Nvidia’s China problem is not over just because Washington opened the door.
President Donald Trump said Beijing is refusing to approve Chinese companies’ purchases of Nvidia’s H200 AI chips, even after the US side reportedly cleared the path for sales. Speaking after a two-day summit with Chinese leader Xi Jinping, Trump said China chose not to sanction the purchases because it wants to build up its own chip ecosystem.
That is a big deal because, according to earlier reporting, the US Commerce Department had already allowed around 10 Chinese companies to buy H200 chips. The names reportedly include major tech players such as Alibaba, Tencent, ByteDance, and JD.com. But despite the approvals, no H200 chips have actually shipped to China.
The US green light is not enough
The export setup is already pretty strict. Lenovo and Foxconn, acting as distributors, have also received approval. Under the rules formalised in January, each H200 chip must go through US territory for third-party inspection before being re-exported to China.
On top of that, Nvidia has to pay a 25% fee to the US Treasury for every H200 sale to China. So yes, even with approval, this is not exactly a normal GPU transaction.
But all those conditions mean nothing if Beijing does not allow its own companies to proceed. Trump said the issue was discussed with Xi and suggested something could still happen, but he did not give details. US Trade Representative Jamieson Greer also said any next move now depends on China.
Commerce Secretary Howard Lutnick made a similar point earlier, saying China is blocking imports to push local semiconductor development. Huawei is one of the obvious names benefiting from that direction. Some Chinese firms had apparently placed orders with Nvidia earlier this year, then later told the company they could not continue.
Jensen Huang’s last-minute trip did not fix it
Nvidia CEO Jensen Huang was not initially part of the US delegation for the Beijing summit. He was reportedly added late and joined during a refuelling stop in Alaska, which naturally raised expectations that H200 sales might finally move.
But based on Trump’s comments after the meeting, the stalemate is still there.
For Nvidia, the financial hit is real, but the company is not banking on a China comeback. Its US$78 billion annual revenue guidance assumes no H200 recovery from China. Analysts, however, believe a working export framework could bring back around US$3.5 billion to US$4 billion in annual revenue from the country.
That is especially painful considering Nvidia once had about 95% of China’s AI accelerator market. Huang has since described its share there as basically zero.
Why Malaysia and SEA should care
At first glance, this sounds like US-China drama only. But for Malaysia and Southeast Asia, it matters because AI infrastructure is becoming a regional arms race.
Malaysia is already seeing more interest in data centres, cloud infrastructure, AI startups, and enterprise AI adoption. If China shifts harder toward homegrown chips, Nvidia supply and pricing dynamics could change globally. If fewer H200s go into China, those chips may be redirected elsewhere — but the bigger picture is messier, because geopolitics can make hardware roadmaps unpredictable.
For SEA companies, the lesson is simple: AI is not just about software anymore. Access to chips, cloud capacity, and export rules can decide who gets to build faster. Malaysian startups and enterprises relying on US cloud providers may not feel the direct impact today, but the global GPU supply chain still affects pricing, availability, and which AI platforms scale fastest.
There is also a strategic angle. China pushing domestic AI chips could accelerate competition against Nvidia alternatives. They may not beat Nvidia immediately on efficiency or software maturity, but if adoption is forced at scale, the ecosystem can improve quickly. That could eventually give SEA buyers more options — or more fragmentation to deal with.
Trump also said AI guardrails were discussed with Xi, while US officials have floated a dedicated bilateral channel for AI talks. The White House has also targeted so-called distillation attacks, where Chinese AI developers are accused of using outputs from advanced US models to train rival systems more cheaply.
So this is not just about one GPU. It is about who controls the next layer of AI power — chips, models, cloud, and rules included.
Source: Tom's Hardware