Nvidia is shifting its China strategy away from graphics processors to a new central processing unit called Vera, which faces fewer regulatory barriers than the company's restricted GPU exports. The company has informed Chinese customers that Vera CPUs for artificial intelligence data centers could be available as early as August, and is already inviting orders, according to three sources familiar with the discussions. This pivot highlights how Nvidia is rapidly adapting to U.S. export restrictions that have effectively frozen its traditional GPU business in China.

Why Is Nvidia Turning to CPUs When GPUs Are Its Strength?

Nvidia's GPU shipments to China have stalled for months, particularly its H200 graphics processing unit, which is its second-most powerful AI chip. The company's chief executive Jensen Huang acknowledged in October that Nvidia's market share in China had effectively dropped to zero, citing the impact of U.S. restrictions on advanced chip exports. Washington has authorized about 10 Chinese firms to purchase the H200 GPU, but none of those orders have been fulfilled so far, as Chinese authorities have not granted approval for shipments.

CPUs, or central processing units, face fewer regulatory barriers than GPUs under current U.S. export controls. This regulatory gap has created an opening for Nvidia to introduce Vera, its first standalone CPU designed specifically for agentic AI, which refers to systems that can execute tasks autonomously without constant human direction. The timing aligns with a broader industry shift from training large AI models to running inference workloads, which respond to user queries and rely more heavily on CPUs and custom accelerators rather than GPUs.

What Are the Early Signs of Chinese Adoption?

Interest from major Chinese cloud providers suggests the strategy may gain traction. One major Chinese cloud provider is preparing an initial order for more than 300 servers configured with two Vera CPUs each, according to one source. The company intends to first use these systems for testing before deciding whether to commit to larger, formal orders. However, another source cautioned that it remains unclear whether this early interest will scale up significantly, citing key challenges including software ecosystems, compatibility issues, and the difficulties of migrating existing workloads developed for domestic AI chips.

Nvidia is projecting $20 billion in revenue from Vera sales by the close of its fiscal year ending in January, positioning the Arm-based CPU in direct competition with Intel and AMD, whose processors are based on the long-dominant x86 architecture. When Vera was introduced in March, Nvidia said large cloud players such as Alibaba and ByteDance were collaborating with it to deploy Vera, though the company did not indicate at that time whether ordering had commenced.

How Does Vera's Pricing and Performance Compare?

• Single Chip Cost: Industry research firm SemiAnalysis estimated that a single Vera processor will be priced "well north" of $20,000 before any volume discounts.
• Full Rack Pricing: A fully built rack incorporating 256 Vera chips would cost roughly $10 million, with the final figure depending on the memory configuration.
• Performance Advantage: Nvidia says the chip delivers up to 1.8 times the performance of comparable processors from competing vendors.
• Target Market: Initial shipments are primarily going into large, turnkey racks favored by hyperscale customers, while simpler servers carrying two processors are expected to scale up later.

How to Understand Vera's Role in the Broader AI Chip Competition?

• Regulatory Advantage: CPUs face fewer export control restrictions than GPUs, allowing Nvidia to maintain a foothold in China despite broader U.S. sanctions on advanced semiconductor exports.
• Market Shift: The global AI industry is transitioning from training models to inference workloads, an area where CPUs and custom accelerators are mounting stronger competition against traditional GPU dominance.
• Supply Pressure: Reuters reported in February that Intel had informed Chinese customers that lead times for server CPU deliveries could extend up to six months, while AMD stated last month that the global CPU market is "tight," with demand running ahead of expectations and supply limitations likely to continue.
• Deployment Strategy: One of the sources said Chinese customers plan to initially deploy Vera chips only in their overseas data centers for testing, a cautious approach that suggests uncertainty about long-term viability.

The Vera rollout intensifies Nvidia's rivalry with established CPU suppliers Intel and AMD, both of which are working to expand server CPU availability for AI data center workloads. For Nvidia, the move represents a calculated bet that regulatory loopholes and shifting AI workload patterns can help it recover lost ground in one of the world's largest AI markets, even as its traditional GPU business remains frozen by export controls.