The US has reportedly opened the door for 10 Chinese companies to buy NVIDIA’s powerful H200 AI chips, but the deal is not exactly jalan yet.
According to Reuters, the US Commerce Department has granted approval to firms including Alibaba, Tencent, ByteDance, JD.com, Lenovo and Foxconn to purchase the H200. The big catch? NVIDIA has reportedly not delivered any of the chips so far.
This is another chapter in the messy AI chip tug-of-war between the US and China. Back in December 2025, the US government allowed NVIDIA to sell H200 processors to selected Chinese customers, after previously blocking those sales over concerns that the hardware could support China’s military technology development.
Reuters previously reported that China agreed in January to import several hundred thousand H200 chips, with the first batch intended for three unnamed Chinese internet companies.
Why the H200 matters
The H200 is not NVIDIA’s absolute top dog — that would be the B200 — but it is still a monster AI processor. It sits above the H20, which had already been cleared for China earlier, and is far more capable for serious AI training and inference workloads.
For companies like Alibaba, Tencent and ByteDance, this kind of hardware is the fuel behind large language models, recommendation systems, cloud AI services, video platforms, search tools and enterprise AI products. Basically, if a tech giant wants to compete in the AI race, chips like these are the expensive ammo.
The approved Chinese companies can reportedly purchase up to 75,000 H200 chips, either directly from NVIDIA or through intermediaries. But Reuters says those firms stepped back after receiving guidance from the Chinese government.
The reason for that guidance is still unclear. Reuters says it was prompted by changes on the US side, but the exact changes have not been detailed.
China may not want to rely on NVIDIA forever
There is also a bigger strategic issue here. Chinese chipmakers have been building their own alternatives after years of US export restrictions, and Beijing has been pushing local companies to use domestic hardware to strengthen China’s own semiconductor ecosystem.
On top of that, Reuters says Chinese officials are reportedly worried that H200 chips sold under this arrangement could contain hidden vulnerabilities. One reason for concern is the way the sales are structured: for the US government to legally receive its 25 percent cut from H200 sales, the chips need to pass through US territory.
That kind of setup is exactly why this story is not just about specs. It is about trust, control and who owns the compute layer powering the next wave of AI.
Why Malaysia and SEA should care
For Malaysian readers, this may sound like a US-China business drama, but it can still affect us. A lot of AI services used by local startups, game studios, agencies, e-commerce teams and creators depend on cloud infrastructure from major global and regional players. If Chinese tech giants get access to stronger NVIDIA chips, their AI tools and cloud offerings could become more competitive across SEA.
At the same time, if China doubles down on its own chips instead, we could see a more split AI hardware landscape: US-led NVIDIA stacks on one side, China-made alternatives on the other. That could affect pricing, availability and which AI platforms Malaysian businesses eventually adopt.
This also matters for gaming and content platforms. Tencent and ByteDance are not small names in SEA — they touch games, social video, livestreaming, ads and creator tools. Better AI compute can improve everything from moderation and recommendations to game development pipelines and generative tools.
NVIDIA CEO Jensen Huang has previously warned that US export controls could weaken NVIDIA’s position in China. He also recently travelled to Beijing with President Donald Trump for a summit with Chinese President Xi Jinping. Whether that trip helps unlock actual H200 deliveries is still belum confirm.
For now, the headline is simple: permission may be granted, but the chips are still not moving.
Source: Engadget