Nvidia Said to Pitch Vera Chips to Chinese Customers: August Launch, Now Taking Orders

TradingKey - On June 12, Eastern Time, Reuters reported, citing three people familiar with the matter, that Nvidia ( NVDA) has informed Chinese customers that its new Vera CPU, dedicated to AI data centers, will be available as early as August and is now open for orders. This Arm-based CPU, unveiled in March, is Nvidia's first standalone processor designed specifically for autonomous AI. Production readiness is complete, and the company is now fully focused on commercialization.

Jensen Huang predicted at a March launch event that Vera would become Nvidia's next multi-billion dollar business. Subsequently, during the May 20 earnings call, Nvidia CFO Colette Kress explicitly stated that revenue from standalone Vera CPUs is expected to approach $20 billion in fiscal year 2027, ending January 2027. Research firm SemiAnalysis estimates that the pricing for a single Vera unit prior to volume discounts will be well over $20,000, with a full-rack configuration of 256 chips totaling approximately $10 million.

Vera is designed specifically for AI agents, requiring robust single-thread performance to support real-time decision-making, tool calling, and inference control—tasks that are precisely the strengths of a CPU.

Nvidia's pivot to CPUs comes against the backdrop of blocked channels for its GPUs in China. Since 2025, Nvidia's market share in China has effectively dropped to "zero." Although the U.S. Department of Commerce adjusted the licensing policy for H200 exports to China from a "presumption of denial" to "case-by-case review" in January, it attached stringent conditions, such as export volumes not exceeding 50% of supply to the U.S. Since then, Nvidia has received almost no valid licenses, and Chinese buyers have yet to purchase a single H200. The CPU sector, where Vera competes, faces relatively more relaxed export controls.

However, selling CPUs to China still faces multiple obstacles. While Chinese customers have shown interest in Vera, initial deployment plans are limited to overseas data centers. Furthermore, major Chinese cloud providers are accelerating domestic substitution efforts; Alibaba Cloud, Tencent, and Baidu have all launched self-developed AI chips, limiting the incentive for large-scale procurement of overseas CPUs in the short to medium term.

From a global supply and demand perspective, AI competition is shifting from model training to inference computing, making the role of CPUs more critical than ever. Intel ( INTC) and AMD ( AMD) have both warned that the global CPU market is tight, with server CPU capacity for 2026 largely sold out and plans to raise prices by 10% to 15%. To ensure Vera shipments, Nvidia recently decided to reduce SOCAMM2 memory from 192GB to 96GB to safeguard CPU production capacity, underscoring that supply constraints have become an industry-wide issue.

Arm ( ARM) CEO Rene Haas noted earlier this month that implementing export controls on AI CPUs is nearly unfeasible. He explained that it is difficult to set clear performance thresholds and memory bandwidth limits as is done with GPUs. If restrictions were actually imposed, they would have to target almost all CPUs, given that their range of applications is far broader than that of GPUs.

Vera opens a new path for Nvidia to sell into China, bypassing some regulatory hurdles. However, whether the $20 billion revenue target can be achieved depends on software ecosystem compatibility, customers' willingness to scale, and uncertainties in the regulatory environment.