like the 12100F or 14400. The memory controllers in these lower-end chips are weaker, sustaining lower DDR5 speeds, so expect less of a difference in performance with them.

Below, you can see the test benches we used. In addition to keeping hardware consistent, we use a frozen OS image with an identical software stack. We’re running the same version of the same apps on the same OS update, with identical drivers.

In the BIOS, we enabled XMP for both kits of memory. We also tested with Resizeable BAR turned on, and Virtualization-Based Security (VBS) turned off. Most motherboard vendors off a profile with the power limits removed on Intel CPUs, which can push the processor outside of warrantied specifications. We stuck with the default “Performance” profile for our testing here.

Is DDR4 still worth it?

It makes sense that we’ve seen a shift back toward DDR4. Although DDR4 prices have risen in lockstep with DDR5 prices, they started from a much lower base. A kit of DDR4 today is not much more expensive than a DDR5 kit of similar capacity a year ago. And, with estimates that the memory shortage could persist through 2030, there’s a strong argument for going with a DDR4 system if you’re in need of an upgrade.

The performance trade-off alone isn’t as big of a problem as it appears at first glance. You’re going to lose about 15% of your average gaming performance with Intel’s LGA 1700 CPUs (particularly Raptor Lake chips). You’ll spend more than twice as much on a 32GB kit of DDR5-6000 as you will on a 32GB kit of DDR4-3200, as you can see in our RAM price tracker. And, you might already have a DDR4 kit, cutting that cost completely out of an upgrade. Dollars spent for performance gained, DDR4 makes complete sense.

However, performance isn’t the only factor at play, and that’s the unfortunate reality of using a DDR4 platform right now. It’s a dead end, both for performance and upgrade potential. As we saw in our recent testing of the Ryzen 7 5800X3D, there’s a performance wall in games with DDR4, one which the 5800X3D often runs up against. The memory is a bottleneck, and it doesn’t seem like there’s a way around that problem without upgrading to DDR5.

We also aren’t seeing any new DDR4 CPUs, at least right now. AMD re-released the Ryzen 7 5800X3D, and there are murmurs of Raptor Lake Next. But given the current landscape, we don’t expect a CPU that will radically change the performance you can get out of a DDR4 platform.

Instead of sticking with DDR4 for a better CPU, the best course of action right now is to bite the bullet on DDR5 and buy a weaker CPU, particularly if gaming performance is your main concern. Given that memory prices are out of control, there are excellent deals on DDR5 CPUs right now, allowing you to offset the cost of an upgrade and giving you a kit of DDR5 to carry forward in future builds.

As an example, AMD’s recently-released Ryzen 7 7700X3D is $80 cheaper than the re-released Ryzen 7 5800X3D, despite coming in nearly 20% faster in average gaming performance. The $230 Ryzen 5 7600X3D shows similar gains, and it’s even more affordable.

In Intel’s camp, the Core Ultra 7 270K Plus and Core Ultra 5 250K Plus represent some of the best all-around value in the CPU world right now, with compelling gaming performance and chart-topping application performance, all for around $300.

You will spend more on a cheaper CPU and a kit of DDR5. There’s no indication that the memory shortage is ending soon, however. We’ve actually seen prices continue to increase despite the surge leveling off. Opting for DDR5 today buys you better gaming performance, but more importantly, it gives you a kit of memory that you can continue to use as new CPUs and platforms are released.

One of the common misconceptions in today's memory market is that once memory modules based on chips from CXMT enter the consumer market, there will be cheaper alternatives to memory sticks running DRAM from the Big Three. While availability of CXMT-powered modules certainly impacts average selling prices, these products are not cheaper than those based on ICs from Micron, Samsung, and SK hynix even in China, as noticed by @harukaze5719.

A 64GB DDR5-5600 RDIMM based on memory from Samsung or SK hynix costs 18,595 CNY ($2,745) at JD.com, whereas a module featuring the same capacity and specification, but using DRAMs from CXMT is priced at 18,999 CNY ($2,805), according to observations by @harukaze5719. It is noteworthy that a Samsung 64GB DDR5-5600 RDIMM made by Samsung and sold by Samsung costs $2,425 at Amazon.com in the U.S.

While the $60 difference seems significant, it is really just 2.2%, which can be considered negligible at these sky-high prices. Nonetheless, many expect CXMT-based memory modules to be cheaper than those carrying chips from Micron, Samsung, or SK hynix, so observing that they are even a bit more expensive than offerings with ICs from renowned manufacturers may be a bit surprising.

Indeed, because CXMT produces memory chips using an outdated fabrication technology, its DRAM ICs consume more power than those made using the latest manufacturing processes, have lower performance potential, and mediocre overclockability. Furthermore, the Chinese government heavily subsidizes both ChangXin Memory Technologies and Yangtze Memory Technologies (YMTC), which enables both to sell their memory at lower prices although their actual costs may be higher compared to those of the Big Three.

As a result, it is reasonable to expect CXMT to charge less for its chips compared to chips from renowned producers. Truth to be told, we do not know CXMT's quotes, they may be as high as those from other manufacturers and the only reason why module producers buy them is that they are the only chips available. Furthermore, companies tend to price products based on what the market will bear, not strictly on production cost or their characteristics. In fact, as everyone is capacity constrained these days, there is little incentive for CXMT to price its DRAMs significantly below those from renowned makers. However, while CXMT may indeed charge less for chips (e.g., because the Chinese government issues an appropriate directive), this lowers costs for module makers, but has a limited effect on retail prices of actual DIMMs or RDIMMs.

Whether potential savings reach end customers depends on market competition, supply-demand conditions, and the pricing strategies of module makers, distributors, and retailers. Furthermore, when prominent companies like Apple, Dell, or Corsair use a memory chip SKU, this one must pass rigorous validation processes and is usually tested in-house, or by contract manufacturers, which adds to costs and somewhat erases the price difference between suppliers.

As a result, CXMT memory chips may make the lives of hardware makers easier, but are not expected to impact the retail prices you pay.

A man by the name of Li and his partner by the name of Cheng have been arrested for a string of RAM stick thefts from various hotels in China where gaming PCs are standard room equipment. This was a pretty small-time operation, with a total of ~ 13,200 yuan (about $1,950) raised by selling 16 stolen RAM sticks on the second-hand goods market. However, the source describes the accused thieves as purveyors of "frenzied hotel room-dismantling" (machine translation).

According to the report, Li began his tech treasure crime spree in May, in Shihezi City, Northern Xinjiang. After checking into an eSports hotel, the light-fingered Li couldn’t resist plucking a pair of RAM sticks from the in-room computer. The solo RAM thief’s initial success spurred him to recruit a partner, and in the second two weeks of May, the dynamic RAM duo of Li and Cheng swept through multiple hotels in the aforementioned “frenzied” fashion.

