For small business owners like Chen Wensheng in China’s Hainan province, investing in artificial intelligence stocks once seemed like a sure bet. Encouraged by government backing for national technology champions, Chen and his friends poured their savings into rising companies such as humanoid robot maker Unitree. Now, a massive sell-off has turned enthusiasm into anxiety. Beijing-based fund manager Dong Baozhen described the rally as a “speculative crazy bull”, adding to concerns that the bubble is bursting. In less than four months, China’s main technology stock index, the STAR50, plunged by 40%, wiping out an estimated $480 billion in market value.
The downturn raises a question that extends beyond China: can the enormous sums invested in artificial intelligence generate the profits investors expect? Developments in the United States suggest that the challenge is not confined to Chinese markets. Anthropic, the American developer of the Claude AI models, has drawn attention for its reported multibillion-dollar losses and substantial spending requirements, highlighting the costs of competing in an industry where technological advances demand enormous investment.
Developing advanced AI models requires expensive computing infrastructure, specialised chips and highly skilled employees. Companies may accept substantial losses while pursuing growth, expecting future revenues to justify their spending. But when costs rise faster than income, investors must decide how long they are willing to finance that gap.
China’s market correction has been particularly severe. According to Goldman Sachs, AI companies account for 42% of the overall value of Chinese stock markets, leaving individual investors heavily exposed to the sell-off. Even after its decline, the STAR50 reportedly trades at around 100 times annual earnings, down from a peak of 250 times. The technology-heavy Nasdaq-100, by comparison, trades at roughly 30 times earnings.
Individual stocks illustrate the speed of the reversal. Unitree shares have reportedly fallen by half since August, despite its stock launch being 8,000 times oversubscribed. Chipmakers MetaX Integrated Circuits and Moore Threads Technology, which surged between 400% and 700% during their market debuts, have surrendered roughly half their market value. Memory maker CXMT has also suffered a sharp decline, while optical equipment manufacturers Zhongji Innolight and Eoptolink Technology have faced steep losses amid concerns over possible US restrictions on Chinese hardware suppliers.
Behind the falling share prices lies a deeper problem: turning AI enthusiasm into commercial profits. Chinese AI start-up funding reportedly fell 67% year-on-year, from $18.4 billion in the first three quarters of 2025 to $6.1 billion over the corresponding period in 2026. Private-market valuations of prominent companies, including Zhipu AI and MiniMax, have also faced significant markdowns.
Customer adoption remains a major obstacle. Less than 12% of Chinese businesses have reportedly adopted AI services, well below the 40% to 50% rates some experts had anticipated. Without enough customers willing to pay for these products, companies may struggle to convert technological advances into reliable revenue.
The United States has attracted substantial investment, with AI start-ups reportedly raising $67 billion in early 2026. Yet strong funding does not guarantee profitability. Anthropic’s reported losses illustrate the tension between growing demand for AI and the expense of developing and operating advanced models. Investors may tolerate heavy spending while a company expands, but their confidence can weaken if revenue fails to catch up with costs or competitors offer comparable services more cheaply.
China’s technology giants are already adjusting their strategies. Baidu has reportedly cut staffing in its AI division, Alibaba Cloud has paused some expansion plans, and Tencent has shifted engineers from experimental projects towards established revenue sources such as gaming and advertising. The moves suggest a growing emphasis on commercially viable products rather than technological ambition alone.
Beijing has also issued guidelines warning against blind speculation and reckless technology projects. Geopolitical tensions add uncertainty: the government recently ordered Meta to reverse its reported $2 billion acquisition of Chinese-founded start-up Manus, while US export controls continue to restrict Chinese access to advanced Nvidia processors.
These developments underscore a shared challenge for China and Silicon Valley. AI companies must demonstrate that customers will pay enough for their products to cover development and operating costs. High valuations, impressive technology and substantial investment cannot guarantee sustainable returns.
For Chinese investors, the immediate consequences are financial losses and a search for safer assets. For AI developers, the longer-term challenge is to establish viable business models before investor patience runs out. Hardware suppliers may face increasingly aggressive price competition, while model developers must balance expensive innovation with commercial discipline.
As Shanghai fund manager Wang Zhuo observed, Chinese AI suppliers will eventually compete on price like traditional manufacturers. The same principle applies to Silicon Valley: technological leadership may provide an advantage, but it cannot guarantee lasting profitability.
The AI industry is unlikely to disappear as valuations fall and companies report heavy losses. The technology may continue transforming industries and creating new markets. But China’s stock-market correction and Anthropic’s financial pressures offer a clear warning: the transition from technological promise to commercial success is neither automatic nor immediate. The ultimate test is not how impressive an AI system appears, but whether the business behind it can generate sustainable returns.
Before publication: Verify the market figures and Anthropic’s reported losses against their original sources, including the relevant financial periods. Distinguish between projected operating losses, realised net losses and cash expenditure.
