AI skews silicon supply

Nvidia is reportedly preparing to skip a new GeForce generation in 2026 as AI chips and memory supply take priority over gaming cards. (digitaltoday.co.kr) Digital Today, citing foreign media reports published April 19 and April 20, said 2026 would be the first break in Nvidia’s annual GeForce launch rhythm since the 1990s. Nvidia told CNBC that gamers remain “hugely important” and said it is “always innovating, testing and releasing” gaming technologies, but CNBC reported no new GeForce generation has been announced after the RTX 50 series debuted at CES in January 2025. (digitaltoday.co.kr) (cnbc.com) At the same time, Microsoft said its Fairwater facility in Mount Pleasant, Wisconsin, went live ahead of schedule. Satya Nadella said on April 16 that the site will link “hundreds of thousands” of Nvidia GB200 systems into one cluster, and Microsoft had previously said the first phase was due online in early 2026. (wccftech.com) (blogs.microsoft.com) A graphics processing unit is the chip that renders game images on a PC and also trains large AI models in data centers. When the same supplier is serving both markets, production choices about chips, packaging and memory can shift capacity away from consumer hardware and toward server racks. (nvidia.com) (cnbc.com) The bottleneck is not only the processor itself. Digital Today said high-bandwidth memory, the stacked memory used in AI accelerators, consumes about four times as many silicon wafers as conventional memory, tightening supply across the broader market. (digitaltoday.co.kr) The money now sits overwhelmingly on the AI side. Nvidia reported fourth-quarter fiscal 2026 revenue of $68.1 billion, including $62.3 billion from data center products, while CNBC reported the data center segment accounted for 91.5% of revenue and Digital Today said compute and networking margins averaged 69% over three years versus about 40% for gaming graphics. (investor.nvidia.com) (cnbc.com) (digitaltoday.co.kr) That squeeze is not limited to Nvidia. Digital Today reported AMD raised Radeon RX 9000 prices by 10% to 17%, and said Intel shelved a planned Arc B770 gaming card in favor of the workstation-focused Arc Pro B70 with 32 gigabytes of memory. (digitaltoday.co.kr) For corporate buyers outside the AI arms race, the effect is slower and less visible than a product launch delay. More memory and advanced packaging are being reserved for AI clusters like Fairwater, leaving finance, design and workstation buyers to compete for the remaining supply, with higher prices and longer procurement cycles. (blogs.microsoft.com) (digitaltoday.co.kr) Nvidia was founded in 1993 to sell 3D graphics into gaming and multimedia, and it introduced the GeForce 256 in 1999. In 2026, the company’s biggest public builds are AI systems such as Blackwell and Rubin, and the supply chain is following that demand. (nvidia.com) (nvidianews.nvidia.com)