iQiyi's chief executive Gong Yu delivers a speech
iQiyi's chief executive Gong Yu delivers a speech during the iQIYI World Conference in Beijing on April 20, 2026. Wang Zhao/AFP

KEY POINTS

  • While cheaper open-weight models from Chinese firms like Moonshot AI and Z.ai are rapidly capturing global user share, surging inference and training costs continue to outpace revenue growth.
  • The popularity of Chinese open models has exposed how compute-constrained many Chinese tech companies are.
  • Like their American counterparts Anthropic and OpenAI, Chinese AI companies spend more money than they make in revenue.

By now, the headline has become familiar: a Chinese AI startup undercuts OpenAI or Anthropic by releasing a vastly cheaper, almost-as-good-or-better open-weight frontier model and policymakers in the U.S. government fret that the country is falling behind China in the global AI race. Most recently, the unveiling of Moonshot AI's Kimi K3 sparked accusations of intellectual property theft from the Trump administration and Washington has been consumed with a debate about banning Chinese AI.

With AI costs rising sharply for businesses, many companies such as Mozilla and Coinbase are switching from expensive, closed-weight AI models created by American companies such as OpenAI and Anthropic to China's cheaper open-weight models. According to the Associated Press, the five most popular models on OpenRouter—a platform that tracks AI model usage—were all developed by Chinese companies.

Yet popularity and profitability are proving to be two very different things.

Chinese companies are now discovering what OpenAI learned years ago with the release of ChatGPT in 2022: Widespread adoption does not automatically produce a viable business. In fact, serving more users can mean incurring higher inference costs and greater demand for computing power.

After the release of Kimi K3 proved popular, Moonshot AI had to pause subscriptions after users complained of slow response times since the company did not have enough compute to run its models reported the South China Morning Post. Moonshot AI was not alone in this—Chinese AI firm Zhipu's similarly slow user response times in February caused the company's stock price to fall 23%, a loss of $9 billion in market capitalization.

Winning the battle for users may make it harder for Chinese AI companies to achieve profitability, if recent financial results are anything to judge by.

Z.ai's revenue surged 132% to $107 million in 2025, but net loss jumped 60% to $694 million as the company ramps up spending on compute and model training. Alibaba similarly reported just $1.3 billion in revenue from AI-related products in the first quarter of 2026 compared to $55 billion in AI infrastructure spending by the end of next year.

"There isn't an A.I. company with a sustainable business model right now," Richard Lin, a vice president at the Silicon Valley company, Datastrato told the New York Times. "It's not a healthy industry."