The crime spree might be one of the first of its kind we’ve read about, but it wouldn’t make the thieves filthy rich. The source notes that 16 RAM sticks in total were taken from eSports hotels during the two-week criminal spree. They would sell the slyly stolen sticks of silicon for a total of ~ 13,200 yuan (about $1,950) in profit via a mix of online marketplaces and brick-and-mortar stores.

In the Chinese language report, it is noted that subsequent guests at the eSport hotels complained about systems with black screens, unable to boot. Customer complaints led to investigations by hotel staff, and the police were called out once the DRAM dirty deeds were definitive. Perhaps the duo would have been more successful if they had decided to just halve the RAM in the raided PCs. Guests who followed them would be less likely to loudly complain about RAM-deficient PCs than completely dead ones.

The Chinese media report ends by saying Li and Cheng are both in custody, with further investigations into the pair’s RAM piracy ongoing.

You're probably already well aware of the three largest memory vendors: South Korea's SK hynix and Samsung, and the United States' Micron. There's a fourth player up and coming in the market, though, and that's China's ChangXin Memory Technologies, more commonly known as CXMT. CXMT is pumping out RAM as fast as it can to soften the effects of the RAMpocalypse on domestic (to China) consumers, and it's also expanding capacity quickly. Given these factors and the size of the Chinese market, it should come as no surprise that motherboard vendors, including Gigabyte, are advertising fresh support for the chip.

MSI appears to have been the first major motherboard vendor to publicly promote BIOS optimizations specifically for CXMT DDR5 memory, initially through China-market BIOS releases and later with global announcements. What Gigabyte is saying is a little different: not only is CXMT RAM supported on Gigabyte motherboards for both AMD's Socket AM5 and Intel's LGA 1851, but it's supported at impressive speeds of up to 8200 MT/s on select boards.

As proof, Gigabyte offers up a CPU-Z screenshot showing a pair of 16GB Lexar modules equipped with CXMT chips achieving a speed of 8200 MT/s with a CAS latency of just 40 cycles on one of its B850M FORCE motherboards. We have to remark that Gigabyte doesn't show the voltage required to hit these speeds, likely for good reason, but the firm did it without disabling the chip's integrated graphics. That doesn't prevent the chips from posting an excellent AIDA64 latency of just 65.4 nanoseconds on the monolithic Ryzen 5 8600G used for the test. For context, typical EXPO kits usually run around 80ns, JEDEC timings are more like 100ns, and LPDDR-based systems rarely come below 110-120ns.

As Gigabyte says, "CXSH (CXMT) chip-based memory offers a welcome additional source of supply for consumers." (CXSH is the JEDEC manufacturer ID code for CXMT) Gigabyte specifically notes that it is modules with 16-gigabit and 24-gigabit ICs that are supported; that means 16GB and 24GB single-rank modules, or 32GB and 48GB dual-rank modules.

While CXMT memory is certainly a welcome additional source of supply, it's also not really clear how much of that memory is actually making its way outside of China, particularly to the US and Europe. CXMT isn't on the US Commerce Department's "Entity List" yet, so the firm's parts aren't actually banned, but the company is once again considered a Chinese military company by the Department of Defense after being briefly removed from the 1260H list in February. Lawmakers here in the states, likely laden with donations from the big three memory firms, want to see CXMT banned, while US tech companies like Corsair, HP, and Dell are already integrating ChangXin RAM into their products, and Apple wants in, too.

The reality is that, as a DIY builder, you are fairly unlikely to know or care if your memory is manufactured by CXMT. You don't usually buy RAM directly from SK hynix, Samsung, or Micron (certainly not since Micron unceremoniously executed Crucial); you buy your RAM from Silicon Power, G.SKILL, Patriot, Corsair, or one of the many other vendors who assemble DIMMs. If the RAM runs at its rated specifications, there's not a lot of reason for the individual consumer to care where it came from. It's nice to hear that the new chips are officially supported by familiar mainboards, though.

While PC builders, enthusiasts, and gamers are some of the most heavily affected by the AI-driven memory chip shortage, experts have also been warning that this will eventually hit the automotive sector. It seems that their fears have come true, as Nikkei Asia reports that two major automakers are increasing their prices precisely for this reason. General Motors said that it expects new vehicle prices to increase by 0.3% this year, changing its previous projection of vehicle costs to remain the same or even slightly more affordable. On the other side of the world, Chinese automotive giant BYD has also raised the prices of the driver assistance features by 20%, which it sells separately from its models. South Korea’s Hyundai also called on domestic chip manufacturers to strengthen the local supply chain.

GM Chief Financial Officer Paul Jacobson said during the carmaker’s earnings call that GM expects its costs to increase by $1.5 to $2 billion, which is mostly driven by increasing prices of vehicle components, especially DRAM. While most people wouldn’t equate cars with memory chips, the increasing number of features and technologies that buyers expect from new models means that engineers and designers must add more memory compared to models from the past.

Micron said in 2023 that the average vehicle used 90GB of DRAM and NAND, and it’s expecting this combined requirement to jump to 278GB by 2026, with high-end models requiring up to 2TB. A British OEM broke down the memory requirements of vehicles released this year — infotainment systems require between 4GB and 16GB, ADAS and safety systems need 8GB to 32GB, while centralized systems often have a minimum requirement of 16GB and could go as high as 64GB or more. And as AI assistants start becoming available on cars, these systems will require at least 256GB of memory. In fact, Micron predicts that cars with level 4 autonomy will need at least 300GB of RAM, as they’re essentially supercomputers on wheels.

Note that automotive memory isn’t just any high-end chip that you can get from wafer production lines. Instead, these are specialized parts that take months or years to be validated for vehicle use. That’s because they must withstand various environmental factors that cars go through, from weather and constant road vibrations to extreme heat and cold.

All this means that the current memory chip shortage will have an outsize effect on the automotive industry and push prices even higher. This is bad news for the average car buyer, who is already facing record-high prices for new vehicles. And if the chip shortages continue, we might experience delays in vehicle deliveries and a lack of availability worse than what we experienced in 2021 during the height of port closures due to the pandemic.

As Apple and other American companies seek to use memory chips from China-based CXMT and YMTC amid massive supply constraints, U.S. lawmakers want to ban exports of Chinese memory chips to the U.S., citing concerns of weakening domestic and allied suppliers and indirectly supporting the development of 3D NAND and DRAM by Chinese companies, reports the Financial Times.

John Moolenaar, Republican chair of the U.S. House China Committee, and Democratic Congressman George Whitesides asked Commerce Secretary Howard Lutnick to prevent U.S. companies from purchasing semiconductors from businesses included either on the Pentagon's Chinese Military Companies blacklist or the Commerce Department's Entity List. They also called on the administration to add CXMT to the Entity List and impose additional restrictions on YMTC.

