Control the chips and you control the intelligence. That is the quiet logic driving the most consequential geopolitical contest of our era — one waged not with armies but with export controls, fab investments, and nanometer tolerances.

In October 2022, the United States government imposed sweeping export controls designed not merely to slow a rival's military, but to structurally cap a nation's ability to develop artificial intelligence. The restrictions targeted advanced semiconductors and the equipment used to make them, effectively placing a ceiling on China's access to the compute needed to train frontier AI models.

Chips — those fingernail-sized wafers of silicon carrying billions of transistors — are no longer merely commercial goods. They are strategic assets, as significant to national power in the 2020s as oil reserves were in the 20th century.

Why Semiconductors Became the Front Line

Training a large language model requires an almost incomprehensible quantity of computation. GPT-4 is estimated to have required roughly 10²&sup5; floating-point operations to train. At scale, this translates directly into GPU-hours — and GPUs are manufactured by a vanishingly small number of firms, on fabs concentrated in just a few places on earth.

Key fact TSMC produces roughly 90% of the world's most advanced chips — those below 7 nanometers. No other company on the planet can do what TSMC does at scale. This single geographic concentration has become the central nerve of the entire AI industry.

90%Share of sub-7nm chip production controlled by TSMC — all on a 36,000 km² island
$500BCommitted US investment in domestic semiconductor manufacturing under the CHIPS Act through 2026
3nmCurrent leading-edge node size — about 1,000× thinner than a human hair, requiring ASML's EUV lithography machines
Partner · Affiliate

Your product name here

One or two sentences on what this product does and why it fits a reader of this journal. Keep the tone of the surrounding writing — it reads better and converts better.

Learn more →

The Chokepoints No One Talks About

ASML, a Dutch company, makes the extreme ultraviolet lithography machines required to print circuits at 7nm and below. There is no alternative supplier. The Netherlands has become a pivotal actor in the chip war simply because ASML exists there.

Similarly, the specialized software used to design chips — Electronic Design Automation tools — is dominated by three American firms: Synopsys, Cadence, and Mentor. China's chip designers depend on these tools. Export restrictions on software have become as significant as hardware bans.

“We thought the battlefield was tanks and aircraft carriers. It turned out to be nanometers and nitrogen-purified silicon.”

Senior policy analyst, cited in Congressional testimony, 2024

China's Response: The Long Game

Huawei's 2023 release of the Mate 60 Pro, powered by a domestically produced 7nm chip from SMIC, surprised analysts who had assumed export controls had set China's chip capability back by years. It hadn't. It had concentrated Chinese engineering talent on solving exactly the problems the restrictions targeted.

Oct 2022US Export Controls — Round One

BIS restricts sales of advanced AI chips (A100, H100) and chip-making equipment to China. The order is broader and more aggressive than expected.

Jan 2023Netherlands Joins the Restriction Regime

Under US pressure, the Dutch government restricts ASML from exporting its most advanced EUV machines to China — a decisive tightening of the chokepoint.

Sep 2023Huawei's 7nm Surprise

The Kirin 9000s chip reveals China's unexpected progress — and forces a reassessment of how effective export controls truly are.

2024–2026Escalation and Fragmentation

Further US restrictions. China counter-restrictions on rare earth minerals. The semiconductor world fragments into allied and non-aligned supply chains.

What This Means for AI Development

The practical consequence for AI researchers is a world in which compute access is increasingly a geopolitical function. Labs in the US and allied nations have relatively unconstrained access to NVIDIA's H100 and successors. Labs in China do not.

Implication The restrictions intended to slow China's AI development may be producing researchers unusually skilled at building efficient models — a capability that will matter enormously as the industry grapples with diminishing returns of raw scale.

Intelligence, it turns out, is not just a software problem. It is a hardware problem. And hardware is made of politics.