Why America cannot simultaneously protect its AI companies, reward investors, and keep its businesses globally competitive.

The Wrong Question

The debate over artificial intelligence is often framed as a race: whoever builds the best models first will dominate the future. That framing made sense when there was a widely perceived technological lead. Today, however, the dynamics look different. Frontier capabilities are increasingly diffuse. The question is no longer simply who can build the best models, but who can provide them most cheaply.

Much of today’s AI policy assumes that maintaining a technological lead is the central objective. But that assumption depends on a belief that the lead can be sustained long enough to justify extraordinary costs.

There is increasing reason to question that premise.

The Superintelligence Assumption

Many investment theses and policy proposals rest on the expectation that today’s frontier models are merely stepping stones toward systems that are dramatically more capable than humans—a level often described as superintelligence. If that future arrives quickly and if one country or one company reaches it first, today’s economics matter much less. Temporary losses would be justified by overwhelming future gains.

But this is a remarkably strong assumption.

It assumes not only that superintelligence is imminent, but that it produces durable economic advantages, that those advantages cannot be replicated, and that technological leadership can be maintained despite global competition.

None of those outcomes are guaranteed.

It is entirely plausible that AI continues to improve rapidly without producing a decisive discontinuity. Models may become steadily better while remaining broadly substitutable. Performance gaps may narrow rather than widen. Frontier capabilities may spread faster than monopolies can form.

If that is the world we are entering, then AI begins to resemble other infrastructure technologies: enormously valuable, but ultimately defined by price, efficiency, and scale rather than exclusive capability.

The Margin Advantage

This is where the strategic asymmetry between the United States and China becomes difficult to escape.

American technology companies are largely expected to generate venture-scale returns. Their investors funded them with the expectation of substantial margins, not merely sustainable businesses. China, by contrast, has repeatedly shown a willingness to compete at much thinner margins. If frontier AI models can be profitably offered at 3% margins—or even lower—that fundamentally changes the economics of the industry.

Once performance converges, price becomes the deciding factor.

The company willing to operate closest to cost effectively determines the global market price.

The Limits of Protectionism

The standard American response is to reach for restrictions. Ban Chinese models. Prohibit their use by American companies. Tighten export controls. Limit access to infrastructure.

But these policies only solve a domestic problem. They do not solve the global one.

The United States can prevent its own firms from using Chinese models. It cannot prevent Brazil, Indonesia, Nigeria, India, Germany, or most of the world from doing so.

If Chinese frontier models are substantially cheaper, businesses everywhere else will naturally adopt them. Software companies will lower operating costs. Manufacturers will automate more aggressively. Startups will gain access to state-of-the-art AI at prices American providers cannot profitably match.

Those cost advantages compound.

A logistics company in Singapore running inference at half the price of its American competitor can optimize more routes. A biotech startup in Europe can run more experiments. A design firm in Mexico can generate more prototypes. Every industry that relies on AI becomes slightly more productive, and those incremental gains accumulate into a broader competitive advantage.

Ironically, restrictions intended to protect American AI companies may weaken American businesses more generally.

The Impossible Tradeoff

The result is an uncomfortable policy dilemma.

If the United States permits unrestricted competition, American AI companies face enormous pricing pressure. Investors who funded these companies expecting venture-scale returns may instead discover that frontier models have become commodities. Capital that once flowed eagerly into AI infrastructure becomes harder to justify.

If the United States instead protects domestic AI providers—through restrictions, procurement preferences, or other forms of insulation—it may preserve returns for investors.

But those returns come at a cost.

American businesses end up paying more for the same capability than competitors abroad. Every additional dollar spent on AI is a dollar not invested in hiring, expansion, or research.

In one world, investors lose.

In the other, the broader economy loses.

Neither outcome is particularly attractive.

Whose Interests Are Being Protected?

This tension reflects a deeper contradiction in industrial policy.

Governments often speak as though the interests of domestic AI companies and domestic businesses are perfectly aligned.

In AI, they may not be.

An AI company benefits from selling expensive inference.

Every other company benefits from buying cheap inference.

The interests are opposite.

Historically, countries tolerated this conflict because technological leadership created temporary monopolies. Pharmaceutical companies benefit from patents. Semiconductor firms possess manufacturing advantages. Software companies enjoy network effects. High margins are sustained because substitutes are scarce.

AI may prove fundamentally different.

If multiple countries can produce frontier-quality models, and those models become increasingly interchangeable for most commercial applications, price competition becomes relentless. The value migrates away from the model itself and toward the industries that use it.

In such a market, the country willing to operate closest to cost sets the global price.

If China is prepared to compete at 3% margins while American firms require substantially higher returns, the equilibrium price is determined by China—not by Silicon Valley.

The Unwinnable Logic

This leaves the United States with no clean strategic response.

Blocking Chinese models domestically protects AI producers while making downstream industries less competitive.

Allowing open competition benefits the wider economy but compresses the profits of the companies that built the frontier in the first place.

The policy objective cannot simultaneously maximize investor returns for AI developers and minimize AI costs for every other business. Those goals become mutually incompatible once technological parity exists and price competition dominates.

If the dream of a decisive superintelligence breakthrough proves overstated—or simply arrives much later than expected—this tension becomes even more acute. The economics of AI will be determined less by who possesses the smartest model than by who can supply intelligence most cheaply to the rest of the world.

That is the unwinnable logic of AI.