"Dependence on Chinese memory manufacturers creates an unacceptable risk for U.S. national security, economic security, and supply chain security," the letter by Moolenaar and Whitesides reads.

Apple has been seeking approval from the Trump administration to source memory from CXMT amid a severe global DRAM supply crisis caused by the rapid expansion of AI infrastructure. Corsair, and some other suppliers of branded memory modules and SSDs have been using DRAM from CXMT and 3D NAND from Yangtze Memory for some time now.

"We are alarmed that Apple and other U.S. tech companies seek to purchase memory from Chinese semiconductor manufacturers, including those with ties to the Chinese military," the U.S. lawmakers wrote.

DRAM maker CXMT is already on the Pentagon's Chinese Military Companies list, whereas 3D NAND producer YMTC is already in the DoC's Entity List. Their presence in the lists does not outright prevent American companies like Apple from buying their products, but at a significant political risk.

Technically, American companies could buy chips from CXMT and YMTC to use inside products bound for China or other countries and continue to use memory from traditional suppliers in products aimed at the U.S. market. To prevent this, Moolenaar and Whitesides also urge the American government to coordinate with Japan, South Korea, and the EU to prevent CXMT and YMTC from taking advantage of the current supply shortage to establish themselves in allied supply chains, which they believe could ultimately leave the West strategically dependent on Chinese memory.

Moolenaar and Whitesides argue that purchases from Chinese memory manufacturers could indirectly support technologies applicable to China's military.

"Leading Chinese memory manufacturers are all closely intertwined with the Chinese military; thus, every memory purchase by a U.S. company will directly subsidize the People’s Liberation Army's development of this critical dual-use technology," the letter stresses.

The lawmakers argue that CXMT and YMTC could repeat China’s playbook in solar, steel, telecom, and EV markets: use state subsidies to undercut foreign rivals, weaken their investments, and ultimately exploit the resulting dependence for strategic leverage. That said, using Chinese memory now could permanently weaken Western production capacity and leave the West strategically dependent on China for a critical component of AI infrastructure, Moolenaar and Whitesides believe.

Homegrown DDR5 memory from China, manufactured by ChangXing Memory Technologies, has been making the rounds lately as more and more vendors start legitimizing it. However, new testing from overclocker Safedisk, shared by Uniko's Hardware, purportedly shows that it actually carries inferior performance compared to similar options from SK Hynix, alongside significant variance in the silicon between different batches.

CXMT began producing DDR5 in late 2025 despite lacking any cutting-edge EUV lithography tools. Fast forward to today, and reports of the company matching Micron's memory capacity by this year are now floating around. If true, China would become the second-largest memory maker in the world. At such scale, it's no wonder that many companies in China have already started sourcing CXMT-made RAM to fill the gap in the consumer market.

Throughout 2026, we've seen motherboard manufacturers verify CXMT's DDR5 with official BIOS optimizations that allow it to run beyond 8,000 MT/s at this point. OEMs such as Dell and HP are using CXMT RAM in their region-bound systems, and even proper PC hardware companies like Corsair are using CXMT modules. Lexar, Kingbank, Netac, Asgard, Gloaway and more are also producing retail DDR5 kits with CXMT chips.

As such, the testing features a Kingbank 48GB (2x24) DDR5-6000 kit running at CL36 and found several weaknesses despite successfully achieving an 8,600 MT/s overclock at CL44. The first revelation is that CXMT modules don't scale with voltage, meaning you can't just increase voltage in hopes of achieving higher clocks. CXMT's DDR5 apparently doesn't respond well to tuning sub-timings either, forcing you to remain stuck with baseline CAS latency (or higher, like in this case).

Different batches of CXMT-equipped memory perform differently, too, so silicon lottery plays a much bigger role than it would with other vendors. Speaking of which, SK Hynix-made DDR5 modules allegedly performed better at identical clock speeds, while CXMT's modules were less susceptible to overclocking in general. We didn't get any comparative benchmarks for any metric, so take these claims with a grain of salt.

Overall, if the testing is to be believed, it serves as counterprogramming against the popular narrative forming around China as the savior of consumer interests. CXMT's strength lies in the fact that it doesn't have to cater to opulent AI clients as much as the Big Three, which reduces opportunity cost, allowing CXMT to produce more DDR5 memory. That doesn't mean it would be cheaper, though, or at least no evidence has suggested that so far.

CXMT has remained limited to the Chinese region for now, and breaking through to the Western market would mean impressing a lot of skeptics. Not only would pricing play a big factor, but the reliability of a new DRAM manufacturer would raise serious concerns. Stories like these certainly don't help, but with CXMT's IPO on the way, it's only a matter of time before it becomes a serious mainstream contender.

ChangXin Memory Technologies (CXMT), China's largest DRAM maker, is on track to match Micron's production capacity in 2026, if Citrini Research's forecasting models are correct. If this happens, China will become the world's second-largest DRAM production base in the coming years.

The bottom-up model estimates that CXMT will finish 2026 with approximately 350,000 wafer starts per month (WSPM) of DRAM capacity, which is just 25,000 WPM less than Micron. According to the analysis, the federal government is pushing CXMT to share its DRAM technology with JHICC, Swaysure, and YMTC's subsidiary XMC to ease domestic shortages. All three companies have either built DRAM capacity already, or will do so in the short-term future, the report claims.

Swaysure has completed construction of a 140,000-WSPM fab in Shenzhen, while JHICC's Jinjiang complex contains enough cleanroom space for 120,000 WSPM, and the initial 60,000-WSPM phase is expected to receive equipment by the end of 2026. YMTC is also projected to operate about 50,000 WSPM of DRAM production at Wuhan Fab 3. If all these facilities initiate operations in the coming years, then China will have a total DRAM capacity of 600,000 WSPM (not counting Samsung's and SK hynix's fabs in China), which is dramatically lower compared to South Korea, but ahead of Japan, Taiwan, and the U.S. combined.

But China is not going to stop developing its DRAM industry, and by 2030, its total capacity will increase to around 1.41 million WSPM, according to Citrini. CXMT alone is projected to build new production capacities in Beijing, Hefei, and Shanghai, to expand its production capability to 950,000 WSPM in 2030, assuming everything goes as planned.

The supply model assumes that about 400,000 WSPM of CXMT output will remain on D1a, another 400,000 WSPM will migrate to D1b, and roughly 150,000 WPM will produce D1c devices.

Enough fab tools?

Citrini admits that producing China's outlook is considerably more difficult than predicting the development of established DRAM makers. On the one hand, there is rapidly expanding fabrication infrastructure in China, abundant state-backed financing, and government-directed technology transfers. On the other hand, among the key near-term limitations remains lithography equipment availability, particularly if the proposed MATCH Act restricts sales of advanced immersion DUV tools to select Chinese companies.

