The Strategy Behind Washington’s China Chip Deal

Why America’s AI dominance depends on keeping China supplied

When Donald Trump’s AI adviser David Sacks — chief architect of the America’s AI Action Plan — set out what “winning” the AI race means, he was blunt: secure 80–90 per cent global market share for the US tech stack; build an advantage that cannot be leapfrogged; and ensure American hardware powers the world’s data centres, much as Huawei once sought to dominate 5G.

By that definition, you cannot “win” without selling into China.

That’s the strategic backdrop to Sunday’s announcement that Nvidia and AMD will hand over 15 per cent of their China chip revenues to the US government in exchange for export licences — a move trade experts have called “unprecedented.”

In effect, Washington has made itself a direct revenue partner in the China business of America’s most important chipmakers, something with few precedents in US commercial policy. The deal could funnel more than $2 billion into the Treasury this year alone, turning export control into both a geopolitical lever and a lucrative cash stream.

It’s not the first time a power has sold essential infrastructure to rivals under its own terms. A century ago, Britain’s undersea cables carried much of the world’s communications traffic, including that of its competitors — at British prices, under British rule. And when war came, that leverage became a weapon.

Within hours of entering the First World War, Britain cut Germany’s cables, forcing its messages onto routes London could intercept and censor. The point wasn’t just to profit in peacetime — it was to ensure that, when it mattered, the world’s communications still ran on British terms. Washington’s new chip deal with China works on the same logic.

The deal also fits into a wider Trump-era pattern of hands-on intervention in strategic industries. In recent months, the White House has insisted on a “golden share” in Nippon Steel’s takeover of U.S. Steel, imposed 100% tariffs on offshore semiconductor production unless it’s moved to the U.S., and tied foreign investment approvals to direct government stakes.

Source: RAND corporation

In any case, last months’ AI Action Plan made no bones about its goal: to “achieve and maintain unquestioned and unchallenged global technological dominance” and make the American AI stack the “gold standard worldwide.” That ambition collides with a simple commercial truth: China is too big a market to abandon entirely, and too sophisticated to be locked out indefinitely.

Experience bears this out. Just last month, the FT reported that more than $1bn worth of Nvidia’s most advanced AI chips entered China in only three months, despite tightened US export controls. These weren’t trickling across borders in hand luggage. They were arriving in bulk — ready-built racks of B200 processors, weighing 150kg each, sold openly on Chinese platforms at a premium. Officially, the sales were legal for Chinese buyers so long as tariffs were paid; the breach was on the export side, far from Washington’s reach.

The lesson is clear: denial alone is porous. Black markets fill the vacuum, enriching middlemen and eroding US leverage. Licensing allows Washington to control, monitor and monetise the flow, while ensuring that the chips at the heart of Chinese AI systems are still designed in California.

Commerce Secretary Howard Lutnick publicly framed the deal as a compromise: “We don’t sell them our best stuff, not our second-best stuff, not even our third-best.” Yet many national security veterans still see it as a strategic misstep.

“You’re selling our national security
for corporate profits.”
Liza Tobin, former National Security Council China director

Elsewhere, a bipartisan group, including Trump-era officials Matt Pottinger and David Feith, called the H20 chip “a potent accelerator of China’s frontier AI capabilities”, not an outdated chip. Nvidia rejects that charge, insisting the H20 is safely below the thresholds that matter militarily.

Regardless, it looks as if Nvidia CEO Jensen Huang’s argument that an outright ban would hand Huawei the Chinese market, letting it dominate AI chip sales and plough the proceeds into R&D, has persuaded President Trump to take the risk. Selling, Huang argued, would keep the American tech stack as the global standard – “just as the American dollar is the standard by which every country builds on” — while also funding US innovation.

In that light, the 15 per cent levy looks less like a compromise and more like a mechanism for sustaining the global spread of US AI hardware that simulataneously forces China to pay for the privilege. It is export controls reimagined as revenue stream, a blend of market capture and fiscal extraction. Is the cash worth it? Washington seems to think so. The licences are Washington’s to grant, the terms are Washington’s to set, and the architecture of China’s AI sector (theoretically) still depends on American silicon.

While some might argue Washington could go further by delivering compute as a cloud service rather than shipping physical chips (therefore keeping the revenue and dependency while retaining full control over usage), in practice China’s data sovereignty laws and mistrust of foreign-managed infrastructure make large-scale U.S. cloud provision far outside the realms of realpolitik.

The levy strategy is a more immediately viable, if looser, form of control — one that still aligns with Sacks’s framework: market share first, enduring advantage second, permanence third. An arrangement balances economic gain with strategic leverage.

As Lutnick himself said, “You want to sell the Chinese enough that their developers get addicted to the American technology stack, that’s the thinking.”

AI is not just another export category; it is critical infrastructure to be owned and defended — the foundation on which empires are attained and retained. Whether the setting is Beijing, Brussels or Bangalore, the same logic applies: it’s better for a data centre to run on a slightly neutered American chip than on a Chinese one.

In this sense, the levy is a way of denying the emergence of a rival castle. By keeping China’s AI sector reliant on U.S. silicon, Washington ensures that Beijing’s walls rest on American-laid foundations. Dependency isn’t a by-product here — it’s the point, a deliberate design to prevent the rise of an independent power centre.