Nvidia returns to China. H200 chip sales breakthrough

Nvidia has successfully broken the geopolitical deadlock by securing key approvals from Washington and Beijing to sell its advanced H200 chips to select Chinese tech giants. At the same time, the American manufacturer is opening a new front in the battle for the lucrative artificial intelligence market by developing an optimized version of its Groq processors designed for inference tasks.

3 Min Read
Nvidia
source: Nvidia

The cold technology war between Washington and Beijing is gradually thawing. After months of regulatory limbo, Nvidia has received approval from both governments to sell its powerful H200 artificial intelligence chips to selected customers in China.

At the same time, the US giant is preparing a locally optimised version of the Groq processors, which signals a completely new strategy in the battle for dominance in the AI inference sector.

For Nvidia, this is a key business moment. The Chinese market has historically generated several per cent of the company’s total revenue, but rising trade barriers forced the company to halt production of the H200 for customers there last year.

However, the situation has improved. CEO Jensen Huang confirmed at a recent conference that supply chains are picking up again, and the company has already received official orders. Initial import approvals were to be obtained from the Middle Kingdom’s biggest tech heavyweights, including ByteDance, Tencent and Alibaba, as well as the rapidly growing startup DeepSeek.

However, the US manufacturer does not intend to stop at just high-powered chips for training models. Market information suggests that Nvidia is preparing an architecture based on technology from the recently acquired startup Groq for entry into the Chinese market.

These chips, specifically designed for inference processes (the generation of responses by already trained AI systems), are expected to make their market debut in May.

Crucially for the industry, the Groq layout variant for China will not be an artificially weakened version. Rather, it will be architecturally adapted to work with alternative systems and external infrastructure.

This will be necessary because US regulations still categorically prohibit the sale in China of Nvidia’s upcoming flagship platform called Vera Rubin, with which Groq chips work as standard.

This dual-track move is clearly a response to increasing competition from local players such as Baidu, who have already managed to deploy their own inference processors. The decision to supply proven H200 chips and Groq’s cutting-edge solutions in parallel shows an extremely pragmatic approach to geopolitical constraints.

TAGGED:
Share This Article