However, the author expects SMEE's domestic immersion DUV scanners to enter volume production around late 2026 or early 2027 following beta testing, as well as SiCarrier/Yuliangsheng introduce its own production-ready DUV platform in 2028. Perhaps a bit optimistically, the analysts predict that availability of lithography tools is not expected to constrain Chinese production beyond 2028, at least for mature logic and DRAM nodes. Yet, for obvious reasons, if the MATCH Act works as planned and disrupts supply of advanced immersion DUV tools to DRAM makers, production capacity expansions will not occur in the next couple of years, the report suggests.

Still, both SMEE and SiCarrier will need time to ramp up production of their lithography systems, whereas DRAM makers must learn how to use them efficiently, so we would not be as optimistic as the authors and would not expect Chinese tools to produce meaningful DRAM volumes before the early 2030s. Still, the key takeaway here is that China is on track to become a major DRAM maker rather sooner than later.

Unprecedented demand

Citrini Research projects total DRAM demand to reach 157.5 exabytes (EB) per year by 2030, including 75 EB of commodity DRAM for agentic AI CPUs, 25 EB of commodity DRAM for conventional cloud servers, 20 EB of commodity DRAM for client devices, and 37.5 EB of HBM4E as well as HBM5 for AI accelerators (15 EB and 22.5 EB, respectively).

Meanwhile, Citrini expects the whole industry to only produce around 37.5 EB of HBM4E/HBM4 memory (mostly by Micron, Samsung, and SK hynix) as well as 91.3 EB of commodity DRAM (including output in China) in 2030, leaving a deficit of 28.7 EB, or roughly 25%.

That said, the rapid expansion of DRAM production in China could be the industry's best hope to maintain relatively low prices of memory, something that will be particularly beneficial for the market of consumer devices that are sensitive to memory prices, analysts from Citrini believe. Yet, the author argues that most of this new capacity would satisfy China's own demand rather than eliminate the global shortage. Furthermore, even if companies like CXMT can expand their fabs faster, that additional capacity will mostly be consumed by domestic needs, according to Citrini.

It should be noted that to make more memory, DRAM makers need more fab tools, primarily 193nm immersion scanners. Yet, companies like ASML, Canon, and Nikon cannot increase output of immersion DUV systems quickly as these are extremely complex machines containing tens of thousands of parts. While Chinese memory companies certainly pin their hopes on local producers like SMEE and SiCarrier, neither has delivered a single commercial immersion system, and after they do, it will take them years to ramp up production of such tools.

SK Hynix CEO Kwak Noh-jung says that 2027 will be the "worst year" for the ongoing memory shortage in comments shared with Reuters. The remark comes on the heels of SK Hynix successfully marking the largest-ever IPO for a foreign company on the U.S. stock market, raising $26.5 billion. Although Kwak points to next year being the worst for RAM shortages, the executive expects the memory crunch to last until 2030.

"We forecast that next year will be the worst year in the industry's history from the supply perspective," Kwan told Reuters. "We still forecast that customer demand will remain higher than our ​supply capacity even beyond 2030. But we are doing our best to solve the problem."

In March, SK Group chairman Chey Tae-won also suggested shortages would last until 2030, and the company has previously pointed to 2027 as a key shortage point, alongside Samsung. DRAM demand is largely driven by the HBM used in AI accelerators, which require far more sophisticated manufacturing and packaging processes compared to consumer DDR5. On top of advanced manufacturing, HBM also consumes more wafer capacity than DDR5, forcing major memory brands to reallocate supply and double down on an already sticky supply situation.

Forecasts like this are tricky. It's in SK Hynix's financial interest for memory shortages to continue, even well beyond 2030. SK Hynix has set a record for quarter-over-quarter revenue, and rival Micron has seen its stock value increase 213% this year, pushing its share price to around $990.

However, Kwan's remarks aren't just a bid to rally behind SK Hynix stock. Over the past few months, we've seen Micron and SK Hynix ink long-term supply agreements (LTAs). These agreements commit supply over multiple years to particular companies and define a price floor and ceiling during the agreement term. Although LTAs don't directly influence market prices, they secure demand, and we've seen a lot of LTAs over the past several months to bind DRAM supply.

Although memory (and NAND) prices will remain elevated for at least the next several months, we've seen some signs of the market cooling. Earlier this month, a TrendForce report showed DRAM contract prices up 15% to 18% quarter over quarter for Q3 2026. That's a large increase, but far lower than the QoQ increases we've seen previously.

We're nearing some semblance of stability in the memory market, just stability at vastly elevated prices. How long that lasts is anyone's guess. Although memory brands like SK Hynix have visibility into market trends, those can rapidly change. Just this year, we've seen a massive pivot toward AI spending going toward CPUs, pushing Intel's stock to record highs while shedding around $1 trillion in Nvidia's market cap; a year ago, that would've been almost impossible to predict.

After announcing it last month, the first AMD EXPO Ultra Low Latency (ULL) memory kits are finally available; and with up to an 80% jump over already inflated RAM prices in tow. HardwareLuxx was able to snag a kit of G.Skill's new Trident Z5 NeoX RGB memory to see how ULL performs, and the results don't quite justify the extra cost. At most, the publication found just a 4% improvement compared to non-ULL kits.

That aligns with AMD's original claims about ULL. When the company announced the initiative last month, it also cited a 4% improvement over standard EXPO. However, AMD claims that ULL offers a 4% improvement on average, while HardwareLuxx only found that large of an improvement in a single game: F1 25. AMD reached out following publication and noted that its 4% average was across a wide variety of titles, and that there are inconsistences between each game, with some titles showing little to no benefit.

HardwareLuxx tested a 2 x 16GB kit at 6,000 MT/s with primary timings at 36-36-36-76. Compared to standard EXPO/XMP DIMMs, a lot of the optimization with ULL DIMMs comes from tuning the subtimings. The primary timings are largely the same as what you'll find on a standard EXPO kit, short of tWR (write recovery), which is lower on the EXPO ULL kit. Better binning of the memory ICs allows for more aggressive subtiming optimization, as HardwareLuxx notes, rather than relying on timings primarily concerned with stability.

In games, the ULL kit showed clear performance improvements, no matter how minor they were. In F1 25, the ULL kit outclassed a DDR5-6000 CL26-36-36-96 kit by 4.2%, and beat out JEDEC standards at 5600 MT/s by nearly 14%. Similarly, in Cyberpunk 2077, the ULL kit was around 3.7% faster than the standard EXPO kit, and 12.7% faster than JEDEC standards. The outlet also tested Arc Raiders, Baldur's Gate 3, and Counter-Strike 2; however, the ULL kit didn't offer a meaningful performance improvement in any of these titles.

