Free Chinese AI models could deflate the US tech bubble, Larry Elliott argues

Moonshot AI released Kimi K3 on July 16 and stopped taking new consumer subscriptions three days later. Demand had pushed its GPUs to the limit within 48 hours. The model competes with paid American products and costs nothing to use, and in the Guardian this week Larry Elliott built a case around that fact: the West spent two decades waiting for China to fail, and China spent them moving up the value chain.

The argument

Elliott sorts the last fifty years into three phases. China 1.0 assembled cheap goods. China 2.0 took solar panels, batteries and electric vehicles, and stopped importing German machine tools because it started making its own. China 3.0 is the race with the United States over artificial intelligence.

The market conclusion follows from there. Wall Street prices US technology companies on the assumption that America wins the AI build-out and keeps the returns. A Chinese lab that matches those models at a lower cost and gives them away weakens the second half of that assumption. Investors who accept the premise have to ask what the earnings look like when the product is free.

“If this strategy succeeds, the third China shock might not just disrupt the global economy, but also the global balance of power.”

— Edoardo Campanella, UniCredit, Quoted by Larry Elliott, The Guardian, 3 September 2026

Quote source: The Guardian, 3 September 2026

The case against

Elliott names the weakness himself. The Chinese model leans on investment and runs excess capacity, and wages and welfare at home are too thin to absorb what the factories produce. Growth depends on exporting the surplus, which is why China took criticism from every side at the last G20 finance ministers’ meeting for failing to lift domestic demand.

Kimi K3 supplies the second gap. Moonshot published the full weights, which sounds like the end of the American advantage until you price the hardware. Running the largest open model yourself takes data centre equipment, and the free tier that overwhelmed Moonshot’s own GPUs shows what serving it costs. Compute constrains both sides of the Pacific, and Washington restricts China’s access to the best of it.

The tests that would settle it

Two things worth watching:

• whether Chinese labs keep frontier models on a free tier once the compute bill lands, or move to paid access the way US labs did
• whether US tariffs on allies push European and Asian buyers toward Chinese models, the risk Elliott flags in Washington’s trade policy

For readers holding technology exposure, the question is narrower than the geopolitics. If a capable model becomes a commodity, the money moves to whoever owns the scarce input, which today is chips, power and data centre space.

Elliott’s last point aims at Britain rather than at markets: China picked its industries, used subsidies, tariffs and state investment without apology, and held the plan for fifty years. He leaves open whether that is worth copying. Anyone tempted by the idea should read the paragraph about wages and welfare first.

This article is for informational purposes only and does not constitute investment advice.