Chip Race: Trump Authorises H200 Sales to China, Dividing American Congress

In early 2026, the Trump administration made a decision that broke one of the rare bipartisan consensuses in American tech policy: it authorised the sale of Nvidia H200 chips, equivalent to the most advanced generation of available AI accelerators, to Chinese companies.

This measure reversed the architecture of export controls that had been progressively built since 2022, creating a regime of increasing restrictions based on the premise that keeping China away from cutting-edge semiconductors was a matter of national security, not merely of trade policy.

What Changed and Why

On January 13, 2026, the American Department of Commerce published final rules that revised the licensing posture for chips equivalent to the Nvidia H200 and AMD MI325X. What was previously a ban transitioned to "case-by-case review", a change that, in practice, substantially facilitates the approval of sales.

The administration's justification mixes economic and geopolitical logic: American semiconductor companies, Nvidia, AMD, Intel, are losing revenue to competitors not operating under the same restrictions. And if China can obtain equivalent chips from other sources, American control would have limited impact.

The premise has partial support in the data: domestic Chinese chips already accounted for 41% of the AI accelerator market in China by 2025, with approximately half of the sales coming from Huawei. The argument is that the restriction was creating a competitive disadvantage for America without significantly slowing down China's AI programme.

Congressional Reaction

The bipartisan consensus that built the restrictions has not disappeared, it remains in Congress.

The reaction was immediate and multi-partisan. Republican and Democratic legislators proposed bills to reverse the decision, including the SAFE CHIPS Act (which would seek to revalidate previous restrictions) and the AI Overwatch Act (which would create oversight mechanisms for the sale of advanced chips).

Chris McGuire from the Council on Foreign Relations contextualises: "The existing bipartisan pressure to reverse the course will intensify as legislators recognise that AI computational power is becoming the world's most critical strategic asset." The question is not whether there will be resistance, but whether it will be swift enough to have an effect.

The Competitive Dimension

Market numbers reveal the complete context of the dispute.

American companies held approximately 93% of global visits to language models in August 2025. However, China's growth has been explosive: visits to Chinese models grew by 460% in two months, driven by DeepSeek, and without cannibalising traffic from other Chinese models, suggesting market expansion, not displacement.

DeepSeek, in particular, altered the perception of what is possible with more limited hardware. The model demonstrated that it is possible to achieve competitive performance with fewer high-performance chips, which partially invalidates the premise that hardware restrictions linearly slow down China.

Beyond Chips: Three Fronts of Competition

The CFR identifies that US-China competition in AI is not a single race, it is a decathlon with at least three simultaneous fronts:

Cutting-edge innovation: who develops the most advanced models. Here, the US leads, but the advantage was only seven months in 2025, a figure that may shrink with access to better chips.

Adoption for national security: China has a structural advantage in the speed of deployment in military and intelligence applications, given its centralised governance model.

Influence on global standards: China is active in international AI standards bodies and proposes the creation of the WAICO (World Artificial Intelligence Cooperation Organization), seeking a position of normative leadership.

Implications for Corporate Strategy

For executives making AI infrastructure decisions, the scenario has concrete consequences:

Export regulatory risk: American companies selling cloud or AI services to customers in countries that operate with Chinese tech partners need to review their export compliance programmes. The regime is in flux, decisions made today could be invalidated tomorrow by legislative reversal.

Supplier risk: Organisations relying on AI hardware (GPUs, TPUs) must monitor American regulatory developments. A reversal in export restrictions could alter prices and availability of cutting-edge hardware, both due to increased Chinese demand and trade reprisals.

Diversification of AI models: the rise of competitive Chinese-origin models (DeepSeek, Qwen, Ernie) creates real price and performance options that are already being adopted by companies outside the US. For CIOs assessing AI strategy, ignoring the Chinese model ecosystem is ignoring a growing share of the available alternatives.