Nvidia sees tepid demand for new RTX6000D AI chip from Chinese tech firms: sources
The chip, mainly used for AI inference tasks, is seen as expensive for what it does, two people with knowledge of purchasing discussions say
The chip, mainly used for AI inference tasks, is seen as expensive for what it does, two people with knowledge of purchasing discussions say
An Nvidia representative, said in a statement that the,
Market is competitive – we offer the best products we can
Alibaba did not address a Reuters query seeking comment. Tencent and ByteDance did not respond to requests for comment.
The extent to which China has access to cutting-edge AI chips is one of the biggest flashpoints in US-Sino trade tensions.
On one hand, the US has retreated from its previous position of more severe export restrictions. But Beijing has become increasingly keen for Chinese firms to switch to domestic chips even if Nvidia’s products are preferred by many companies.
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.
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.
It remains to be seen whether the planned B30A will be approved in Washington. Also based on Blackwell architecture, it is expected to deliver up to six times more performance than the H20, while costing only twice as much, Reuters reported this month.
READ the latest news shaping the Nvidia market at Newsvidia
Nvidia sees tepid demand for new RTX6000D AI chip from Chinese tech firms: sources, source
