Nvidia’s New RTX6000 AI Chip Met With Dismal Demand From Chinese Firms

Nvidia’s RTX6000D, its newest artificial intelligence chip tailored for the Chinese market, has seen lukewarm demand with some major tech firms going so far as not placing any orders, the SCMP reported citing sources. 

The reason for the snub: the RTX6000D, designed mainly for AI inference tasks, is seen as expensive for what it does (just wait until the rest of Nvidia’s produce lineup gets the same treatment when China reverse engineers its products and sells them for 90% off). They added that testing of samples showed its performance lags the RTX5090 – a chip banned by the US government for use in mainland China, but which is still readily available through grey market channels at less than half the RTX6000D’s price of around 50,000 yuan (US$7,000).

Chinese technology giants – including Alibaba, Tencent and ByteDance – are also waiting for clarity on whether orders for Nvidia’s H20 chip will be processed, separate sources said earlier this month. The US firm regained permission to sell the H20 in July, but shipments have yet to restart.

Chinese firms are also hoping that Nvidia’s B30A, a much more powerful graphics processing unit (GPU) than the H20, will be approved by Washington.

Those three chips are downgraded versions of GPUs sold outside China, developed to comply with export restrictions put in place by the US, which wants to rein the mainland’s tech progress and retain its lead in AI development.

The muted demand for the RTX6000D contrasts with optimistic projections from sell-side analysts. JPMorgan said in a report last month that it expected some 1.5 million RTX6000Ds to be produced in the second half of this year, while Morgan Stanley predicted in July that Nvidia would have 2 million RTX6000Ds in its pipeline. 

Nvidia began shipping the RTX6000D this week, according to one of the sources. 

The RTX6000D is based on Nvidia’s latest Blackwell architecture with conventional graphics double data rate memory and a memory bandwidth of 1,398 gigabytes per second, just below the 1.4 terabyte threshold set under restrictions laid out by the US in April. It was developed in part to fill a void left by the H20, which was banned from sale in April before that decision was reversed.

The H20, which is priced between US$10,000 and US$12,000, uses older Hopper architecture but has a greater memory bandwidth of 4TBs per second. Shipments of the H20, however, have not started for several reasons, including Nvidia’s need to sort out some issues related to a recent deal to give the US government a portion of its China sales.

In unrelated news, on Monday Beijing accused Nvidia of violating China’s anti-monopoly law, casting more uncertainty on its business in the world’s second-biggest economy. The move came as delegations from both sides are meeting in Madrid this week to discuss a trade agreement.

Chinese authorities have also summoned companies, including Tencent and ByteDance, over their purchases of the H20, asking them to explain their reasons and expressing concerns over information risks, sources have said. Nvidia has said its products do not pose any back door risks that would give anyone a remote way to access or control them.