Voice & Audio OpenAI Opens GPT-Live-1.1 to the API and Prices Silence at 2 Cents a Minute Coding Assistants Replit Pitted 32 Claude Agents Against NumPy Tuning: 18.9x Speedup While GPT-6.2 Swarms Collapsed at Scale Security GAO Covert Tests: All Four Top Chatbots Produced Viable Attack Plans in Under 12 Hours Once Guardrails Fell Meta Meta Puts Its 2030 Net-Zero Target Under Review as AI Data Centers Push Net Emissions Up 45 Percent Business Daniel Ek Is Backing Europa Fusion, a Stealth UK Startup Led by a Former DeepMind Researcher Betting on Pulsed Power Coding Assistants Stripe Tested 14 Coding Agents on 69 Payment Scenarios: 78% of Criteria Passed, but Only 29% of Runs Met Every Requirement Security A Transcription Firm Used by Police and Prosecutors Left Interview Transcripts and Body-Cam Footage on an Open S3 Bucket News UN Killer-Robot Talks End With a 2028 Deadline and No Binding Rules. About 100 States Wanted a Treaty; the US, Russia and Israel Blocked It Voice & Audio OpenAI Opens GPT-Live-1.1 to the API and Prices Silence at 2 Cents a Minute Coding Assistants Replit Pitted 32 Claude Agents Against NumPy Tuning: 18.9x Speedup While GPT-6.2 Swarms Collapsed at Scale Security GAO Covert Tests: All Four Top Chatbots Produced Viable Attack Plans in Under 12 Hours Once Guardrails Fell Meta Meta Puts Its 2030 Net-Zero Target Under Review as AI Data Centers Push Net Emissions Up 45 Percent Business Daniel Ek Is Backing Europa Fusion, a Stealth UK Startup Led by a Former DeepMind Researcher Betting on Pulsed Power Coding Assistants Stripe Tested 14 Coding Agents on 69 Payment Scenarios: 78% of Criteria Passed, but Only 29% of Runs Met Every Requirement Security A Transcription Firm Used by Police and Prosecutors Left Interview Transcripts and Body-Cam Footage on an Open S3 Bucket News UN Killer-Robot Talks End With a 2028 Deadline and No Binding Rules. About 100 States Wanted a Treaty; the US, Russia and Israel Blocked It

China Is Doubling AI Capex to $139 Billion While Its Labs Earn Less Than 8 Percent of It Back

China Is Doubling AI Capex to $139 Billion While Its Labs Earn Less Than 8 Percent of It Back

A Rhodium Group analysis, reported by The New York Times, finds 13 listed Chinese tech firms will lift AI infrastructure capex 103 percent this year to 932 billion yuan ($139 billion), while AI lab revenue covers less than 8 percent of that spend. China's AI build-out sits at 15-20 percent of the US level, and financing — not chip curbs — may be the binding constraint.

The money question behind China's AI race just got its clearest answer yet. A Rhodium Group analysis of China's AI financing, reported Friday by The New York Times, projects that 13 listed Chinese technology companies will spend 932 billion yuan — about $139 billion — on AI infrastructure this year, a 103 percent jump, with spending forecast to exceed 1.2 trillion yuan ($193 billion) in 2027.

The revenue side is much thinner. Across those companies, AI lab revenue covers less than 8 percent of this year's capex, Rhodium estimates. The annual recurring revenue of all of China's AI models combined is roughly $10.7 billion — about a tenth of what OpenAI and Anthropic report between them. "When you look at their numbers, it looks like they're running on a treadmill that's speeding up and getting harder to run," Kyle Chan, a fellow at the Brookings Institution, told the Times.

Rhodium partner Logan Wright, who co-wrote the report, argues the real bottleneck is not Washington's export curbs but money itself: financing constraints, he told the Times, "will be probably more decisive" in the long run. Rhodium puts China's AI build-out at just 15 to 20 percent of the US level. Alibaba, Baidu and Tencent — the backers behind several of China's top labs — now carry negative cash flows, and the STAR 50 index has fallen 43 percent since late July.

The open-source bet has squeezed margins further. When Z.ai released GLM-5.3 Flash in August it halved prices for two weeks; by November, OpenRouter data cited by the Times showed roughly 20 rival providers charging less than Z.ai does for its own model. Compute itself is scarce: on Goofish, a Chinese second-hand marketplace, sellers have listed Nvidia's DGX B300 server for nearly $3 million — about eight times the US price, the paper found.

Even DeepSeek, the best-capitalized Chinese lab, is rationing ambition. In a leaked transcript of an investor call cited by the Times, founder Liang Wenfeng said Huawei's chips are only about two years behind Nvidia's — but it would take four Huawei 950 chips to match one Nvidia GB300. "We simply can't afford to train the largest models available right now," he said, according to the transcript. DeepSeek did not respond to the Times' request for comment. The same report notes DeepSeek has raised more than $20 billion, including a $15 billion round backed by Tencent and CATL — a round whose corporate registration was confirmed this week.

For scale, Stanford's AI Index counts $12.4 billion of private AI investment in China in 2025 against $285.9 billion in the US — and that likely understates China's total, because state guidance funds also pour in money. Foreign capital has largely dried up, partly because Washington bars US venture investment in some Chinese technologies.

None of this means China's AI effort stalls. Its researchers keep finding cheaper ways to train and run models, and the state can mobilize resources few markets can match. But as Han Lin of the Asia Group told the Times, frontier technology also needs risk capital that can absorb failure — and that is exactly the resource China's accelerating treadmill is short of.

Comments (0)

Log in to join the discussion

Log In

No comments yet