BREAKINGVIEWS-China can crack AI code without Nvidia

Reuters12-12 11:01
BREAKINGVIEWS-China can crack AI code without <a href="https://laohu8.com/S/NVDA">Nvidia</a>

The author is a Reuters Breakingviews columnist. The opinions expressed are her own.

By Robyn Mak

HONG KONG, Dec 12 (Reuters Breakingviews) - Does China need Nvidia NVDA.O? On first glance, the answer looks like a resounding "yes". Chinese tech firms Alibaba 9988.HK and ByteDance are eyeing large orders for the U.S. titan's H200 artificial intelligence processors after U.S. President Donald Trump on Monday said he would lift a ban on exports to the People's Republic. But that looks set to change as Huawei and domestic peers gradually close the AI chip gap.

Washington's years-long sanctions and export controls have left China's tech industry struggling with an acute shortage of leading-edge semiconductors. That has resulted in a thriving black market for off-limits silicon smuggled into the country, as well as legal workarounds like training large-language models in overseas locations. And while mainland silicon outfits like the $85 billion Cambricon Technologies 688256.SS and recently listed Moore Threads 688795.SS are making substantial progress in chips primarily used for responding to queries from users – a process known as inferencing – they have yet to produce a competitive processor for training models.

Trump's latest move will come as a relief for Nvidia boss Jensen Huang. AI chip sales in China are on track to top $189 billion by 2029, up from $54 billion in 2026, according to research firm Frost & Sullivan. Crucially for Huang, Chinese regulators have signalled they will permit some sales of the H200, the $4.5 trillion company's second-most powerful AI chip, provided buyers justify why they need them, the Financial Times reported, citing sources.

But if Huang is hoping for a thaw in Beijing, where officials have blasted Nvidia over security concerns and launched an anti-trust investigation into the company, he will be disappointed. The Chinese government is stepping up its campaign to wean the country off foreign technology; recently, it banned foreign AI chips from state-funded data centres, Reuters reported, citing sources, a move that could divert billions of dollars of sales to local manufacturers. These efforts are paying off: Qingyuan Lin at Bernstein reckons local supply of inference chips will exceed demand by 2028, with domestic chipmakers accounting for over 90% of the market.

Moreover, Huawei has laid out a three-year roadmap to challenge Nvidia in processors used for training. Currently, the privately-held Chinese group's most advanced chip significantly lags the H200 in both computing power and memory bandwidth, accordingto the Institute for Progress. But Huawei is making impressive advances in so-called networking and interconnect technology used to link thousands of less-powerful processors together to work as one, potentially erasing Nvidia's individual chip advantage. The U.S. giant's return to China may be short-lived.

Follow Robyn Mak on X.

CONTEXT NEWS

The United States will allow Nvidia's H200 processors to be exported to China, U.S. President Donald Trump said in a post on Truth Social on December 8. In exchange, the U.S. government will collect a 25% fee on sales of the company's second-best artificial intelligence chips, Trump said.

Chinese companies ByteDance and Alibaba are keen to place large orders for the H200, should Beijing give them the green light, Reuters reported on December 10, citing sources.

Chinese regulators gathered representatives from companies including Alibaba, ByteDance and Tencent and asked them to assess their demand for the H200, The Information reported on December 10, citing sources.

Beijing is planning to limit access to Nvidia's chips, the Financial Times reported on December 9, citing sources. Buyers may be required to go through an approval process to explain why domestic providers were unable to meet their needs, the report added.

Chinese chipmakers fill the AI inference chip shortage https://www.reuters.com/graphics/BRV-BRV/dwpkqjazxpm/chart.png

(Editing by Peter Thal Larsen; Production by Ujjaini Dutta)

((For previous columns by the author, Reuters customers can click on MAK/ robyn.mak@thomsonreuters.com))

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Comments

We need your insight to fill this gap
Leave a comment