The outlet also tested 7-Zip, though with only minor differences between ULL and non-ULL memory. The most interesting results are from the microbenchmarks available in AIDA64, which HardwareLuxx also ran. ULL showed largely similar copy and read throughput, but write throughput was 9.4% higher with ULL compared to stock EXPO.

Although there's a performance benefit, it's minor, and HardwareLuxx notes that "manual tuning still allows for the maximum possible optimization."

The particular kit that HardwareLuxx tested doesn't have the extreme ULL price increases we've seen elsewhere. It's currently available for sale for $530, which is only $20 more than a kit of G.Skill's Trident Z5 Neo RGB memory at DDR5-6000 CL36. It's actually gone down in price (it was originally listed at $550), while kits with more aggressive timings have increased in price.

The NeoX DDR5-6000 CL26 kit, for example, has jumped up $50 to $1,150 — yes, that's for a 2 x 16 GB kit still — while the CL28 kit has jumped up to $1,030 (a $30 price increase). Two weeks ago, we saw non-ULL kits selling at $560 and $700 for CL28 and CL26, respectively, creating a large delta in price between ULL and non-ULL kits. Now, those kits are selling for $700 and $900, respectively.

The introduction of ULL couldn't have come at a worse time, as the ongoing DRAM shortage continues to raise the cost of building a PC around the world. Adding a premium on top of those already inflated prices is tough to justify, even if that premium is modest — especially for mainstream CL36 and CL30 kits, the "ULL tax" is essentially null. The good news is that you can largely achieve what ULL offers on your own, at least given that you have the patience to sit through tuning your memory for single-digit gains.

JEDEC has released its new specification that aims to push down the pricing of the ultra-expensive HBM that powers the fastest AI processors. While the new standard will not help relieve the DRAM shortage as it uses large HBM4 DRAM devices, it can make high-bandwidth memory a bit cheaper as it enables attaching SPHBM4 memory stacks without advanced packaging and using inexpensive organic substrates.

The standard's body published the specification of SPHBM4, Standard Package High Bandwidth Memory (JESD330-4), that combines HBM4 DRAM ICs with standard packaging and a fast 'narrow' 512-bit interface. Here are the details.

HBM4 performance with a 512-bit wide interface

Although 1024-bit and 2048-bit interfaces used by HBM3 and HBM4 memory deliver unbeatable performance, their wide interfaces consume significant silicon area inside processors, they require expensive interposers, and advanced packaging technologies with limited capacity, such as TSMC’s CoWoS, for integration with host processors. The upcoming SPHBM4 memory continues to use the same HBM4 DRAM stacks as JESD270-4, but swaps the conventional HBM base die for a new SPHBM4 PHY/buffer die featuring a narrower 512-bit interface that enables mounting on standard organic substrates without using sophisticated packaging methods for integration. To offset the effect of the narrower interface, SPHBM4 supports considerably higher data transfer rates ranging from 22.4 GT/s to 46.0 GT/s.

Instead of connecting to the host processor using a 2048-bit memory interface like HBM4, SPHBM4 uses 32 independent 16-bit DDR channels organized into eight Quad Channels. Since 'Quad Channel' is a new term, let us explain how things work. Internally, an HBM4 stack contains 32 memory channels, each 64 bits wide, for a total external interface width of 2048 bits. SPHBM4 needs to 'convert' the 2048-bit internal I/O onto a 512-bit external interface, which is why it groups every four HBM4 channels into a Quad Channel. As a result, externally, a Quad Channel exposes 64 data pins (4 × 16 bits), which replace the 256 data pins (4 × 64 bits) that those four HBM4 channels would normally require. To preserve bandwidth, these 64 pins operate at four times the data rate of the original HBM4 interface.

While SPHBM4 dramatically increases I/O bandwidth, it does not make the DRAM array itself faster. The HBM4 memory core retains the same fundamental architecture and timings, including core frequency, row activation, precharge, and refresh operations, though the additional PHY is expected to introduce some latency. For example, the DRAM core runs at only one-quarter of the external interface frequency, which means 2 GHz in the case of SPHBM4 with a 32 GT/s speed bin.

The major change is the new base die, which implements a high-speed SerDes-like PHY that maps each 16-bit external channel to four conventional 64-bit HBM4 channels. As a result, SPHBM4 introduces equalization, lane training, BER requirements, and other high-speed signaling features that are unnecessary in HBM4’s slower, wide parallel interface. To support transfer rates of up to 46.0 GT/s/s per pin, each Quad Channel uses a shared command/address interface protected by forward error correction (FEC), while data transfers rely on dedicated differential write (WCK) and read (RCK) clocks, as well as ECC and error-reporting signals.

When it comes to capacity, SPHBM4 can use stacks containing 4, 8, 12, or 16 DRAM dies featuring 24 Gb or 32 Gb densities, so the largest standardized SPHBM4 configuration is a 64 GB memory stack built from sixteen 32 Gb DRAM dies, identical to the maximum capacity supported by HBM4E.

Cheap HBM at last?

The standard supports bump pitches greater than 90 µm and channel reaches up to 20 mm, which are two features that enable dropping the expensive interposer and using less-expensive organic substrate routing. However, getting rid of the interposer and CoWoS (or similar) packaging does not automatically make SPHBM4 inexpensive. SPHBM4 still requires massive HBM4 DRAM ICs, 2.5D packaging, a complex base die (which is likely costlier than the one used by conventional HBM4), and advanced package assembly with through-silicon vias. In addition, SPHBM4's narrow interface consumes significantly less die perimeter and silicon area inside processors, which makes it more attractive to companies that strive to install more compute capability and/or intend to install more memory stacks around their processors. However, we are still talking about a niche high-performance memory technology that will address select applications and will barely rival HBM4 directly.

When it comes to maximum performance, HBM4 moves the data at 8 GT/s (though most controllers and chips support higher data rates), so one HBM4 stack can offer bandwidth of 2 TB/s. HBM4E is set to up data transfer rate to 12 – 12.8 GT/s, therefore increasing peak bandwidth to 3 – 3.3 TB/s per stack. By contrast, one SPHBM4 with a 46 GT/s interface can hit 2.944 TB/s, though do not expect the initial versions of SPHBM4 to hit the maximum speed. Therefore, it is likely that HBM4, HBM4E, and C-HBM4E will maintain a performance lead in terms of bandwidth over SPHBM4 in the foreseeable future.

HBM4 latency will still probably have an edge over SPHBM4. HBM4 essentially connects to its host processor almost directly through a very simple interface. By contrast, SPHBM4 inserts a much more sophisticated PHY that performs serialization/deserialization, lane training, FEC handling, and other operations that can add a few nanoseconds of latency. This may not be a big problem for some applications, but inference benefits a lot from low latencies.

