Nvidia's desktop PC chip holdup purportedly tied to Windows delays — ongoing chip revisions and weakening demand also blamed

News
By published

More reasons for the alleged delay have emerged

Nvidia
(Image credit: Nvidia)

MediaTek and Nvidia have reportedly pushed back the launch of their much-anticipated N1X AI PC platform to the first quarter of 2026, according to a new DigiTimes report citing supply chain sources. The latest reasons given contrast with an earlier report, which still informed us of the delay, but attributed it to critical hardware defects requiring a silicon respin. While chip revisions at Nvidia are still part of this latest story, the new information points toward a broader set of factors, including Microsoft’s slower-than-expected OS roadmap and weakening demand across the notebook market.

The N1X platform, originally believed to be scheduled for Q3 2025 with both consumer and commercial models, never materialized at Computex earlier this year, fueling speculation about its readiness. DigiTimes now reports that MediaTek and Nvidia are prioritizing enterprise-class systems for the initial rollout, banking on stronger commercial adoption before expanding into the volatile consumer segment.

Moreover, DigiTimes' report also offers a glimpse into Nvidia and MediaTek’s expanding collaboration. Beyond PCs, the two companies are said to be advancing joint efforts in automotive AI via MediaTek’s Dimensity Auto platform, as well as edge AI development with Nvidia’s TAO toolkit and MediaTek’s NeuroPilot SDK. Digitimes says the companies have also co-developed the DGX Spark personal AI supercomputer and are key partners in Nvidia’s NVLink Fusion ecosystem for custom AI silicon.

Anyhow, the new reported Q1 2026 timeline for N1X could suggest that Nvidia and MediaTek are refining both hardware and strategy. While SemiAccurate’s earlier claims of hardware defects may still hold some truth, the new report paints a picture of a more calculated delay, aligning with Microsoft’s OS updates, fixing remaining chip-level issues, and waiting for commercial demand to stabilize. In the meantime, Nvidia’s GB10-based AI workstations, which are still on track for release, may serve as the company’s first real-world testbed for consumer-facing AI PC hardware before N1X arrives.

Hassam Nasir
Hassam Nasir
Contributing Writer

Hassam Nasir is a die-hard hardware enthusiast with years of experience as a tech editor and writer, focusing on detailed CPU comparisons and general hardware news. When he’s not working, you’ll find him bending tubes for his ever-evolving custom water-loop gaming rig or benchmarking the latest CPUs and GPUs just for fun.

20 Comments Comment from the forums
  • abufrejoval
    Well, who would have thought that this would be the reason to push this back: it only runs well with Linux/SteamOS!

    Possibly even with MacOS, if anyone wanted that.

    It's also the only reason I'd ever want to buy one of those, if the price was right. But looking at Strix Halo, everybody is just expecting to get rich from doing Fruity Cult things.

    And you shouldn't let an LLM proof-read itself, there is so much cut & paste garbage in there, it becomes hard to read.
    Reply
  • SomeoneElse23
    Windows is delayed because MS is busy spending money buying crap.
    Reply
  • bit_user
    The current version of this article has almost a second copy concatenated onto the end. Someone is either sloppy with cut-and-paste or trying to pad out the word count. Either way, it's clear the final version wasn't proofread.
    Reply
  • EzzyB
    abufrejoval said:
    Well, who would have thought that this would be the reason to push this back: it only runs well with Linux/SteamOS!

    Possibly even with MacOS, if anyone wanted that.

    It's also the only reason I'd ever want to buy one of those, if the price was right. But looking at Strix Halo, everybody is just expecting to get rich from doing Fruity Cult things.

    And you shouldn't let an LLM proof-read itself, there is so much cut & paste garbage in there, it becomes hard to read.
    This seems a bizarre thing to say. Given it's target market it pretty much has to run on Windows. "(MediaTek and Nvidia are prioritizing enterprise-class systems for the initial rollout)" Why would Linux or MacOS be modified for it? Nvidia could be using a custom Linux to test it I guess, but more likely a pre-release version of Windows for ARM. Apple would not seem to be involved at all. Best guess is this works better on whatever Windows build MS has provided and not at all on Linux or MacOS.

    It's another attempt to crack the enterprise laptop market which is Intel's last bastion and hasn't, so far, worked too well for for Snapdragon X (for the same reasons I state below.)

