KEY POINTS
- Nvidia holds about 80 percent of AI accelerator sales, while AMD has risen from 2 percent in 2022 to 7-10 percent in 2026.
- China now makes up 8 percent of Nvidia's revenue, down from 20 percent in 2024.
- AI tools that can port code to rival hardware may weaken the CUDA software advantage, according to analysts cited by WIRED.
Nvidia is facing a growing field of competitors at a delicate moment, as Chief Financial Officer Colette Kress has taken over leadership following the removal of CEO Jensen Huang, WIRED reported. The outlet describes a company that still leads the AI hardware market but faces pressure from several directions.
According to figures from Silicon Analysis cited by WIRED, Nvidia captures roughly 80 percent of AI accelerator sales. AMD has steadily gained ground, moving from about 2 percent of the market in 2022 to between 7 and 10 percent in 2026. Google, Amazon and Microsoft are also building their own specialized chips, partly to cut costs and reduce their reliance on Nvidia, and WIRED says each has seen similar gains.
Startups add to the pressure. Cerebras sells very large chips designed for very large models, and AMD recently acquired Taalas, which burns a model's weights directly into silicon for greater efficiency. Such chips usually require substantial code rewrites to run advanced models. An anonymous AI company chief executive told WIRED that products sold today were designed two to three years ago, so Nvidia's next leader must anticipate where the industry will be in two years.
Analyst Patrick Moorhead of MoorInStrat told WIRED that AI agents are making inference, the running of already trained models, more important. Nvidia has less of an edge there against cloud providers and startups, though it bought a specialized chip startup in December 2025. Demand for general-purpose CPUs is also rising, a segment where Nvidia barely competes.
China is another weak spot. Nvidia cannot sell its most powerful chips there, only earlier generations, and China now accounts for 8 percent of revenue versus 20 percent in 2024, per the company's filings. WIRED notes that Huawei has built supercomputers linking many chips, and that some Chinese AI firms, including Z.ai, say they trained models entirely on Huawei hardware. It also reports that Singapore-based Megaspeed reportedly imported between $4.6 billion and $7.5 billion of Nvidia chips into southeast Asia, with reports suggesting Chinese AI firms may have obtained access.
The article also points to a risk to Nvidia's software lead. CUDA has long kept developers tied to its chips, but AI models can now rewrite code for other hardware quickly. Wafer, a startup that ports models to AMD chips, says it can run a new open-weight model at least 80 percent as efficiently within hours, according to its CEO Emilio Andere, who said his engineers barely need to review the code.
Despite these threats, WIRED says Nvidia retains a large lead built on Huang's early bet on AI. Revenue rose from $26.91 billion in January 2022 to $215.94 billion in January 2026. Chip analyst Dan Hutchinson suggested Nvidia could be approaching an experience like Intel's after its 2019 move to a finance-focused chief executive, when attention shifted from engineering to sales.