Nvidia, Qualcomm and MediaTek Move In on Intel-AMD CPU Turf

The CPU used to be the settled part of the AI stack. It ran the OS, marshalled data, and mostly stayed out of the AI conversation while GPUs ate all the attention. That framing is now breaking, according to a piece from Nikkei Asia arguing that Nvidia, Qualcomm and MediaTek are all squeezing into a data center CPU market Intel and AMD have owned for years.

The specific moves are worth pinning down. Nvidia CEO Jensen Huang has said his company is entering the market for PC and server CPUs with the intention of "reinventing" the industry, and has framed the overall CPU market as heading into what he calls a $200 billion industry. On the data center side, Nvidia's new Vera CPU carries 88 high-performance cores. On the PC side, Nvidia is launching a chip called RTX Spark, a joint effort with Taiwan's MediaTek, which debuts on a fresh line of Windows PCs from Microsoft, Dell, HP, ASUS, Lenovo and MSI. Qualcomm, separately, has unveiled its own data center chip lineup to challenge Nvidia in the AI processor race.

The pressure on the incumbents is real for a boring, physical reason: they can't make enough of what customers already want. AI-driven server demand has forced Intel and AMD to prioritize capacity for general and storage servers, squeezing PC supplies and pushing quoted CPU prices up an average of 10 to 15% so far this year, with wait times in some cases stretched from one or two weeks to eight to twelve. Every quarter Intel and AMD spend rationing existing product is a quarter Nvidia and Qualcomm get to pitch OEMs on a full-stack alternative.

The China half of this is a different sort of squeeze. Beijing is reportedly drafting a plan on the order of $295 billion to build a nationwide AI computing grid running mostly on domestic chips, and nine categories of home-grown AI chips from vendors including Huawei, Alibaba, Shanghai Biren and Moore Threads have already cleared a Chinese government security review that clears them for sensitive deployments. That is a wall being built, not a market being opened.

The honest caveat is that a chip announcement is not a design win. The reporting doesn't tell you how Vera or RTX Spark benchmark against Intel's and AMD's shipping parts on real workloads, what pricing looks like at volume, or how much of the $200 billion figure Huang cites is Nvidia's own framing rather than an independent number. What is worth watching is the OEM list. When Dell, HP and Lenovo are shipping Nvidia CPUs at the same time hyperscalers are qualifying Vera racks, the part of the incumbents' position that erodes first is pricing power.