China's brutal AI economics hold lessons for US

For Chinese companies, extracting greater performance from models while meeting growing demand increasingly rests on bulking up in-house computing capacity and expertise. Yet the sums remain tiny when compared with the U.S. infrastructure splurge. Alibaba racked up $10 billion in capital expenditure in the second quarter, up 75% from a year ago and dragging the e-commerce giant's cash flow into the red. But the company has a cloud computing business to cross-sell APIs, storage, higher margin IT services and even excess computing power to its already sizable customer base. That puts boss Eddie Wu in an enviable position: Alibaba says it expects a three-year payback period on its AI capex. Standalone AI labs with fewer revenue sources will probably struggle to match that. For all the frantic competition, no Chinese AI company has established a durable advantage. AIibaba, DeepSeek, Z.AI and Moonshot constantly trade places atop AI leader boards. The vast [majority](https://global.chinadaily.com.cn/a/202507/28/WS6887269fa310c26fd717c174.html) of model developers will probably not survive, despite Beijing's growing [warnings](https://www.cac.gov.cn/2026-08/21/c_1789061449039952.htm) against destructive competition and disorderly expansion. Investors currently favour faster-growing Z.AI and MiniMax over incumbents Alibaba and Tencent, which trade on lower valuation multiples. The bet may be that the survivors can carve out higher-margin niches beyond coding. Either way, there are lessons for OpenAI and Anthropic. Slowing the pace of development would force their existing models to compete on price and efficiency with bargain-basement alternatives. The biggest lesson from China's AI industry is that once the spiral of relentless competition starts, there are rarely winners. Despite these handicaps, China's AI hopefuls are keeping pace with their stateside rivals. Industry analysts [estimate](https://epoch.ai/data-insights/us-vs-china-eci) the best Chinese models are on average about seven months behind the most sophisticated releases from OpenAI and Anthropic. Even so, the industry's economics are very different. Chinese companies compete mostly on efficiency and volume in a brutal race to the bottom. As of early August, DeepSeek's latest V4-Flash model was more than 100 times cheaper to run than Anthropic's Claude Fable 5, according to research firm Artificial Analysis. This and similarly low-cost offerings from Alibaba 9988.HK and others appeal to users seeking a value-for-money alternative. The cut-price approach means upstarts are still bleeding cash. In the first half of 2026, the $40 billion Hong Kong-listed Z.AI 2513.HK, among China's top model developers, spent 2.1 billion yuan ($313 million) on research and development despite earning less than half that sum in revenue. Rival $10 billion MiniMax 0100.HK raked in $117 million in sales but sunk more than two-and-a-half times that figure in R&D over the same period. Skimpy revenue reflects a fundamentally difference in strategy. Chinese AI labs develop open-weight models that are free for anyone to download onto private servers, modify and use themselves. Alibaba in August announced that its models accumulated more than 3 billion global downloads in the past six months, surpassing Meta Platforms META.O and Alphabet's GOOGL.O free offerings. Though giving away the product may seem foolish, it is simpler and often cheaper for enterprises to use those models remotely by paying for an application programming interface (API). Transactions on OpenRouter, a popular marketplace for APIs, show that DeepSeek, Z.AI, Tencent 0700.HK and Alibaba held a combined 45% market share in the week to September 14. Chinese AI labs are also developing other sources of revenue. MiniMax, which makes roughly 60% of its sales from overseas, generates subscriptions from consumer apps. Moonshot, the owner of the Kimi models, is demanding revenue-sharing agreements with major users of its models - a practice that is set to become more common. So long as revenue grows faster than the costs of training new models, the payback period on R&D should shorten. The other promising trend is in inference, where trained models process and respond to prompts. China's obsession with efficiency and scale is starting to pay off as serving APIs - the biggest drag on gross profit - is getting cheaper. Z.AI says per-unit inference costs are down 80% since the start of the year, lifting gross margins at its API division to 25% in the six months to June. That's a notable improvement from the negative 0.4% a year earlier. Separately, MiniMax credits "improving infrastructure efficiency" for a 5.8 percentage point increase in overall gross margins over the same period, to just under 18%. These levels of profitability are still puny, though. Anthropic's gross margin in the second quarter was 44%, PitchBook [estimated](https://pitchbook.com/news/articles/anthropics-gross-margin-ipo). Even that falls far short of the 70%-plus level earned by many of the cloud software specialists that AI is meant to replace. What would an artificial intelligence industry without breakneck progress look like? Anthropic CEO Dario Amodei's [call](https://darioamodei.com/post/we-must-pace-the-frontier) for AI labs to slow development and erect guardrails to manage apocalyptic threats from rogue chatbots and agents has garnered endorsements from rival bosses Sam Altman and Elon Musk. But scaling back spending on sophisticated frontier models would also upend the industry's business plans and financial returns. China offers clues for what may come next. At first glance, the trajectories for machine intelligence in the world's two largest economies could not be more different. Until a few days ago, labs like OpenAI and Anthropic appeared locked in a cash-burning battle to build ever more powerful and expensive AI models. In doing so they had the support of cutting-edge Nvidia NVDA.O semiconductors, vast new data centres built by tech giants such as Microsoft MSFT.O and Amazon AMZN.O, and a fire hose of Silicon Valley capital. Competitors in the People's Republic, meanwhile, have been hobbled by Washington's export controls on advanced chips and equipment and shrinking investment from overseas. Between 2023 and mid-2026, when venture capital funding into AI topped $380 billion in the United States, Chinese upstarts received barely a tenth of that sum, [estimates](https://www.bcg.com/publications/2026/us-and-china-ai-strategy-causing-global-ai-divide) BCG. U.S. tech titans are forecast to splash out nearly $1 trillion on chips, data centres and other capital expenditures in 2027, according to Moody's analysts; their counterparts in China are on track to spend just $165 billion. Local champion DeepSeek is [looking](https://www.bloomberg.com/news/articles/2026-08-06/deepseek-resumes-8-billion-round-with-monolith-in-the-running) for roughly $8 billion, a fraction of the $122 billion OpenAI secured earlier this year.

China's brutal AI economics hold lessons for US

Lessons for OpenAI and Anthropic

Open-weight models and thin margins

Revenue growth and inference are improving

China's AI race is cheaper, but still brutal

Related News

American Water’s Illinois unit buys Venice, Illinois wastewater collection system for $770,000

Opaleye Management reports USD 1.07 million Alpha Cognition common share purchase

First Bancorp director Katharine Caldwell Nevin acquires $14,833.13 in common shares

Flexsteel director William S. Creekmuir reports disposal of USD 321,389.03 in common shares

Hinge Health director Elliott Robinson reports disposal of common shares worth $13.86 million