When it comes to power and voltages, HBM4 and SPHBM4 share the same DRAM core voltage because SPHBM4 reuses standard HBM4 DRAM stacks. However, I/O is different: HBM4 leaves the interface voltage up to memory vendors and allows implementations at 0.7V, 0.75V, 0.8V, or 0.9V, depending on the desired balance between power, speed, and signal integrity. By contrast, SPHBM4 standardizes the external I/O at 0.75V.

Also, HBM4 moves data over a very wide interface with many slow parallel links that tend to be very energy efficient. By contrast, SPHBM4 moves the same amount of data through one-quarter as many wires, which run roughly four times faster. High-speed data transfer tends to be less energy efficient than 'slow' data transfers over a wide interface. Keeping in mind SPHBM4's rather sophisticated PHY that converts a wide interface into a narrow interface, which is likely a power-hungry process. Nonetheless, the 4X lower number of drivers and receivers could tangibly reduce the power consumption of SPHBM4. That said, without implementation details from DRAM makers or a processor developer, it is impossible to conclude which memory type has lower power consumption.

Last but not least, SPHBM4 essentially trades manufacturing challenges that arise from using silicon interposers for an engineering challenge of developing an extremely sophisticated base die/PHY. Developing and manufacturing such a base die should not be a problem for foundries. However, it remains to be seen whether DRAM makers can design and produce SPHBM4 with decent power efficiency. After all, both Micron and SK hynix work with TSMC to build C-HBM4E and HBM4E base dies, whereas Samsung's memory division uses base dies produced by Samsung Foundry.

China factor

One interesting aspect of SPHBM4 is whether Chinese developers of AI accelerators can benefit from this technology. In theory, Chinese developers like Biren, Huawei, Moore Threads, and other blacklisted companies that cannot use TSMC's chip manufacturing or packaging services could become one of the biggest beneficiaries of SPHBM4, perhaps even more so than the U.S.

First up, a smaller shoreline directly benefits chips that are made using trailing nodes, as it enables packing more compute capability into them without sacrificing memory bandwidth or capacity. Secondly, Chinese OSATs currently do not offer CoWoS-like technologies, so eliminating the interposer and using advanced organic substrates is a benefit.

However, SPHBM4 still requires HBM4 DRAM stacks, and today, Samsung, SK hynix, and Micron are the only companies capable of producing them, while China-based CXMT can barely make HBM2E. Furthermore, building a 46 GT/s PHY is very hard and will likely be challenging for Chinese IC developers.

Nonetheless, assembling SPHBM4 packages on organic substrates is arguably more aligned with China's existing manufacturing base, so if local DRAM makers eventually develop competitive HBM4-class memory, SPHBM4 could substantially reduce one of the country's remaining infrastructure gaps.

Summary

JEDEC's SPHBM4 looks like a promising standard that can potentially address a broader range of applications than HBM4 itself due to lower integration cost. Still, HBM4, HBM4E, and C-HBM4E will maintain performance leadership, which will make them a preferable choice for flagship AI accelerators in the coming years.

ChangXin Memory Technologies, or CXMT, has just received official validation from MSI for its high-speed DDR5 memory modules. The manufacturer has released new beta BIOSes across its AM5 lineup, unlocking stable frequencies up to 8,200 MT/s for 3GB CXMT chips on dual-DIMM motherboards. Previously, RAM using these modules was limited to around 6,800 MT/s despite the hardware itself being capable of much more.

The test BIOS comes from MSI China — there's no announcement on global channels for some reason, and it's only available for select motherboards at the moment. MSI tested region-bound retail kits from Lexar and KingBank on boards with both two slots and four slots, but we only have screenshots for the former, courtesy of Videocardz.

One test was conducted using 24GB sticks (2x24) on the MEG X870E Unify model and a Ryzen 7 9700X CPU. This was mostly a standard kit since it came with a 6,000 MT/s EXPO profile. The other config consisted of 16GB sticks (2x16) with an EXPO profile already running at 7,200 MT/s, so it was somewhat cherry-picked silicon, paired with a Ryzen 5 9600X on a MAG B850 MPower motherboard.

The results showed that on dual-DIMM motherboards, 24Gbit (3GB) modules from CXMT were able to clock up to 8,200 MT/s, passing MemTest with 101% coverage. Conversely, 16Gbit (2GB) chips were also stable at 8,000 MT/s on the same test bench with 101% coverage. Moving over to quad-DIMM, the patch notes for the BIOS say they've "also been raised to DDR5-7200," where the limit was stuck at 6,800 MT/s prior.

This is not the first time MSI has optimized Chinese RAM for some of its motherboards. Earlier this year, the company did the same thing for Intel's 800-series models in China. Anyway, these specific BIOSes are meant for the AM5 socket and they're based on existing stable releases — just patched with unlocked memory overclocking features. More motherboards should be supported soon, but for now you can check out the MSI China's community release channel if you want to try one yourself.

Memory prices will keep climbing through the third quarter of 2026, but the blistering increases of the past few quarters are set to cool as consumer buyers reach the ceiling of what they can afford, according to a detailed TrendForce report. The research firm's latest memory pricing survey projects conventional DRAM contract prices to rise 13% to 18% quarter-over-quarter (QoQ) in Q3 2026, with NAND Flash contract prices increasing 10% to 15%; substantial gains, but a marked slowdown from the roughly 60% jumps recorded in the second quarter.

According to the report, the cooldown is driven by consumer electronics manufacturers' unwillingness and inability to absorb higher memory costs after months of relentless price increases, rather than by improved supply. In other words, memory remains in short supply, but consumers are no longer willing to keep paying ever-higher prices.

AI continues to be the market's driving force. Demand for AI inference systems and hyperscale data centers remains strong enough to keep both DRAM and NAND supply constrained, while memory manufacturers continue shifting production capacity toward higher-margin server products. This leaves less capacity available for consumer memory, preventing prices from falling even as demand from PCs and smartphones weakens.

As a result, the memory market is increasingly split between enterprise and consumer customers. On the server side, TrendForce expects demand to remain healthy through 2027 as improving CPU availability supports continued deployments of AI servers built around x86 processors and registered DIMMs (RDIMMs). Although server DRAM is expected to remain undersupplied during the third quarter, price increases should moderate because a portion of purchases is covered by long-term supply agreements.

Consumer markets appear to paint a very different picture. Notebook manufacturers are expected to keep replenishing inventories, but higher memory costs are gradually feeding through into retail pricing, a trend TrendForce believes could weigh on PC shipments for the rest of the year. Smartphone vendors face similar pressure, with many expected to raise handset prices to offset persistently high LPDRAM (low-power DRAM) costs while simultaneously becoming more cautious with production plans as consumer demand softens.

