Nvidia just made a move that almost nobody predicted: it's actively pitching its brand-new Vera CPU to Chinese cloud providers, with first deliveries expected as early as August 2026. The price tag is $20,000 per chip, or roughly $10 million for a full 256-chip rack. The strategic logic — selling the CPU while the GPU remains blocked by US export controls — is either brilliant or a regulatory tightrope walk that ends badly. Here's the full picture.
Why Nvidia Is Selling CPUs to China When GPUs Are Banned
US export controls imposed since late 2022 have progressively tightened the GPU chips Nvidia can sell to Chinese buyers. The H100, H200, and other flagship AI accelerators are effectively off-limits. This has cost Nvidia an estimated $15–20 billion in potential revenue from the world's second-largest economy, per industry analyst estimates cited by TrendForce in their June 12, 2026 report.
Enter the Vera CPU. Unveiled by Nvidia in March 2026, Vera is an 88-core processor designed for agentic AI workloads — the inference and orchestration tasks that run AI agents in production. Critically, it's a CPU, not a GPU, and currently falls outside the most restrictive export control categories. Nvidia is explicitly pitching Vera as a standalone product to China — while the Vera Rubin GPU, which pairs with it in Nvidia's full platform, would trigger immediate regulatory scrutiny.
Who's Buying — and How Much
One major Chinese cloud provider has reportedly placed a preliminary order for more than 300 servers, each equipped with two Vera processors. At $40,000 per server in CPU cost alone, that's $12 million in chips for a single customer before factoring in server hardware, cooling, and networking.
When Nvidia unveiled Vera in March, the company noted that Alibaba and ByteDance were already working with the firm to deploy the chip. Both companies run massive AI inference workloads serving hundreds of millions of users — Vera CPU is purpose-built for exactly this use case. The strategic fit is clear.
This compares sharply with the situation 18 months ago: as we reported in our analysis of how Nvidia navigated the first wave of AI chip export bans, the company initially tried to design downgraded chips (H800, A800) to comply with controls. Those were subsequently banned too. The Vera CPU play is a more sophisticated approach.
The Regulatory Risk Nobody Is Talking About Loudly Enough
The US Commerce Department's Bureau of Industry and Security (BIS) has shown a pattern of progressively expanding export control categories to close loopholes. When Nvidia sold the H800 (a downgraded H100) to China, BIS eventually banned that too. The question is whether CPUs purpose-built for AI inference eventually face the same fate.
Industry analysts at Igor's Lab noted in their June 2026 analysis that "the CPU plays a critical role in agentic AI infrastructure, and selling it to Chinese hyperscalers effectively provides meaningful AI compute capability, even without the GPU." This is exactly the kind of argument BIS uses to expand control lists. Nvidia is betting the window stays open at least through 2026–2027.
What This Means for the Global AI Chip Race
Nvidia's China pivot changes the competitive calculus in several ways. First, it signals that Nvidia is not giving up on Chinese revenue — it's finding creative compliant paths. Second, it puts pressure on AMD and Intel, both of which also sell CPUs to China, to develop their own AI-optimized CPU plays. Third, and most importantly, it demonstrates that the US-China AI chip war is not a clean bifurcation — it's a messy regulatory chess game where rules keep changing and technology evolves faster than policy can track.
What This Means for You
If you own Nvidia stock, this is a cautiously positive development — incremental China revenue at a time when the GPU business there is blocked. But the regulatory risk is real; any BIS expansion of controls to include AI CPUs could reverse this quickly. For enterprise buyers building agentic AI infrastructure, Nvidia's Vera CPU is now in full production and available. For the broader tech industry, this saga is a live case study in how geopolitics and technology strategy collide — and neither side wins cleanly.
Frequently Asked Questions (FAQs)
Q: What is the Nvidia Vera CPU and when is it available?
A: The Vera CPU is Nvidia's 88-core processor purpose-built for agentic AI workloads. Unveiled in March 2026 and now in full production, it's priced at $20,000 per chip, with a full 256-chip rack running approximately $10 million. Chinese clients can begin ordering with August 2026 delivery.
Q: Why is Nvidia selling CPUs to China when GPU exports are banned?
A: Current US export controls primarily target advanced AI GPUs like the H100 and H200. The Vera CPU, while designed for AI workloads, currently falls outside the most restrictive export categories, giving Nvidia a legal path to sell it to Chinese cloud providers like Alibaba and ByteDance.
Q: Could the US government ban Nvidia's Vera CPU sales to China?
A: Yes. The Bureau of Industry and Security has previously expanded control lists to close similar loopholes. Analysts consider this a significant regulatory risk, though the timing of any such action is uncertain. Nvidia is betting the window stays open through at least 2026–2027.
Q: How does the Vera CPU compare to AMD and Intel CPUs for AI?
A: Nvidia's Vera CPU is specifically optimized for agentic AI inference tasks and is designed to pair with Nvidia's Rubin GPU platform. AMD's EPYC and Intel's Xeon are general-purpose data center CPUs with AI acceleration capabilities but are not purpose-designed for the same agentic AI use cases Vera targets.
Nvidia's Vera CPU play is audacious, legally defensible for now, and potentially very lucrative. Whether it becomes a sustainable China strategy or a regulatory flashpoint depends on policy decisions being made in Washington right now. Watch the BIS rulemaking calendar in the second half of 2026 — that's where this story gets resolved.
