H200 exports mark reset in US-China chip conflict

On January 13, 2026, the Trump administration formally authorised restricted gross sales of Nvidia’s H200 AI chips to China, topic to the approval of non-military use circumstances and the overview of chips by a third-party testing laboratory.

This determination is being hailed as a major improvement in US-China expertise coverage as a result of, since 2022, Washington has restricted Beijing’s entry to superior AI chips with computing energy past the A100. The H200 is Nvidia’s second-best chip and, till now, Chinese language firms may solely procure the H20, a degraded model of the H200 with almost seven occasions much less complete processing energy (TPP).

Lifting export controls on the H200 successfully ends this coverage of containment, nevertheless it additionally seems to be a rigorously orchestrated shift pushed by the truth that Huawei is quick rising as a possible competitor to Nvidia in AI chips.

For instance, Pakistan has been in talks with Huawei’s vp concerning the event of GPU-based AI infrastructure in Pakistan beneath the federal government’s cloud-first coverage. Different creating nations are more and more searching for options to American cloud suppliers, pushed by price concerns in addition to geopolitical alignment preferences. On the Belt and Highway Boards for Worldwide Cooperation, China has repeatedly emphasised its International AI Governance Initiative. On the similar time, China’s open-source mannequin, DeepSeek-R1, has been described as “China’s Sputnik second”.

Nevertheless, if we benchmark the efficiency of Nvidia’s chips in opposition to Huawei’s, the US authorities’s assumption that Huawei is changing into a robust competitor seems weak. Huawei has numerous product strains and doesn’t have a specialised area of interest in GPUs similar to Nvidia’s. Based on DeepSeek researchers, Huawei’s top-of-the-line chip, the Ascend 910C, performs at solely about 60% of Nvidia’s H100, the predecessor of the H200. Equally, Huawei’s annual chip manufacturing charge is estimated at beneath a million models, which can not meet the rising demand of Chinese language firms for high-performance chips.

As per Nvidia’s roadmap, the reminiscence bandwidth of its top-of-the-line chips is anticipated to extend from 8 TB per second in 2026 to twenty TB per second in 2027, whereas Huawei’s roadmap suggests its chips will attain a bandwidth of round 9 TB per second by 2028. In contrast, Nvidia’s Rubin Extremely is anticipated to exceed 50 TB per second by 2028.

To take full benefit of those loosened export controls, Chinese language expertise firms have reportedly already positioned orders for greater than two million H200 chips, priced at round $27,000 every and paid for upfront, bringing the full export invoice to roughly $54 billion. With Nvidia’s stock reportedly operating low at round 700,000 models, the corporate has already ramped up manufacturing to fulfil new orders, resulting in a scarcity of high-performance reminiscence chips as a consequence.

This coverage reversal, from containment to conditional opening, has offered a much-needed reset for Chinese language firms. China already leads in a number of different features of the AI worth chain, together with knowledge availability, algorithms, human sources and electrical energy era. Entry to high-end GPU chips had been the first bottleneck over the previous three years. Whereas the USA continues to carry commanding benefits in semiconductor design and manufacturing partnerships with TSMC and different main foundries, Beijing now has a chance to study, adapt and compete on extra equal footing.

That stated, Beijing stays cautious about absolutely reopening its market to Nvidia. Authorities have reportedly knowledgeable expertise firms that purchases shall be authorised solely beneath particular circumstances, comparable to for analysis functions. Native companies are additionally being discouraged from stockpiling US chips with out approval, in an effort to cut back dependence on overseas expertise the place attainable.

In a nutshell, by permitting H200 exports, the US has successfully pressed the “reset” button on the chip wars. For Nvidia, it represents a $54 billion validation of its technological supremacy. For China, it removes the first bottleneck constraining its AI ambitions. Whereas the US at present retains the lead in uncooked processing energy and reminiscence bandwidth, Beijing’s cautious method to stockpiling suggests it’s unwilling to commerce long-term self-reliance for short-term {hardware} good points.

On this high-stakes sport of silicon diplomacy, the US is betting on the velocity of its innovation, whereas China is betting on its skill to combine these chips right into a superior, state-led AI ecosystem.

THE WRITER IS A CAMBRIDGE GRADUATE AND is working AS A STRATEGY CONSULTANT