TLDR:
- Nvidia launches a lower-cost AI chipset tailored for China with mass production starting June.
- The new GPU uses less advanced tech, making it cheaper but less powerful than existing models.
- U.S. export restrictions limit Nvidia’s ability to supply high-end chips to China.
- Nvidia ramps up AI chip and supercomputer production in the U.S. alongside new China strategy.
Nvidia is preparing to introduce a new AI chipset designed specifically for the Chinese market, priced significantly lower than its premium H20 models.
According to Reuters, mass production of the new GPU will commence in June, marking a strategic move to maintain Nvidia’s presence in China despite growing export restrictions imposed by the U.S. government.
The new chipset, built on Nvidia’s latest Blackwell architecture, will cost between $6,500 and $8,000, considerably cheaper than the $10,000 to $12,000 price range of the H20 series. The reduced price reflects the adoption of more modest technical specifications and streamlined production processes.
Technical Trade-Offs Behind the Affordable GPU
Unlike Nvidia’s flagship AI processors, this new GPU uses standard GDDR7 memory instead of the more advanced high-bandwidth memory (HBM), a choice that limits its performance capabilities. The chipset is based on the RTX Pro 6000D server-class GPU but does not incorporate TSMC’s cutting-edge Chip-on-Wafer-on-Substrate (CoWoS) packaging technology. This simpler design approach enables Nvidia to meet cost targets while conforming to export regulations.
The U.S. government has enforced strict controls on GPU memory bandwidth for chips exported to China, capping it around 1.7 to 1.8 terabytes per second (TB/s). In contrast, the H20 chipset delivers bandwidth of up to 4 TB/s, making it significantly more powerful but restricted from direct export without special licensing.
Navigating U.S. Export Restrictions
Notably, since 2022, the U.S. has limited the export of high-performance AI chips to China, prompting Nvidia to develop the H20 specifically tailored to these constraints.
However, recent policy changes have tightened these controls further, requiring Nvidia to obtain licenses to export H20 GPUs to China, a process expected to cost the company around $5.5 billion by the end of its fiscal quarter.
An Nvidia spokesperson acknowledged the challenges stating;
“Until we select a new design and receive approval from the U.S. government, we are effectively locked out of the $50 billion data center market in China.”
Despite this, Nvidia remains committed to serving Chinese customers within the framework of these restrictions.
Additionally, Nvidia is reportedly developing another Blackwell-based chip for China, slated for production in September, though details about this new processor remain under wraps.
Nvidia’s Approach amid US/China Trade Wars
While navigating export hurdles abroad, Nvidia is simultaneously accelerating its AI manufacturing efforts in the United States. Last month, the company announced the opening of over one million square feet of new manufacturing space in Texas, dedicated to building and testing Blackwell chips and AI supercomputers.
In partnership with firms like Foxconn and Wistron, Nvidia plans to open factories in Houston and Dallas within the next 12 to 15 months. The Phoenix, Arizona facility, operated by TSMC, has already started chip production. Collaborations with packaging specialists Amkor and SPIL are also underway to optimize chip assembly and testing.
Nvidia forecasted that its investments could yield up to $500 billion in AI infrastructure production in the U.S. over the next four years. The company further that this expansion will create hundreds of thousands of jobs and bolster America’s economic security in the long term.
