Nvidia Launches Affordable AI Chip for Chinese Market
Nvidia is gearing up to introduce a new artificial intelligence chip for the Chinese market, intended to be offered at a notably lower price than the recently limited H20 model, as reported by Reuters based on information from knowledgeable sources.
New AI Chip Pricing and Specifications
The upcoming GPU, which is part of Nvidia’s latest Blackwell architecture for AI chips, is projected to be priced between $6,500 and $8,000. This marks a significant reduction from the H20’s pricing of $10,000 to $12,000. Sources indicated to Reuters that this lowered price corresponds to less robust specifications and more straightforward manufacturing processes.
Technical Features of the New Chip
The chip will reportedly utilise Nvidia’s RTX Pro 6000D, a GPU designed for servers, and will use standard GDDR7 memory as opposed to the more sophisticated high-bandwidth memory (HBM). It will not incorporate TSMC’s advanced Chip-on-Wafer-on-Substrate (CoWoS) packaging technology, according to the insider information. This level of detail about its pricing and production strategy has not been disclosed before.
Production Timeline and Market Impact
Mass production of the new chip could commence as soon as June, as per three unnamed sources who wished to remain anonymous due to the nature of their information.
An Nvidia spokesperson informed Reuters that the company is still investigating its limited options, stating that until a new product design is finalised and approved by the U.S. government, Nvidia remains effectively barred from accessing China’s $50 billion data centre market. TSMC chose not to comment on the matter.
Nvidia’s Significance in the Chinese Market
The Chinese market is crucial for Nvidia, contributing to 13% of the company’s revenue in the last fiscal year. This is the third occasion Nvidia has had to redesign a GPU in response to tightening U.S. export restrictions aimed at limiting China’s technological progress.
Though the new GPU is expected to be less powerful than the H20, it is viewed as a tactic to sustain Nvidia’s competitiveness in a market where it has seen a substantial loss due to export limitations. Huawei, with its domestically manufactured Ascend 910B chip, has positioned itself as a significant competitor.
Future Developments and Industry Perspectives
According to one investment director at White Oak Capital Partners, domestic technologies from companies like Huawei are anticipated to catch up with the performance levels of downgraded versions within the next one to two years. He noted that Nvidia’s distinction lies in its CUDA software platform, which aids in AI application development and has cultivated a robust developer ecosystem.
As reported by two Reuters sources, Nvidia is also working on a second Blackwell-based chip specifically designed for the Chinese market, with production potentially beginning in September; however, the specifications for this chip remain unclear.
Challenges and Architectural Considerations
Nvidia had previously considered creating a downgraded version of the H20 but shelved the plan due to architectural constraints imposed by the latest U.S. regulations. The company’s CEO has confirmed that the older Hopper-based architecture, which supports the H20 model, can no longer be adapted to meet current export standards.
The final designation for the new chip is still pending confirmation, but a Chinese brokerage, GF Securities, has speculated in a recent analysis that it might bear the name 6000D or B40, although no sources or pricing specifics were provided.