China has finally approved purchases of Nvidia’s H200 AI chip, ending months of “yes, but maybe not” for its local customers.
The go-ahead came as Nvidia chief executive Jensen Huang rolled into China, after the Trump administration said late last year that the chips could be sold but left everyone guessing whether Beijing would actually sign off.
The first approval reportedly covers several hundred thousand H200s, worth about $10 billion, with Alibaba and ByteDance among the first to get the nod.
Beijing’s move is being read as a small thaw with Washington ahead of President Donald Trump’s China trip, pencilled in for April, following the two sides’ October trade truce.
Chinese buyers have had to submit paperwork explaining how they will use the chips, and officials have been nudging them towards “necessary” work like advanced AI research and development.
Officials are pushing firms to buy domestic silicon too and are requiring homegrown chips for some training tasks and most inference workloads, the bit where AI spits out answers like a chatbot.
Beijing wants its best AI developers shipping models quickly, but it also wants a self-reliant semiconductor industry, and it is sick of everyone defaulting to Nvidia’s hardware and software stack.
Huang has been doing his usual pre-Lunar New Year circuit, hitting Shanghai, Beijing and Shenzhen, where Nvidia has offices, and getting spotted touring markets and handing out kumquats at a company party.
Despite the photo ops, he has not met any senior Chinese officials on this trip, a change from last year, when he reportedly met Vice Premier He Lifeng.
This latest clearance breaks a logjam that started last April when Washington initially banned sales of H20 chips, the cut-down China-friendly cousin of the H200, then reversed course while Beijing told companies to steer clear anyway on “cybersecurity” grounds Nvidia said were nonsense.
After an October meeting between Trump and Chinese leader Xi Jinping in South Korea, Washington cleared the way for the H200, while insisting that Nvidia ensure US supply stays healthy and that buyers can prove they have proper security before anything ships.
Huang did not hide the damage, saying in October that the company’s share in China’s GPU market for AI “had fallen to zero from 95 per cent” while the two governments played tug-of-war.
Chinese rivals have been scrambling to fill the gap by bundling piles of weaker chips, and some researchers reckon restricted access to top-end Nvidia parts could still widen the US-China AI divide.
Huang is expected to head to Taiwan next to talk manufacturing with suppliers, because nothing says “careful export policy” like lining up more H200 supply to feed a market that both governments keep trying to throttle.