    Nvidia really seems to be doing this backwards. THE toughest nut to crack is getting a new product directly into enterprise. Those are the absolute most conservative types because literally everything piece of software in the enterprise has to be tested against that hardware/software before they'll buy it. You are even less likely find AMD processors there because they are THAT conservative. If your job is to sign off on a buy of say, 5000 laptops, you have to be.
    Reply
  • abufrejoval
    EzzyB said:
    This seems a bizarre thing to say. Given it's target market it pretty much has to run on Windows. "(MediaTek and Nvidia are prioritizing enterprise-class systems for the initial rollout)" Why would Linux or MacOS be modified for it?
    Well Linux already runs as well with Nvidia GPUs on ARM as it does on x86: you might say it's pretty near done, but ARM on Windows is still a nightmare, I keep hearing. From my experience on RP5 and Orange Pi 5+ the Windows OS core is fine, but anything device related is missing: much easier to run as a VM on ARM-Proxmox with RDP.
    EzzyB said:
    Nvidia could be using a custom Linux to test it I guess, but more likely a pre-release version of Windows for ARM. Apple would not seem to be involved at all. Best guess is this works better on whatever Windows build MS has provided and not at all on Linux or MacOS.
    I have no insight into the portability of today's MacOS: I'm pretty sure the Darwin base and BSD userland is ARM-proof, while the GPU side may be trouble to port from Apple's GPU to Nvidia. no matter the CPU ISA.
    EzzyB said:
    It's another attempt to crack the enterprise laptop market which is Intel's last bastion and hasn't, so far, worked too well for for Snapdragon X (for the same reasons I state below.)
    On Snapdragon's Windows troubles, I only have second-hand information. My impression is that every laptop maker tends to do their own thing for everything outside the CPU/APU/SoC and rely on UEFI abstractions to fill the gap to the OS.

    That layer just doesn't exist for Snapdragon and it's causing mortal overhead on Windows and Linux implementations, because it would require scale to make it worth development effort.

    Nvidia may be in a better position to pressure Microsoft than Qualcomm, but with M$ shedding staff, it may just not scale that well and I really don't want to look into Qualcomms balance sheets: they got really lucky they didn't loose the court case with ARM.
    EzzyB said:
    Nvidia really seems to be doing this backwards. THE toughest nut to crack is getting a new product directly into enterprise. Those are the absolute most conservative types because literally everything piece of software in the enterprise has to be tested against that hardware/software before they'll buy it. You are even less likely find AMD processors there because they are THAT conservative. If your job is to sign off on a buy of say, 5000 laptops, you have to be.
    They are also after Fruity Cult profit margins. And that's a bit of a sonic wall or chicken and egg issue.

    I have no idea if Apple sells in some corporate environments or layers, but they didn't get rich only on private consumers. In any case it's a market that doesn't support commodity or competition, as current court cases well attest.

    And no, the enterprise market value is actually dying from both all directions, compute saturation, AI layoffs etc., which is why Microsoft invented nonsensical hardware criteria to revitalize it via Windows 11.

    Even the lowliest Atoms outrun most entreprise software needs by far and whoever needs leading edge does it in the cloud or on low volume workstations. Enterprise purchasing departments are by nature penny pinchers not innovators, I agree there, but not on Nvidia making that a priority.

    Perhaps they can't really help themselves from acting like Intel during their glory days: delivering the best user product isn't nearly as important when you're #1 as depriving the competition of opportunities to grow and fester and they can afford to invest a lot of money into that.

    But mostly I believe it's Mediatek and the Asian market pushing here and those guys are far more ready to ditch not just x86 but also Microsoft. Or US domination in general. Nvidia is smart enough not to say no when they can just sell more. They have the Switch, but would probably like the Steam consoles and then everything upwards.

    Can't fault them both for that as a European, nor do I think they have anything backwards.
    Reply
  • thestryker
    Delaying a hardware launch (if its actually ready to go) due to nebulous "OS" issues seems unlikely unless it's a problem with their driver stack. A chunk of the issues Qualcomm had with the Snapdragon X Elite launch were certainly on them not Microsoft. Windows on Arm has been in wide release for years and even if you point at the the X Elite as the true launch still over a year.

    I still think this is a product almost nobody wants at a consumer level. That's likely to be their biggest issue and the performance level isn't going to be high enough to get gamers onboard as these are unlikely to be appropriately priced.
    Reply
  • Notton
    DGX Spark is $4000, which is already enterprise pricing.
    Are they going to price it like Apple?
    N1X Ultra $4000-$14100
    N1X Max $2000-$4000

    I guess it'll depend on how well the GPU can game, but if their top bragging point is TOPS, then it doesn't look promising.
    Reply
  • abufrejoval
    Notton said:
    DGX Spark is $4000, which is already enterprise pricing.
    Are they going to price it like Apple?
    N1X Ultra $4000-$14100
    N1X Max $2000-$4000

    I guess it'll depend on how well the GPU can game, but if their top bragging point is TOPS, then it doesn't look promising.
    If I remember correctly, we'll know more any day now, as supposedly ASUS is presenting a variant of the Spark this month.

    And the DIGITS is likely to have a premium pricing, because one USP is an integrated dual port ConnectX-7 adapter that supports both Infiniband and Ethernet personalities. It allows a certain degree of scale-out with whatever software transparency Nvidia is able to provide via CUDA or else.

    With a theoretical peak RAM bandwidth of 273GB/s as quoted by Nvidia, that falls below the 320GB/s of an RTX 5050, but better latency and 128 instead of 8GB of capacity.

    CPU is 10 ARM 925 and 10 ARM 725 which is somewhere along the Snapdragon X Elite in terms of compute power and the GPU below an RTX 5050.

    To me that's essentially the same horse power as my LOQ notebook with a Zen 3 8-core APU and an RTX 4060 I got for around €750 a year ago, except that that's only 64GB of 128 bit of 50GB/s DDR5-5200 DRAM and 8GB of 320GB/s VRAM.

