Nvidia faces AMD, cloud chips and AI code porting as Huang exits

Nvidia faces AMD, cloud chips and AI code porting as Huang exits

Wired argues that Nvidia's timing is poor. Its recently ousted CEO Jensen Huang, who foresaw the AI chip boom earlier than most, is gone, and CFO Colette Kress has taken the reins, while competitors with new technology are closing in. An anonymous CEO of a large AI company told Wired that "everything sold today was designed two to three years ago", so Kress, or whoever replaces her permanently, must chart where the chip industry will be two years from now.

Nvidia still dominates. It captures 80 percent of AI accelerator sales, according to Silicon Analysis, and its revenue went from $26.91 billion in January 2022 to $215.94 billion in January 2026. Huang cofounded the company in 1993 to design graphics chips for PCs, saw the promise of AI in the mid-2010s, funded academic research, built software for AI engineering and gave hardware to labs and startups. When ChatGPT took AI mainstream in 2022, chip sales skyrocketed. Wired says the challenges Nvidia faces may not be existential, but they add up.

The competition comes from several sides. AMD's share of the AI accelerator market has crept up from two percent in 2022 to between 7 and 10 percent in 2026. Google, Amazon and Microsoft are designing their own silicon to cut costs and gain some independence from Nvidia, and each has seen similar gains. Startups offer specialised chips: Cerebras sells giant ones for giant models, and AMD recently bought Taalas, which burns models' weights into silicon, though using those chips requires rewriting a lot of code. Patrick Moorhead, chief analyst at MoorInStrat, says AI agents make inference (running already trained models) more important, an area where Nvidia has less of an edge against hyperscalers and startups. He adds that demand for general purpose CPUs is growing, and Nvidia barely competes there. Nvidia did acquire the specialised-chip startup Groq in December 2025.

China is the second front. Nvidia may not sell its most powerful chips there but can sell previous generations. Sales of those chips and of non-AI chips doubled over the past year, yet China now accounts for only eight percent of Nvidia's revenue, against 20 percent in 2024, according to its filings. Huawei has built supercomputers that harness many more chips than usual, and some Chinese AI companies, such as Z.ai, say they trained their models entirely on Huawei chips. If Chinese open-weight models gain traction abroad, that could create a market for Chinese chips elsewhere. It remains unclear how many chips exported to other countries end up in China or are used remotely by Chinese firms.

Wired says Huang's dealings with China played a major role in his downfall. He had a romantic relationship with Huang Le, also known as Alice Huang, a Chinese executive at the Singaporean company Megaspeed, whom the US government suspects of being an asset of China's Ministry of State Security. Megaspeed and its subsidiaries reportedly imported between $4.6 and $7.5 billion of Nvidia chips into southeast Asia, and reports suggest the company may have given Chinese AI firms access to them. Babak Minovi, a senior analyst at the Eurasia Group, says Huang was right to argue that cutting off China would empower domestic rivals: "Huang was taking a pragmatic line, and he knows competition."

The article's last section is about trends. Models built on Nvidia chips are automating coding, data analysis and research that help design chips, and OpenAI and Anthropic have recently touted their newest models' ability to design electronics from scratch. Moorhead says this may not hurt Nvidia much because "they have access to AI too". The bigger worry, he says, is that AI can now rewrite code to run on other hardware, which may prove a problem for CUDA, the software long seen as a moat keeping AI companies on Nvidia chips. Wafer, a startup that uses AI to port models to AMD and serves them more cheaply, says its revenue has grown spectacularly in recent months. CEO Emilio Andere says that whenever a major new open-weight model appears, Wafer can have it running on AMD hardware at least 80 percent as efficiently within hours, and that his engineers barely have to edit or review the code.

Veteran chip analyst Dan Hutchinson says Nvidia may be entering its "Intel swan moment", referring to Intel's 2019 replacement of engineering-focused CEO Brian Krzanich with CFO Bob Swan. Wired notes that Krzanich also resigned after a relationship with an employee, while Huang is alleged to have had multiple. After that change, it writes, Intel's focus shifted to sales over technical progress, bureaucracy ballooned and revenue tanked. Hutchinson's verdict: "Nvidia is now an Intel waiting to happen."

Key facts

  • Nvidia captures 80 percent of AI accelerator sales per Silicon Analysis, and its revenue rose from $26.91 billion in January 2022 to $215.94 billion in January 2026.
  • AMD's AI accelerator share grew from two percent in 2022 to between 7 and 10 percent in 2026; Google, Amazon and Microsoft are building their own chips and, per Wired, have seen similar gains.
  • China is now eight percent of Nvidia's revenue, down from 20 percent in 2024; Wired says Huang's China dealings, including a relationship with a Megaspeed executive, played a major role in his downfall.
  • AI-driven code porting threatens the CUDA moat: Wafer's CEO says new open-weight models can run on AMD at least 80 percent as efficiently within hours.
  • Analyst Dan Hutchinson warns of an Intel-style 'swan moment' after a CFO takes over from an ousted CEO, as happened at Intel in 2019.

Why it matters

Nvidia is the main supplier of AI accelerators, with 80 percent of sales per Silicon Analysis, and its revenue grew from $26.91 billion to $215.94 billion in four years. Wired's point is that leadership is changing at the moment pressure is building on several fronts at once: AMD, in-house chips from Google, Amazon and Microsoft, specialised startups, Chinese rivals, and AI tools that make switching away from CUDA easier. The Intel comparison, a CFO replacing an engineering-minded CEO in 2019, frames the worry that focus could drift from technical progress.

Who it affects

Nvidia and its leadership, starting with CFO Colette Kress, who has taken the reins. AMD, Cerebras, Wafer and the hyperscalers (Google, Amazon, Microsoft) stand to gain from any loss of Nvidia's lead. AI companies that buy chips are affected through supply and cost, and Chinese firms such as Z.ai and Huawei are part of the picture through domestic chips and remote access to Nvidia hardware.

How to use it

Read it as a map of the pressure points on Nvidia rather than a forecast. Things to watch, per the article: inference demand as AI agents spread, demand for general purpose CPUs, how fast AI-assisted porting moves workloads from CUDA to AMD, whether Chinese open-weight models create demand for Chinese chips abroad, and what the permanent CEO decides about chips that will ship in about two years.

How solid is it

This is a reported analysis, not a company announcement. The 80 percent share comes from Silicon Analysis, with no period stated, and the revenue and China-share figures are drawn from Nvidia's filings. The anonymous AI-company CEO, and the claims about Wafer's growth and 80 percent efficiency, rest on single sources; the Wafer claims come from its own CEO, who also says revenue has grown spectacularly without giving figures. No Nvidia response or comment is quoted. The article does not state that Huang's ouster itself caused any loss of sales or market share, and it gives no date for it.

Risks and caveats

Wired itself says the challenges may not be existential. The Megaspeed material is suspicion and reporting: the US government suspects Huang Le of being an asset of China's Ministry of State Security, and reports say Megaspeed may have given Chinese AI firms access to chips. How many exported chips reach China directly or remotely remains unclear. The Intel comparison is an analyst's warning, and Moorhead notes AI chip design may not threaten Nvidia much since it can use AI too. Several rivals, such as Taalas-style chips, need a lot of code rewriting to work with cutting-edge models.

“Nvidia is now an Intel waiting to happen”

— Dan Hutchinson, veteran chip analyst, quoted by Wired