Boosting Nvidia’s bottom line won’t make Americans safer

IN DECIDING where to draw the line on sales of advanced semiconductors to China, the White House should prioritise the security of Americans over the interests of companies. Its latest moves risk getting that balance wrong.

Earlier this month, US President Donald Trump granted Nvidia permission to ship its H200 artificial intelligence (AI) chips to China, while loosening restrictions on rivals Intel and Advanced Micro Devices. Although not Nvidia’s most cutting-edge offering, the H200 is at least a generation ahead of Chinese-made alternatives.

The Institute for Progress noted that the chip provides 32 per cent more processing power and 50 per cent more memory bandwidth than the Ascend 910C, the best chip from Huawei Technologies. It is also cheaper to use and more reliable.

The Commerce Department estimated that Huawei will produce only 200,000 Ascend chips this year, compared to the 14 million AI chips deployed by US companies.

Proponents of the new policy argued that Huawei is catching up to US rivals by networking more of its chips together, and could expand production dramatically next year.

They warned that too-tight controls will effectively cede the Chinese market, depriving Nvidia and others of revenue they could invest in designing more powerful semiconductors.

AI czar David Sacks said the White House aims to slow the mainland’s own progress in chipmaking by keeping Chinese customers “addicted” to US technology. More crassly, the US has also claimed a 25 per cent “cut” of any H200 sales.

The logic is dubious. Few believe the company can come close to meeting domestic demand, which means China can import H200s to fill the gap while continuing to force companies to buy any available Huawei chips.

As those locally made products become more capable, demand for US semiconductors can be choked off. The government has already said companies must apply for licences to import H200s.

Meanwhile, any H200 sales will erode the greatest US advantage over China in AI – aggregate computing power. Chinese AI researchers are world-class, they have access to more data and they face fewer energy constraints.

What they lack is compute: As at mid-2025, the US accounted for nearly three-quarters of global AI supercomputer capacity, critical for training large-scale models, compared to 14 per cent for China.

Chinese authorities are likely to direct H200 imports to the highest-performing companies and, through civilian intermediaries, to defence research. That would speed advances in Chinese frontier models and weapons development, and could help Chinese cloud providers steal business from US rivals globally.

Tellingly, AI companies such as Anthropic PBC and hyperscalers such as Amazon.com and Microsoft Corp have all supported export controls in the past.

Any pushback will have to come from Congress, where there remains bipartisan support for stricter controls. As legislators consider whether to freeze current limits in place for a set period, they should challenge the White House to back up its estimates of Huawei’s technological progress and seek testimony from AI developers and cloud providers about the impact of loosened export controls.

The administration should also be questioned about pushback from allies such as Japan and South Korea, who have faced pressure to restrict their own technology exports only to be blindsided by shifts in US policy.

Above all, Congress should refocus attention on White House policies that are impeding US competitiveness in AI.

Antipathy towards clean-energy projects needlessly limits US power supplies. An overly harsh immigration crackdown is deterring talented researchers. Disrupting funding for basic research will impede US innovation over time.

If the administration really wants to remove roadblocks to American AI dominance, those would be better places to start. BLOOMBERG