    Throw in €100 for the other 64GB we are talking about charging €3000 for the ConnectX-7 and the exclusive design in terms of hardware cost, which is certainly not reflected in gaming performance, no matter what they manage in terms of software support.

    It's not at all designed for gaming, even if you could get game vendors to port to ARM or Valve to reuse part of their current Apple efforts, which I guess include an x86 emulator for games that aren't native ARM.

    For me Strix Halo and the DGX Spark offer roughly similar raw capabilities, but the Strix Halo could just about manage one 400Gbit/s Mellanox port with its 16 lanes of PCIe v4. The internal connectivity on the DGX Spark is not disclosed today, but I'd personally doubt it will reach the 32 lanes of PCIe v5 the biggest ConnectX-7 chips support. Halo doesn't run CUDA, but with IB and software abstractions it might actually matter little.

    Personally I believe the overhead of this software scale-out is unattractive for end-user or production use at this 'dwarf size' hardware level. However, it would allow widening the software engineering base working on scale out techniques and abstractions significantly, including universities or research labs: those guys don't need to run 100% on production hardware (which needs to run near 100% sellable utilization to pay for its crazy cost), if they have a local dwarf variant that functionally behaves the same, that's a productivity boost for them.

    But IMHO all that engineering effort is around designing scale-out unicorns for both training and inference, not for end-users expecting 'commodity shrink wrapped AI software', if there ever was such a thing.

    Ultimately a niche within a niche and asking a 200-300% surcharge for that means it's not for the masses, but Nvidia gets to skim an attractive amount of cream from the crème de la crème in AI software engineering.
    Reply
  • bit_user
    abufrejoval said:
    I have no idea if Apple sells in some corporate environments or layers,
    Yes. At my job, you can get an Apple laptop at your manager's discretion. I think they found their way into corporations because some execs simply demanded it. Another reason might be that any iPhone apps development must be on Macs. So, if the company maintains any apps, there are going to be at least some Macs on their network.

    abufrejoval said:
    Even the lowliest Atoms outrun most entreprise software needs by far
    You have no idea how much bloat-ware they load onto our machines. The lowest spec laptop we can get is an i5.

    abufrejoval said:
    Enterprise purchasing departments are by nature penny pinchers not innovators, I agree there, but not on Nvidia making that a priority.
    We get them on lease, which used to be 3 years and recently got stretched to 4. They negotiate a deal with a big OEM (e.g. Dell) and spec out several different machines.

    The engineering laptop I have includes a Nvidia dGPU, which I really don't need, but we're unable to customize the spec and I'd lose too much on the CPU front if I dropped to the fastest model with an iGPU. It's ultimately not the IT department's problem how much a machine costs, because that comes out of your manager/department's budget. What the IT department optimizes for is to have a relatively small number of different configurations to support.

    abufrejoval said:
    I believe it's Mediatek and the Asian market pushing here and those guys are far more ready to ditch not just x86 but also Microsoft. Or US domination in general. Nvidia is smart enough not to say no when they can just sell more. They have the Switch, but would probably like the Steam consoles and then everything upwards.
    This is actually the best argument I've heard for why Nvidia partnered with Mediatek on this. Nvidia might not want their client-side fortunes too closely tied to the x86 platform. While Nvidia is perfectly capable of making whole SoCs, Mediatek probably has way more business relationships in the Asian client market and can potentially drive a lot more design wins in Asian OEMs than if Nvidia tried to shop its own SoCs there.
    Reply
  • abufrejoval
    bit_user said:
    This is actually the best argument I've heard for why Nvidia partnered with Mediatek on this. Nvidia might not want their client-side fortunes too closely tied to the x86 platform. While Nvidia is perfectly capable of making whole SoCs, Mediatek probably has way more business relationships in the Asian client market and can potentially drive a lot more design wins in Asian OEMs than if Nvidia tried to shop its own SoCs there.
    In my opinion Mediatek is hugely underestimated in the West in terms of what they are capable of.

    We tend to think of them as a 2nd line contender for cheaper phones, but ignore just how big they are in global mobile radio coverage: AFAIK they are toe-to-toe with Qualcomm in that respect and that's the only two in that league.

    To my understanding the Fruity Cult still has to rely on Qualcom, who it mostly considers an arch-enemy, for its mobile modems, because they didn't manage to do their own. And that's years after they had bought Initel's mobile radio division for the purpose of gaining autonomy in that rather important part of the phone business.

    And Samsung, who used to be a primary contender, isn't doing much better: they never managed to include US and Chinese mobile radio support in their Exynos chips.

    Mediatek has always gone for true ARM designs in CPUs and GPUs, while Qualcomm preferred spinning their own for both, which at times almost killed the company, e.g. as they fumbled the 64-bit transition.

    But those ARM, Inc. CPUs never were all that bad, and even the GPUs, with technology originally purchased from AMD, seem to have become rather good, so Nvidia might be motivated to ensure that those don't go much further, unless they can finally buy ARM as they had planned.

    The PRC might want their fingers on Mediatek just as much as they want them on TSMC, because without mobile radio IP their Kirins can't talk.
    Reply