Apple's rumored M7 Ultra targets 1.5TB of memory and Blackwell-class AI performance, report claims — monster 2028 offering would depend on memory shortage easing

News
By
Published

Bloomberg’s Gurman describes a compressed M7 timeline, but its top memory config hinges on shortage relief.

Apple logo
(Image credit: Getty / NurPhoto)

Apple's planned M7 Ultra chip is being designed to support up to 1.5 TB of unified memory and to push AI performance toward the class of Nvidia's Blackwell accelerators, according to a new Bloomberg report published by Mark Gurman. But whether the lofty top memory config can ship at all will depend on the state of the memory market, and the part isn't expected until 2028. The same report says Apple has compressed its Mac silicon timeline, taping out the M7 roughly six months after the M6.

We've already heard that Apple plans to release a base M6 chip this fall for entry-level Macs, then skip the Pro, Max, and Ultra versions of that generation and move straight to the M7 line. However, Gurman now reckons that we'll see a base M7 in the first half of 2027, M7 Pro and M7 Max at the end of 2027, and the M7 Ultra in 2028. Apple reportedly began taping out the M7 about six months after it started the same process for the M6, which is what has enabled the company to pull the schedule forward.

Swipe to scroll horizontally
Apple's rumored M-series roadmap

Chip

Rumored timing

Reported details

M6 (base)

Fall 2026

Entry-level Macs only; Pro/Max/Ultra skipped this generation

M7 (base)

H1 2027

Taped out roughly six months after M6

M7 Pro / M7 Max

End of 2027

N/A

M7 Ultra

2028

AI performance "closer to" Nvidia Blackwell-class accelerators; up to 1.5TB memory (~2x the M5 Ultra's planned capacity), supply-dependent

M8 (Soko)

By 2028

Built on a 1.4nm process; further AI gains

Cardinal

2028 generation

High-end Macs

The 1.5 TB target for the M7 Ultra is roughly twice the capacity Apple has planned for the M5 Ultra, per Gurman, who tied the configuration directly to memory availability. Apple already pulled the 128GB Mac Studio this year over supply constraints as DRAM prices climbed, and a 1.5 TB part would call for far more of the same scarce, high-cost memory.

Latest Videos FromTom's Hardware

Apple's current M3 Ultra already reaches 819 GB/s of memory bandwidth by fusing two Max dies, and it's the Ultra tier, not the base chips, that carries the heaviest local-AI workloads. Gurman describes the M7 Ultra as a large step up in AI performance rather than stated parity with Nvidia's data-center silicon. "I'm told the processor dramatically upgrades AI performance, bringing it closer to the class of dedicated AI accelerators such as Nvidia Corp.'s Blackwell," Gurman wrote in his report.

Apple is also preparing an AI server built on the M5 Ultra under the code name J246 for deployment soon, with a second server chip based on the M7 Ultra planned for 2029, according to the report. The 2028 generation, which includes an M8 chip code-named Soko and a high-end Mac part called Cardinal, moves to a 1.4nm process. That aligns with TSMC's A14 node, which the foundry has scheduled for mass production in the second half of 2028.

None of the dates or specifications have been confirmed by Apple.

Google Preferred Source

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

TOPICS
Luke James
Luke James
Contributor

Luke James is a freelance writer and journalist.  Although his background is in legal, he has a personal interest in all things tech, especially hardware and microelectronics, and anything regulatory. 

10 Comments Comment from the forums
  • thestryker
    The capacities being talked about here seem extremely unlikely unless they're going to use a lot of packages which is something they haven't done so far. To reach 768GB on a 1024-bit interface they would need 64, 32 or 16 packages of 12GB, 24GB or 48GB capacity. This puts them in the crosshairs of enterprise LPDDR5X packaging. The only way they could double this up is by using the same packaging as is used by nvidia for SOCAMM.
    Reply
  • usertests
    thestryker said:
    The capacities being talked about here seem extremely unlikely unless they're going to use a lot of packages which is something they haven't done so far. To reach 768GB on a 1024-bit interface they would need 64, 32 or 16 packages of 12GB, 24GB or 48GB capacity. This puts them in the crosshairs of enterprise LPDDR5X packaging. The only way they could double this up is by using the same packaging as is used by nvidia for SOCAMM.
    M3 Ultra had 512 GB on 1024-bit. They apparently used semi-custom LPDDR5 packages to get there: 8x 128-bit 64 GB.

    That is as niche as it gets and could help explain why they discontinued it.

    I don't know how they'll get to 768 GB or 1.5 TB, but it needs to remain compact to fit into that small chassis. Maybe they will use SOCAMM. I don't think the technology behind a SOCAMM module is inherently expensive, unlike HBM. And of course, even if shortages eased, top-spec'd M5/M7 Ultra would be sold for well over $10k.
    Reply
  • acadia11
    How? LPDDRx is already slower than HBM3/e, so how? Is this Blackwell level … also slower than gddr7 vram?

