A worker stands at the booth for Chinese AI startup Moonshot's Kimi K3 during the World AI Conference in Shanghai, on July 17.Ng Han Guan/The Associated Press
Jacob Cooke is the chief executive of WPIC Marketing + Technologies, a Beijing-based firm that advises global brands in Asian markets.
There was another Sputnik moment with the recent release of Kimi K3, a cutting edge artificial intelligence model from the Chinese startup Moonshot AI.
Rivalling the performance of Anthropic and OpenAI’s top models, Kimi K3 sent Silicon Valley into a frenzy, prompting allegations that Chinese firms are using American models to train their own.
But those allegations, unproven as of now, are not the most important part of the story. Neither is the mere fact that Chinese AI is approaching parity with American AI.
The real story is that China is putting forth a distinct framework for how AI should be diffused across society. Generally speaking, AI in China is open, low-cost, and application-focused. Meanwhile, Chinese firms and government agencies are taking proactive steps to address potential social harms of AI.
It’s a framework with significant merit.
The Kimi ecosystem is open weight. That means any company or developer can download Kimi, run it on local servers, and apply it to their specific needs. The same is true for models from Chinese players such as Alibaba, DeepSeek, Z.ai, MiniMax, and others. Firms monetize through adjacent services, like hosting or enterprise setup, but broadly speaking the technology is accessible.
On the other hand, America’s frontier AI models are closed. Access is metered, and firms are shelling out enormous token fees to Anthropic, OpenAI, and Google.
Chinese models are generally more efficient than their American counterparts. U.S. limits on the sales of cutting edge Nvidia chips to China were intended to contain its AI development. Instead, these export restrictions forced Chinese labs to optimize for efficiency. Now, many Chinese models can achieve high performance at a fraction of the compute cost of American models.
Meanwhile, to meet skyrocketing token demand, the U.S. is building out data centres at a historic scale. That is straining electricity grids, inspiring local opposition, and damaging the environment.
Another difference centres on the pursuit of artificial general intelligence, or AGI. Silicon Valley firms are pouring billions into achieving superintelligence. Sky-high capital expenditure has rendered American AI models largely unprofitable, despite high fees. These firms’ enormous valuations are based on a bet that achieving AGI in the future would pay enormous dividends. But even if that bet pays off, the winnings would accrue to a narrow few, according to Adam Tooze, an economic historian and a professor at Columbia University.
Chinese AI model takes U.S. tech industry by surprise with abilities rivaling Claude and ChatGPT
Chinese firms are not necessarily focused on creating AGI, but instead on developing capable models that can be diffused across the economy through applications. The lower cost of Chinese models, combined with their open weight nature, allows other companies to build customized apps that run on local servers. As Alibaba’s chairman Joseph Tsai said last year, open source allows businesses, individuals, and entrepreneurs to innovate with AI, rather than AI staying concentrated in a few dominant players.
Here’s how that can work in practice. My firm helps overseas consumer brands enter the market in China. We’ve used Kimi and Alibaba’s Qwen to build our own applications that run on our own servers. These applications help manage e-commerce stores, optimize delivery logistics, and execute marketing campaigns. To do this with American AI’s proprietary models would have been prohibitively expensive for many companies. Businesses across the economy in China are embracing AI in this fashion, and the efficiency gains are more widely enjoyed.
The divergence also extends to public sentiment. According to Stanford’s 2026 AI Index, 83 per cent of people in China see AI as more beneficial than harmful, versus just 39 per cent of Americans. Much of the public conversation in China focuses not on job losses but on concrete gains. For example, AI-powered robots could ease staffing shortages in the China health care sector in the next few years.
That positive sentiment likely stems in part from the Chinese public and private sectors getting ahead of AI’s potential externalities. That was made clear at the World AI Conference in Shanghai, which ran from July 17 to 20, when China’s President Xi Jinping insisted that AI must serve humanity and called for precise oversight of any negative consequences.
In China, new rules this month banned AI “companion” services for minors. Chinese courts have ruled that a company cannot dismiss workers to replace them with AI. The e-commerce giant JD.com has pledged to retrain its staff even as the company pushes to automate warehouses with AI.
China’s framework – AI as accessible, low-cost, and focused on positive applications, with guardrails around employment and other negative externalities – is perhaps best suited to meeting this moment.