The same pattern is beginning to emerge across storage products. PC manufacturers accumulated substantial client SSD inventories during the first half of 2026, reducing their willingness to accept another round of price increases. Suppliers have responded by taking a more flexible approach during contract negotiations, helping to moderate SSD pricing even as enterprise storage continues to benefit from AI infrastructure spending.

Not every segment is seeing the same level of demand. TrendForce notes that NVIDIA's RTX PRO 6000 Blackwell has yet to generate the expected wave of GDDR7 demand, while weaker notebook shipments have also softened demand for graphics memory. At the opposite end of the market, retail products such as USB flash drives and memory cards remain sluggish as higher upstream costs become increasingly difficult to pass on to consumers.

For PC builders, the report suggests that meaningful price relief is still some way off. Memory prices are continuing to rise because AI infrastructure remains the industry's top priority. However, the pace of those increases is slowing as consumer demand reaches its breaking point.

SK hynix this week announced that it would invest an additional KRW 100 trillion ($64 billion) in its Cheongju campus to expand production of 3D NAND and HBM packaging at the site. Given the vast investment, expect the company to add some massive production capacity, but unfortunately that production capacity is going to kick in only several years down the road. But that investment pales in front of the company's plan to invest $712.5 billion in its South Korean operations.

The massive KRW 100 trillion ($64 billion) in its Cheongju campus investment is only a part of SK hynix's grand plan to invest KRW 1.1 trillion ($712.5 billion) in a variety of projects in South Korea. In particular, the company intends to invest KRW 400 trillion ($259.5 billion) in its all-new Southwestern semiconductor cluster as well as KRW 600 trillion ($389.3 billion) in its Yongin site. While the Cheongju investment is considerably lower than investments in other campuses, it is the only project that is actually detailed enough.

$64 billion go to Cheongju to support NAND and packaging

SK hynix claims that it intends to build a 3D NAND fab, install manufacturing equipment, and expand its advanced packaging capabilities for HBM back-end processing at its Cheongju campus in the Chungcheong region. The company intends to start building its M17 fab next year, so the earliest timeframe it comes online is sometimes in 2029 at the earliest. The fab will cost around KRW 80 trillion ($51.8 billion), whereas the new P&T7 packaging and test facility will cost KRW 20 trillion ($12.945 billion).

SK hynix's campus in Cheongju houses some of the company's primary fabs that manufacture 3D NAND flash, including M11, M12, and M15, and historically it was the company's main 3D NAND memory manufacturing center. However, because multi-layer 3D NAND and high-bandwidth memory (HBM) stacks use similar packaging technologies, it is now evolving into a site that also makes HBM stacks: M15X produces actual DRAM dies, whereas P&T3 performs packaging operations.

But the investment in SK hynix's Cheongju NAND and HBM assembly operations pales when compared to how much money the company plans to pour into other projects.

$389.3 billion go to Yongin Semiconductor Cluster to boost DRAM output

SK hynix plans to invest approximately $389.3 billion in the Yongin Semiconductor Cluster, which is the company's largest investment commitment ever and which will make the campus its largest DRAM production site. Meanwhile, Yongin is a greenfield site today.

The first fab in Yongin is expected to commence operations in May 2027, while the remaining fabs will be added sequentially. It takes about a year or 1.5 years or so to fully ramp a DRAM fab, so expect the facility to impact the memory market in 2028 – 2029. Under the company's newly announced plan, construction of all four fabs is now targeted to complete the fourth fab by 2033, instead of the original 2045 timeline. The $389.3 billion investment extends beyond 2033.

$259.5 billion go to Southwestern Semiconductor Cluster

Unlike Yongin, the Southwestern Semiconductor Cluster does not even exist. It is currently a planned project, and SK hynix has not even selected a specific site within southwestern Korea. The company says the exact location will be determined after evaluating land availability, electricity, water, transportation, and other infrastructure requirements in consultation with central and local governments.

The cluster is envisioned as SK hynix's next major manufacturing base after Icheon, Cheongju, and Yongin. For now, the planned investment totals approximately $259.5 billion, though given that the project's completion is decades away, we can expect that number to change upwards or downwards depending on the market conditions and the cost of wafer fabrication equipment.

The investment will be phased over many years and cover land acquisition, fab construction, and production tools. SK hynix says preparations must begin now because developing a new semiconductor cluster — including site selection and infrastructure — takes many years. For example, development of the Yongin Cluster took about nine years, according to SK hynix.

Not alone

SK hynix is not alone in investing in South Korea. Samsung on Thursday announced plans to spend some KRW 140 trillion ($90.98 billion) in its operations in South Korea’s Chungcheong region.

Under the plan, Samsung Display will expand OLED production in Asan; Samsung Electronics will build five HBM production lines in Onyang and modernize HBM-related facilities in Cheonan; Samsung SDI will establish a battery production line in Cheonan to validate next-generation technologies before deploying them globally; and Samsung Electro-Mechanics will expand AI server package substrate manufacturing in Sejong.

The price of DDR5 memory is setting new highs these days as demand badly outstrips supply. In a bid to save money, Meta is recovering legacy DDR4 memory from used servers and is installing it into new machines using its in-house developed Vistara ASIC that enables it to connect old memory modules to its latest servers running AMD EPYC 'Turin' processors that only support DDR5 memory.

Interestingly, Meta is not the only company developing such a solution. Panmnesia, a startup from South Korea, has developed an off-the-shelf CXL controller and switch that enables servers to attach considerably larger memory pools without extending latency, which differentiates Panmnesia’s solution from competing CXL offerings.

Custom ASIC enables DDR4 memory to work with new servers

Vistara is Meta’s first-gen custom CXL memory expander ASIC designed to attach outdated DDR4 memory to modern servers. The chip implements a CXL 2.0 Type-3 memory expander over a PCIe 5.0 x16 interface and bridges standard DDR4 RDIMMs to host processors. Each ASIC supports two independent 72-bit DDR4 memory channels and can provide up to 256 GB of capacity using 64 GB DIMMs. At present, Meta deploys 128 GB per ASIC using 32 GB DDR4 modules recovered from decommissioned servers.

Meta deploys Vistara in its MemServer platform, where two ASICs connect to a single 158-core AMD Turin processor over PCIe 5.0 x8 links. Each server combines 768 GB of DDR5-6400 local memory with 256 GB of CXL-attached DDR4-2400, which expands memory capacity to 1 TB. The software stack transparently exposes CXL memory as a separate NUMA node and enables Linux to migrate cold pages to the slower DDR4 tier (with 76 GB/s of bandwidth) and retain frequently accessed data in local DDR5 (with 614 GB/s of bandwidth).

