Nvidia Returns to China With the Vera CPU — Can It Succeed Where the H200 Failed?

Nvidia has opened a new front in its effort to revive a Chinese business that CEO Jensen Huang once described as having "effectively fallen to zero." According to a Reuters exclusive, the company has informed major Chinese cloud providers—including Alibaba, ByteDance, and Tencent—that its new Vera central processing unit (CPU) could be available as soon as August 2026 and that they can now place orders .

The sales push marks a strategic pivot for Nvidia, relying not on the powerful GPUs that have been blocked by U.S. export controls, but on a brand-new category of standalone server processors purpose-built for the age of AI agents.

A New Product for a New Regulatory Pathway

The Vera CPU is fundamentally different from the H200 GPU, which received U.S. government clearance for sale to vetted Chinese buyers in January 2026 but whose shipments have since stalled . Vera is Nvidia's first fully custom-designed, standalone data center CPU—an Arm-based processor with a different set of capabilities and, potentially, a different set of export rules.

This architectural distinction is central to Nvidia's re-entry strategy. While the H200 is a GPU that accelerates AI training and inference, Vera is a general-purpose CPU designed specifically for agentic AI workloads: the reinforcement learning and sandboxed task execution that happen when an AI agent plans, reasons, and uses tools. Because it is a CPU rather than a GPU, it may not trigger the same U.S. export restrictions that have crippled Nvidia's graphics processor business in China .

The Pitch to Chinese Hyperscalers

The outreach is targeted and concrete. Nvidia has specifically named Alibaba, ByteDance, and Tencent among the cloud giants being courted, and one major Chinese cloud company is already planning an order for more than 300 servers equipped with Vera CPUs .

Chinese customers are interested. The same sources who confirmed the August timeline told Reuters that clients have shown genuine enthusiasm for the chip, which Nvidia claims runs up to 1.8 times faster than competing x86 processors in agentic sandbox benchmarks . When Huang first disclosed the Vera CPU in March 2026, he said leading cloud companies including Alibaba and ByteDance were collaborating with Nvidia to deploy it .

At the same time, the memory of the stalled H200 rollout provides a cautionary tale. Despite receiving U.S. approval to sell H200 GPUs to Chinese customers—subject to a 25% tariff and mandatory third-party security audits on each unit—actual revenue from those sales remains absent . Clearance on one side of the Pacific does not guarantee purchases on the other.

What Makes the Vera CPU Different

Unveiled at GTC 2026 as Nvidia's first standalone server processor, Vera is built from the ground up for a new class of AI workloads . Key specs include:

• 88 custom Arm-based Olympus cores with Nvidia's Spatial Multithreading, enabling 176 threads per socket .
• Up to 1.2 TB/s of memory bandwidth from paired LPDDR5X memory, which is crucial for the high-throughput data movement that agentic AI demands .
• 1.8x performance advantage over the fastest x86 CPUs in agentic sandbox benchmarks, according to Nvidia's headline claims. The company's own developer documentation is more conservative, citing a 1.5x improvement under full-socket load, reflecting different measurement scenarios .
• Rack-scale density: A single liquid-cooled rack can house 256 Vera CPUs, sustaining over 22,500 concurrent execution environments for AI agents .

The economics are equally ambitious. A single Vera chip costs "well north" of $20,000 before bulk discounts, and a fully outfitted 256-chip rack runs approximately $10 million, according to SemiAnalysis .

The Financial and Strategic Stakes

CFO Colette Kress has anchored the company's CPU ambitions to eye-popping numbers. On the Q1 FY2027 earnings call, she disclosed that Nvidia has visibility into roughly $20 billion in standalone Vera and Grace CPU revenue for the fiscal year—a figure that would make Nvidia the world's largest CPU supplier by revenue, surpassing both Intel and AMD . For context, Intel reported $5 billion in data center and AI revenue last quarter, while AMD reported $5.8 billion .

Huang went further, describing Vera as a "major new growth driver" and the company's "primary growth catalyst." He pegged the total addressable market for CPUs at $200 billion—a market Nvidia has never addressed before—and when asked directly in Taipei whether that forecast includes China, he replied, "I would think so" .

The Two-Sided Regulatory Uncertainty

The largest question hanging over Nvidia's China CPU strategy is not technological but political—and it has two distinct fronts.

From the U.S. side, it remains unclear whether Vera will fall under current export controls on advanced semiconductors. While the chip is a CPU rather than a GPU, the Biden administration's technology restrictions are broad, and Nvidia has not publicly confirmed that it has obtained the necessary licenses for mass export .

From the Chinese side, Beijing's regulators must decide whether to permit large-scale Vera purchases. China is actively pushing domestic chip development, including Huawei's Ascend processors and homegrown CPU alternatives, and has urged local tech companies to reduce dependence on foreign silicon . Whether the government chooses to green-light a major new Nvidia product—even a CPU rather than a GPU—while simultaneously building domestic capacity is an open question.

The stalled H200 situation illustrates how even a product with U.S. clearance can stall. Chinese officials did issue in-principle approval for H200 orders from Alibaba, Tencent, and ByteDance in early 2026, but with restrictions barring the chips from military, sensitive government agencies, and critical infrastructure . That conditional approval did not translate into shipments, and the same pattern could replay with Vera if Chinese regulators impose their own constraints or simply delay.

A Pivot That May Redefine the Chip Market

Whether or not Vera achieves large-scale sales in China, the strategic implications extend far beyond one geography. By entering the data center CPU market with a chip purpose-built for agentic AI, Nvidia stands to reshape a sector historically dominated by Intel and AMD. If the company hits its $20 billion revenue target, it would represent one of the fastest market entries in semiconductor history—moving from a near-zero position in data center CPUs to industry leadership in a single fiscal year.

The China outreach is the first concrete test of whether that ambition can survive the geopolitics of advanced computing. The orders are being taken, the August timeline is set, and the hyperscalers are at the table. Whether the chips actually ship—and in what volume—will be determined by the intersection of chip architecture and international regulation, a collision that has defined Nvidia's fortunes in China for the past two years and shows no sign of resolving cleanly.