Nvidia Opens a New Door Into China — Through the CPU Side
With its flagship AI accelerators still locked out of the world's largest emerging market, Nvidia has found a different path back in. The company has begun telling Chinese cloud and data center customers that its new Vera server CPU can be ordered now and shipped as early as August — a move that sidesteps the export restrictions that have blocked its advanced GPU business in China since last year.
The development, first reported by Reuters on June 12 citing three sources familiar with the matter, represents a significant pivot in Nvidia's China strategy. While Washington and Beijing have effectively conspired — through different motivations — to freeze H200 GPU deliveries, the Vera CPU sits in a different regulatory category altogether, one that carries far lighter U.S. export controls.
What the Vera CPU Is — and Why It Matters
Vera is not a graphics chip. It is an Arm-based server CPU, originally unveiled as the processor half of Nvidia's Vera Rubin AI supercomputer platform, which the company announced at CES 2026 in January. At GTC San Jose this past March, Nvidia broke Vera out as a standalone product, pairing it with a rack design that fits 256 liquid-cooled Vera CPUs and can sustain more than 22,500 concurrent CPU environments. The chip delivers a claimed 1.8x improvement in task completion over x86 processors on agentic workloads, according to Nvidia, and its predecessor, the Grace CPU, has shipped nearly 2.5 million units to date.
The Vera Rubin platform as a whole — combining the Vera CPU with Rubin GPU accelerators — is positioned by Nvidia as the foundation for next-generation AI infrastructure, with particular emphasis on inference and agentic AI, which require more general-purpose processing than GPU-heavy model training.
CEO Jensen Huang has been bullish about the CPU business. At Nvidia's May earnings call, the company disclosed visibility into nearly $20 billion in CPU revenue by the end of its fiscal year in late January 2027 — a figure that would position Nvidia as the world's largest CPU supplier by revenue, ahead of Intel and AMD. Analysts have estimated that Nvidia is already on track to deliver 4 million Vera CPUs in fiscal 2027.
"Conceding an entire market the size of China probably does not make a lot of strategic sense," Huang said in an April appearance on the Special Competitive Studies Project's podcast, referring to the broader impact of export restrictions. "I think that has already largely backfired."
The China Equation: Two Governments, One Blocked Market
The situation Nvidia faces in China has been shaped by obstacles on both sides of the Pacific. After the Biden administration expanded chip export controls in April 2025, even the China-designed H20 GPU was blocked. The Trump administration subsequently licensed approximately ten Chinese firms — including Alibaba, JD.com, and ByteDance — to purchase H200 GPUs. But as of this writing, not a single unit has been delivered: Chinese authorities, mindful of nurturing domestic semiconductor champions like Huawei and Cambricon, have declined to approve the imports at their end.
That double-block has left Nvidia with effectively zero AI GPU market share in China, down from a dominant 95% share before restrictions tightened, according to Huang's own public statements. The company took a $4.5 billion charge in its first fiscal quarter tied to excess H20 inventory and purchase obligations it could no longer fulfill.
The Vera CPU maneuver is notable precisely because it navigates around this impasse. Server CPUs are not classified under the same advanced-chip export restrictions that govern AI accelerators like the H200 or H20. Nvidia can legally sell the chip to Chinese buyers without a special license — a meaningful distinction at a time when GPU allocations require government sign-off on both sides of the ocean.
Orders, Deployments, and the Overseas Caveat
According to Reuters, Chinese cloud companies are already testing more than 300 Vera servers, and at least one major provider is prepared to place a commercial order. When Nvidia unveiled Vera in March, it noted that Alibaba and ByteDance were already working with the company to deploy the chip. One source told Reuters that initial deployments would be restricted to those companies' overseas data centers — a practical workaround that keeps U.S. silicon out of Chinese domestic infrastructure, limiting the political exposure for both sides.
Tom's Hardware noted that the August timeline for Chinese customers aligns with what Nvidia said at GTC Taipei during Computex, where the company indicated that Vera systems would reach customers through system builders and cloud partners starting in the fall. Offering Chinese buyers August availability during a global server CPU shortage, the outlet observed, suggests they are sitting near the front of the allocation queue.
The supply environment is working in Nvidia's favor. Intel has warned its Chinese customers of lead times of up to six months for server CPUs. AMD has said the global CPU market remains tight, with demand continuing to exceed its forecasts. TSMC manufactures the Vera CPU on its 3nm node, and SK Hynix supplies the memory — giving Nvidia a high-end process advantage over incumbents.
"Telling Chinese buyers they can have silicon in August, during a global server CPU shortage, suggests they're sitting near the front of the allocation queue for a product line Nvidia expects to generate $20 billion in revenue," Tom's Hardware reported.
Analysis: A Tactical Wedge, Not a Full Comeback
The Vera play is strategically elegant but limited. A CPU, however capable, is not a substitute for the H100, H200, or B30-class accelerators that Chinese AI labs actually need for large-scale model training and inference. Nvidia's China revenue story remains on hold as long as high-bandwidth GPU accelerators stay blocked.
What Vera does accomplish is threefold: it keeps Nvidia's brand and engineering relationships alive with major Chinese cloud customers; it captures real revenue in a segment (server CPUs) where China's domestic alternatives are far weaker than in accelerators; and it builds infrastructure-level stickiness at a time when Huawei's Ascend chips are actively competing for loyalty.
Whether Beijing permits or tolerates Vera deployments in Chinese domestic data centers — beyond the overseas workaround — will be the next test. The restriction to overseas infrastructure suggests both sides are feeling their way carefully. An unrestricted domestic rollout would carry political risk for Chinese officials trying to signal support for homegrown silicon.
What to Watch Next
Three threads will determine how this develops. First, whether Beijing formally signals approval or objection to Vera deployments in Chinese facilities — the overseas-only restriction looks like a provisional arrangement, not a settled policy. Second, whether the H200 GPU license logjam breaks; if Chinese approvals come through, the GPU revenue story changes entirely and Vera becomes a complement rather than a substitute. Third, whether Nvidia's claimed $20 billion CPU revenue forecast holds up as AMD and Intel compete for the same agentic AI infrastructure buildout — AMD's upcoming Zen 6 "Venice" processor has already drawn attention for claims of superior rack-level performance. Nvidia's re-entry into China, even through the CPU door, will test whether brand loyalty and supply priority can outlast regulatory uncertainty.
"Conceding an entire market the size of China probably does not make a lot of strategic sense. I think that has already largely backfired."— Jensen Huang, CEO, Nvidia