The ASIC is based on three RISC-V processor cores for secure boot, device initialization, firmware management, and health monitoring. Meta claims it has optimized its CXL controller and memory pipeline to reduce protocol overhead, minimize queuing delays, and lower idle round-trip latency to around 50ns. The chip also incorporates advanced reliability features, including Reed-Solomon two-symbol error correction and x4 chip-kill support.

Not only Meta's Vistara

Meta is not the only company that wants to attach legacy DDR4 memory to newer servers that rely on DDR5 memory and save some money. While Vistara is available exclusively to Meta, there is a new CXL expander solution from Panmnesia that will be available to other companies.

"There has been a perception that putting a switch between the CPU and devices makes it hard to meet the memory-access latency these systems expect, so directly attached multi-headed devices (MHDs) stayed the norm even though they were harder to scale," said Myoungsoo Jung, chief executive of Panmnesia. "Our work shows this is not an inherent limit of CXL or CXL switches — it is a trait of early-stage CXL, and one that fades as the standard and the products around it mature. With a fabric switch that carries our next-stage CXL controller, scalability, low latency, and stable performance can come together."

CXL is a protocol that sits on top of the PCIe physical interface. As a result of this, many early CXL implementations were built by modifying existing PCIe IP, which is why such implementations inherited architectural characteristics optimized for PCIe rather than for memory-semantic communications, which added substantial latency, according to Panmnesia. By contrast, its new CXL controller IP features a redesigned data path that replaces separate per-layer buffers with shared buffers to eliminate much of the synchronization overhead. In addition, it features additional latency optimizations throughout the protocol stack that offset the additional hop introduced by the switch.

The accompanying CXL fabric switch introduces Port-Based Routing (PBR), which removes the tree-topology limitations of conventional Hierarchy-Based Routing (HBR) used by PCIe and early CXL implementations. The fabric switch still supports both PBR and HBR to enable flexible system topologies, optimized traffic routing, and stable performance. In practice, it enables companies like Meta to install more DDR4 memory into their modern servers without major performance degradation because of high latency.

Panmnesia claims that while early CXL deployments could connect only a handful of compute nodes to shared memory pools, its fabric scales to up to 64 nodes, which means greater flexibility for hyperscalers that tend to run thousands of servers, but which now have to rationalize usage of expensive DRAM.

Panmnesia says its next-generation CXL technologies are progressing toward commercialization. The company has pre-release silicon for its PCIe 6.4/CXL 3.2 Fusion Switch and has completed development of its PCIe 7.0/CXL 4.0 Combo IP, which supports the latest features introduced by the CXL 4.0 specification.

Edit 07/02/2026 1:00 am ET: Corrected the DRAM voltages and pricing with new information provided by G.Skill.

G.Skill’s AMD EXPO ULL (Ultra Low Latency) memory kits, strong contenders among today's best RAM options, have arrived on the retail market. While AMD previously stated that EXPO ULL memory kits would be “effectively the same price” as vanilla EXPO memory kits, this has proven by far not to be the case. In reality, these new EXPO ULL kits carry a substantial premium that can scale up to 80% over their regular counterparts, according to the latest Newegg listings.

The Trident Z5 NeoX is G.Skill’s latest memory series featuring ULL optimizations. The new lineup is available in three striking color options, black, glossy black, and white, that cater to a variety of system aesthetics. G.Skill has not released the official product pages for the Trident Z5 NeoX series, so detailed information on the full range of planned capacities and speeds is unavailable. Newegg is currently selling the Trident Z5 NeoX memory kits in a 32GB (2x16GB) dual-channel configuration clocked at DDR5-6000.

To make it a meaningful comparison, we compared the Trident Z5 NeoX memory kits to their direct counterparts, the standard Trident Z5 Neo series. We deliberately excluded similarly specced lineups such as the Aegis or Ripjaws series, which are tailored for budget-conscious buyers, as well as omitting the Trident Z5 Royal Neo, which targets a more posh crowd.

According to G.Skill, the EXPO ULL memory kits share similar pricing to the non-ULL counterparts. However, there may be a price difference between ULL memory kits and regular memory kits in stores. The reason is that some retailers likely still have older inventory that was bought at a lower cost. They may try to balance the prices of new and old stock, resulting in the price gap. This may occur specifically when prices are on the rise.

According to Newegg, the Trident Z5 NeoX DDR5-6000 C26 memory kit presently carries a 57% premium over the standard Trident Z5 Neo DDR5-6000 C26 memory kit. The price gap widens further for the DDR5-6000 C28 variant, which commands a 79% premium. Meanwhile, the Trident Z5 NeoX C30 and C36 versions are 14% and 9% more expensive than their regular Trident Z5 Neo counterparts, respectively.

On a superficial level, it’s basically impossible to grasp the true extent of the level of optimization that goes into AMD EXPO ULL memory kits just by looking at product listings. It’s customary for memory manufacturers and retailers to only list the four primary memory timings: CAS Latency (CL), Row Address to Column Delay (tRCD), Row Precharge Time (tRP), and Row Active Time (tRAS). Unfortunately, these memory timings only tell part of the story, since most of the secondary timings for AMD EXPO ULL certification, including Refresh Interval (tREFI), Row-to-Row Delay Short (tRRDS), and Write Recovery Time (tWR), are not publicly visible.

Returning to the example of the Trident Z5 NeoX memory kits, what immediately stands out is the dramatically reduced tRAS value, which is up to 67% lower than that of standard Trident Z5 Neo memory kits. In short, tRAS dictates how many clock cycles must pass after a row opens before the memory controller can close it. The reason for high tRAS values in DDR5 resides in the architecture itself. Unlike DDR4, which operates with a single 64-bit channel, DDR5 employs two independent 32-bit subchannels. This change, alongside the doubling of burst length from BL8 to BL16, enables each memory command to transfer twice as much data through the channels in a single operation. As a result, the memory rows remain open longer, leading to the higher tRAS values typically seen in DDR5 modules. What makes the AMD EXPO ULL memory kits unique is that they drive tRAS values back down to levels reminiscent of high-performance DDR4 from the old days.

Another significant advantage of AMD EXPO ULL memory kits is their ability to maintain tight timings at the same voltages. Lower DRAM voltages translate to lower power consumption, lower operating temperatures, and more headroom for overclocking.

If you’re looking for the simplest explanation as to why AMD EXPO ULL memory kits carry a hefty premium over regular AMD EXPO memory kits, it boils down to the extensive optimization and meticulous binning process involved in their production. Memory vendors spend more time testing and sorting memory chips that can operate at the tightest timings and lowest voltages. It’s a very time-consuming and labor-intensive process. As a result, the overall manufacturing costs for EXPO ULL kits are higher than for standard kits. In many ways, you aren’t just paying a premium for the hardware itself, but also for the vendor’s time and the guarantee that the memory modules run at near-maximum potential right out of the box.