    So not understanding the claim?
    Reply
  • usertests
    acadia11 said:
    How? LPDDRx is already slower than HBM3/e, so how? Is this Blackwell level … also slower than gddr7 vram?

    So not understanding the claim?
    If they are using LPDDR6 variants for these, my guess is they would increase bus width by 50% since LPDDR6 uses 12-bit subchannels. So they could be working with 1536-bit memory.

    With increased memory speeds, I think they could match the RTX PRO 6000 Blackwell (which has 96 GB GDDR7 and 1.792 GB/s). Good enough?
    Reply
  • thestryker
    usertests said:
    They apparently used semi-custom LPDDR5 packages to get there: 8x 128-bit 64 GB.
    They've done this several times over the years with the packaging for more than just the Ultras. The only thing I can come up with is that they're doing something side by side for most of them. In the case of a 128-bit/64GB package that's using 16Gb packages so I don't even want to know what sort of a mess the packaging was.
    usertests said:
    I don't know how they'll get to 768 GB or 1.5 TB, but it needs to remain compact to fit into that small chassis. Maybe they will use SOCAMM. I don't think the technology behind a SOCAMM module is inherently expensive, unlike HBM. And of course, even if shortages eased, top-spec'd M5/M7 Ultra would be sold for well over $10k.
    Higher density packages are possible and SOCAMM uses up to 32-bit/64GB packages. As far as I've seen this is the only place such high density packaging is available though. The largest client packages have gotten up to 24GB, but I believe those have only appeared in phones and the like.
    Reply
  • ezst036
    Admin said:
    Apple's planned M7 Ultra chip is being designed to support up to 1.5 TB of unified memory and to push AI performance toward the class of Nvidia's Blackwell accelerators.

    Apple's rumored M7 Ultra targets 1.5TB of memory and Blackwell-class AI performance, report claims — monster 2028 offering would depend on memory s... : Read more
    🤔

    I'm imagining the retail or B2B prices Apple would be charging for 1.5 TB of memory.

    I see where I can put together kits and maybe make it $70-50,000. So this is Apple. Apple upcharges like crazy.

    1.5TB Ram = over $200,000 from Apple??
    Reply
  • usertests
    ezst036 said:
    🤔

    I'm imagining the retail or B2B prices Apple would be charging for 1.5 TB of memory.

    I see where I can put together kits and maybe make it $70-50,000. So this is Apple. Apple upcharges like crazy.

    1.5TB Ram = over $200,000 from Apple??
    Plug in a craaazy $50/GB for LPDDR6(X), and that would still only be around $75,000 of memory.

    I think the 512 GB model was going for $15k to $25k:
    The Apple Mac Studio with the M3 Ultra chip and 512GB of unified memory has a base manufacturer price of $14,099 when configured with the maximum 16TB of storage. This represents a $4,000 premium for the 512GB memory upgrade on the top-tier 32-core CPU/80-core GPU model.

    Secondary Market: Due to high demand for AI and local LLM workloads, resale prices for the 512GB model have surged, with listings on platforms like eBay and refurb markets reaching $14,609 to over $25,000 depending on storage size and warranty status.
    I see listings for exactly $25k on ebay, but others for under $10k (very suspicious).

    So nah, not $200k unless economic and geopolitical pains rise to unprecedented heights.
    Reply
  • vossile
    The line of thinking here is all based on what is available to consumers now.

    Apple has access to other "channels" and we must not forget that once the AI crunch is over (because there are many things taking place simultaneously to overcome it because there are loads of cash to be made) memory prices are going to collapse dramatically. So Apple is counting on things to change by then or close to when the M7 Ultra becomes available. That is likely based on insider information that does not leak to you or me.

    And, of course, Gurman could be entirely wrong.
    Reply
  • ezst036
    usertests said:
    Plug in a craaazy $50/GB for LPDDR6(X), and that would still only be around $75,000 of memory.
    Do you mean just regular "this is what I can go buy today at any storefront" prices or do you specifically mean Appleworld markup pricing?
    Reply
  • thestryker
    ezst036 said:
    Do you mean just regular "this is what I can go buy today at any storefront" prices or do you specifically mean Appleworld markup pricing?
    $50/GB would be an absolutely insane price which is why they chose that as an example. The most expensive memory currently available that anyone would conceivably buy right now is ~$35/GB and that's for a ridiculously high cost EXPO ULL kit which is using brand new memory. This is of course a retail kit of desktop memory which is a significantly higher price than Apple themselves would be paying. Based on what I've seen premium LPDDR5X going for I'd be surprised if Apple would ever be in a position to pay more than $20/GB for any version of LPDDR.
    Reply