US Commerce Secretary Howard Lutnick said the chipmaker “must live with” strict licensing conditions attached to H200 exports during a congressional hearing.
The terms were negotiated with the State Department and include Know-Your-Customer requirements intended to prevent the chips from being diverted to military use.
The H200 is Nvidia’s second-most-advanced AI processor cleared for export to China under an October 2025 trade truce brokered by Presidents Donald Trump and Xi Jinping.
That agreement included a US pledge to postpone additional restrictions on technology shipments to Chinese firms by one year.
Despite authorisation in January 2026, Nvidia has reportedly not fully signed off on all the proposed conditions.
Under the framework, the Commerce Department’s Bureau of Industry and Security will review H200 export licence applications on a case-by-case basis.
Chinese end-users such as ByteDance, Alibaba, Tencent and DeepSeek face verification protocols designed to ensure shipments do not undermine semiconductor capacity available to US allies. We think he means Russia, Israel, Argentina and Hungary.
It is arguably insane because Washington is trying to micromanage the downstream use of a general-purpose processor in one of the most complex supply chains on Earth. AI accelerators are not uranium centrifuges.
Once an H200 leaves the factory floor and lands in a hyperscale data centre, verifying that it will never train, fine-tune or serve a model with dual-use potential is practically impossible. Know-Your-Customer rules might look tough on paper, but policing how clusters are partitioned, virtualised and re-tasked in production is a compliance nightmare.
It is economically self-defeating. Nvidia has already taken multibillion-dollar hits from previous export controls, and every extra layer of restriction accelerates Chinese substitution. Huawei and other domestic players do not need to outperform Blackwell. They just need to be good enough.
With the US throttling legal sales while China pours money into homegrown silicon, the US risks funding its competitor through lost revenue, weaker economies of scale and fragmented standards.
He declined to detail Chinese compliance, saying the “complex relationship” between the two countries is handled at the highest levels by the president.
China has historically been one of Nvidia’s biggest markets, but repeated export controls have cost the company an estimated $8bn in potential sales.
In January 2026, the White House imposed a 25 per cent tariff on certain advanced computing chips, including the H200 and AMD’s MI325X, although chips supporting US infrastructure buildout were exempt.
Nvidia’s newer Blackwell chips, already in use by US customers, are reportedly up to 10 times faster than the H200 for some workloads, while the forthcoming Rubin architecture remains excluded from Chinese export approvals.
Chinese chipmakers such as Huawei and Cambricon have gained ground despite US restrictions, with Huawei’s Ascend 910C positioned as a domestic alternative, though it is described as a generation behind.
