Nvidia’s big AI chip return to China is looking a lot messier than expected.
According to comments from U.S. Commerce Secretary Howard Lutnick reported by Tom’s Hardware, Nvidia still has not sold any H200 AI GPUs to Chinese customers, even though the U.S. lifted the ban around four months ago. The reason, according to Lutnick, is not just Washington red tape. Beijing is also making it tough for local companies to bring those chips in.
The short version: China wants its own semiconductor ecosystem to grow, and letting every major tech firm rush back to Nvidia does not exactly help that mission.
Why the H200 matters
The H200 is not a gaming GPU you slot into your PC next to your RGB RAM, bro. It is a high-end AI accelerator built for data centres, model training, inference, and the kind of compute-heavy workloads powering modern AI services.
For big Chinese companies like Alibaba and ByteDance, access to Nvidia AI hardware can directly affect how fast they build and scale AI products. Tom’s Hardware notes that major Chinese firms were reportedly ready to order hundreds of thousands of Nvidia AI GPUs after the U.S. opened the door again, with exports allowed under a 25% fee.
But Lutnick says no H200 chips have actually been sold to China yet. That clashes with Nvidia’s own position from GTC 2026, where the company said it had received orders and export licences for several Chinese customers. So right now, the picture is blur: orders may exist, licences may exist, but actual shipments appear to be stuck.
There is also another possible bottleneck on the U.S. side. Reports say AI chip export applications are being closely reviewed at the Bureau of Industry and Security, with undersecretary Jeffrey Kessler personally checking individual cases.
China is pushing local chips hard
Beijing’s move is not surprising. China has been encouraging companies to use domestic AI chips from players like Huawei, Alibaba, Baidu, Cambricon, Moore Threads, and others.
Tom’s Hardware says Chinese customs officials have been told to block H200 imports in practice, while universities and R&D labs have had more room to get access. That is a very clear signal: commercial tech giants should support China’s local chip industry first.
This is painful for Nvidia. The company’s China market share has reportedly fallen to under 60%, compared with around 95% before U.S. sanctions disrupted the market. Jensen Huang has publicly argued against a full AI chip export ban, warning that pushing China away from U.S. hardware could allow Chinese AI chipmakers to dominate their own market instead.
Even with the H200 already behind Nvidia’s newer Blackwell platform and upcoming Vera Rubin GPUs, demand in China is still strong. The article even notes that some companies are looking at black market routes, which tells you how valuable these chips remain.
Why Malaysia and SEA should care
For Malaysia, this is not just another U.S.-China tech drama happening far away. AI infrastructure is becoming a major regional race, and Malaysia is trying to position itself as a serious data centre and cloud hub in Southeast Asia.
If Nvidia supply into China stays restricted, demand does not magically disappear. It can shift pressure across the wider region: cloud pricing, GPU rental availability, data centre planning, and enterprise AI costs may all feel the ripple effect. Malaysian startups, universities, banks, telcos, and media companies using AI services may not buy H200s directly, but they depend on platforms that do.
There is also a bigger strategic angle. SEA countries are watching both sides: U.S. chips remain the gold standard, while Chinese AI hardware is improving fast and may become cheaper or more accessible across the region. If China accelerates its local semiconductor push, SEA buyers may eventually get more options — but also more complicated choices around compatibility, software ecosystems, and long-term support.
For now, Nvidia is still the boss fight in AI hardware. But China clearly wants to build its own endgame build, and that could reshape the chip market that Malaysia’s AI ambitions depend on.
Source: Tom's Hardware