China's tech giants double AI spend to $139B. Labs earn under 8%
Beijing's biggest platforms are pouring a record $139bn into AI while their own labs return less than 8% of it, and a chip squeeze is making the math worse.

China's largest technology companies will double their AI spending this year to $139bn, while the AI labs they fund bring in less than 8% of that outlay, according to Rhodium Group figures reported Friday by The New York Times' Sui-Lee Wee and Li You. For executives and boards, it is a live case study in what happens when capital outruns monetization and the hardware needed to deploy it stays scarce.
China's biggest technology companies are set to double their spending on artificial intelligence this year, pushing the total to $139bn, while the AI labs they bankroll bring in less than 8% of that money back. The figures come from a Rhodium Group report on China's AI economy, reported Friday by The New York Times' Sui-Lee Wee and Li You. The headline number is the doubling. The real story is the gap underneath it: China's AI industry now has a money problem stacked directly on top of its chip problem.
Run the arithmetic and the shape of the problem snaps into focus. For every dollar China's largest platforms commit to AI, their own labs return less than eight cents. That is not a profit margin, it is a revenue ratio, and it describes an industry spending at a pace its customers have not yet matched. The spending itself is real and heavy: data centers, accelerators, power contracts, and the compensation required to keep elite researchers from walking out the door. The revenue is not, at least not at anything close to the same scale. When a company doubles its outlay while its own AI unit earns a fraction of it, the gap has to be funded from somewhere, and that somewhere is the profitable core business.
The chip shortage is what makes the doubling so expensive. China's access to the most advanced AI accelerators has been constrained by US export controls, which means its largest firms are buying fewer chips at higher prices, or routing around the shortage with domestically produced silicon and engineering workarounds. When supply is tight, the same capability costs more. A doubling of spend therefore does not buy a doubling of compute, which is one reason the return line looks so thin relative to the investment line. Historically, compute scarcity has a way of turning capital expenditure into a treadmill: you run faster and stay roughly in place.
There is a second, quieter reason the labs earn so little. China's AI market has typically been shaped by fierce price competition and a consumer-first dynamic, where models are given away or priced near zero to win users, and enterprise contracts that actually carry margin arrive later and slower. That pattern is familiar from earlier Chinese internet cycles, where scale in users preceded scale in revenue by years. The difference this time is the size of the bill. The platforms are not funding a lightweight app layer; they are funding power-hungry infrastructure with depreciation schedules that start ticking immediately, whether or not a paying customer shows up.
For the executives running these companies, the strategic bind is genuine. Slow the spending and you risk losing the model race to a domestic rival or to a US lab that is already monetizing aggressively. Keep the spending and you are asking shareholders to underwrite a bet whose payoff is, by the report's own numbers, less than 8% of the cost so far. Boards in this position tend to reach for the same levers: pushing AI units toward enterprise and government contracts, bundling AI into existing paid products where the customer already exists, and quietly capping the most speculative research lines. None of that is a cure. It is triage.
The second-order effects reach well beyond the platforms themselves. If compute is the scarce input, then whoever controls access to chips, power, and data center capacity holds the real leverage in China's AI stack, not whoever ships the flashiest model. That shifts bargaining power toward infrastructure suppliers and domestic chipmakers, and it makes energy contracts a strategic asset rather than a line item. It also raises the odds of consolidation among smaller labs that cannot absorb rising compute costs, and of a talent market where a handful of well-funded players set the price for everyone else.
The uncomfortable parallel for Western operators is that the spending-versus-revenue gap is not uniquely Chinese. US hyperscalers have faced versions of the same question from investors, though they generally have larger, more mature cloud businesses to absorb the cost. What makes China's case sharper is the combination: a doubling of spend, sub-8% returns from the labs themselves, and a hardware shortage that inflates the price of every step forward. The number to watch next is not the $139bn. It is whether that 8% moves, and how fast, before the patience funding it runs out.
This story's Key Insights and Take-aways are locked.
Create a free account to unlock Executive Actions for one credit.
Register to UnlockAlways free for Executives Club members. Join the Club
More in Technology
Anthropic Mathematician Claims Claude Helped Prove Riemann Hypothesis in Six Weeks
Levent Alpöge says the AI tool helped him develop a solution to one of math's most important unsolved problems in about six weeks, a claim that, if verified, would reshape AI's role in research.
Anthropic mathematician claims Claude helped crack the Riemann hypothesis
If the proof holds, AI has toppled math's most famous open problem; if it does not, it is a case study in announcing breakthroughs before anyone verifies them.
AI's $4 Trillion Wipeout Erases Nasdaq's Gains for the Year
A brutal week took roughly $4 trillion off the AI sector and left the Nasdaq flat for the year, and the damage ran the full length of the supply chain.


