Nvidia Halts China Chips, Pivots to New 'Vera Rubin' Platform
Nvidia has reportedly halted production of its H200 AI chips for China, shifting resources to accelerate its next-gen "Vera Rubin" platform. The new platform includes the Vera CPU, a chip optimized for AI reasoning workloads, signaling a strategic pivot away from the uncertain Chinese market due to U.S. export controls.
The pivot away from China-specific chips follows years of escalating U.S. export controls that began in October 2022, aimed at restricting China's access to advanced semiconductor manufacturing equipment and high-performance GPUs. This regulatory pressure has created significant uncertainty, impacting Nvidia's revenue, which saw a 21.2% decline in China for the fiscal year ending in January 2026. The "Vera Rubin" platform represents a strategic shift to a rack-scale system treated as a single computational unit. This integrated design includes six distinct chips: the Rubin GPU, the Vera CPU, an NVLink 6 Switch, the ConnectX-9 SuperNIC, a BlueField-4 DPU for data processing and security, and a Spectrum-6 Ethernet Switch. At the heart of the new architecture is the Vera CPU, featuring 88 custom-designed ARM cores, and the next-generation Rubin GPU. The platform is engineered to slash the cost of AI inference by up to 10 times compared to the already formidable Blackwell generation, a critical metric for large-scale AI deployment. The platform is named for Vera Rubin, the American astronomer whose observations of galaxy rotation curves provided the first direct evidence for the existence of dark matter. Her work revealed that the observed motion of galaxies was impossible without a vast amount of unseen mass, revolutionizing cosmology. This strategic pivot has direct manufacturing implications, with reports indicating that Nvidia is reallocating H200 production capacity at its foundry partner, TSMC, directly to the new Vera Rubin architecture. The move signals a prioritization of next-generation platforms for major U.S. tech clients like Google and OpenAI over a Chinese market with mounting regulatory hurdles. Simultaneously, Beijing has been actively encouraging its domestic AI firms to prioritize locally produced chips, fostering a competitive landscape with companies like Enflame Technology and Moore Threads. This push for self-reliance complicates the market for any foreign semiconductor firm, even for chips compliant with U.S. export rules.
