Nvidia is planning to produce an entirely new AI accelerator chip for the Chinese market, replacing its banned H20 and using a different architecture at a significantly lower price than the previous China-focused chip, Reuters reported.

The chip, set for mass production as early as June, is to use Nvidia’s latest Blackwell architecture and is expected to be priced at $6,500 (£4,793) to $8,000, compared to $10,000 to $12,000 for the H20, the report said, citing unnamed sources.

The H20, which the US effectively banned for export last month, uses the Hopper architecture that preceded Blackwell.

New architecture

Nvidia initially considered downgrading the H20 to comply with the latest US regulations, but reportedly found the architecture could not accommodate further modifications.

Reports from earlier this month had indicated Nvidia was readying a modified version of the H20.

The new chip’s lower price reflects its lower specifications and simplified manufacturing process.

It is to be based on Nvidia’s RTX Pro 6000D server graphics processor with conventional GDDR7 rather than more advanced high bandwidth memory (HBM), and is planned to use a simpler process than TSMC’s advanced Chip-on-Wafer-on-Substrate (CoWoS) packaging technology, Reuters’ report said.

The company is also reportedly developing a separate Blackwell-based chip for China that is scheduled for mass production in September, and about which little is known.

GF Securities estimated that Nvidia’s upcoming China-focused GPU would achieve a memory bandwidth of roughly 1.7 terabytes per second using GDDR7 memory, just inside an estimated cap imposed by the new regulations, compared to 4TB/s for the H20.

Nvidia reiterated earlier complaints that US regulations are limiting its access to the Chinese chip market and benefiting domestic Chinese competitors.

‘Effectively foreclosed’

“Until we settle on a new product design and receive approval from the US government, we are effectively foreclosed from China’s $50bn data centre market,” the company said in a statement.

The firm said last week that its market share in China has plummeted from 95 percent before 2022 to 50 percent currently.

Aside from its raw computing power, however, Nvidia also has a significant edge in its CUDA programming architecture, which is broadly used and makes it challenging for organisations to shift from one hardware platform to another.

Huawei is Nvidia’s leading competitor in China with its Ascend 910B and other upcoming AI chips.