It's hard not to feel a sense of déjà vu. Last year, DeepSeek shocked investors by proving China could build a world-class AI model despite US efforts to block its access to the most advanced chips. This month, it happened again.
Chinese startup Moonshot AI unveiled Kimi K3, an open-weight model that briefly topped a leading independent coding leaderboard – Arena.ai's Frontend Code Arena – becoming the first Chinese model to perform better than Anthropic's Claude. On broader intelligence rankings, it now sits just behind the US frontier, at around third place – still a remarkable showing for a freely downloadable model.
That rattled US tech stocks because it challenged one of the AI boom's biggest assumptions: that America's lead in chips, spending, and computing power would naturally translate into a durable advantage in AI – and ultimately, into lasting profits. The recent slide in China chip and AI stocks shows how quickly sentiment can shift when that assumption is questioned.
Two very different games
The two countries are playing very different games. America's AI leaders are mostly keeping their best models behind closed doors, charging customers for access and tying the large language models (LLMs) closely to giant cloud platforms. China's firms, meanwhile, have taken the opposite approach, pushing increasingly capable models into the market at much lower prices and generally letting anyone run them.
This matters a lot now that OpenAI and Anthropic are both moving toward public listings – potentially giving everyday investors their first real shot at owning two of America's leading AI developers. But before putting a price tag on either company, there's a more fundamental question to answer: what does winning the AI race actually mean?
In other words, if AI models become cheaper, easier to copy, and easier to switch between, does owning the best model create a lasting moat at all? Or do the profits ultimately end up somewhere else – with the cloud providers, the chipmakers, or the businesses putting AI to work?
What it means for investors
For everyday investors, the key takeaway is that the smartest AI model might not be the smartest investment. The history of technology is littered with companies that had the best product but failed to capture the profits – think of Netscape in the early internet era or BlackBerry in smartphones.
Instead, investors should watch who keeps getting paid after the smartest model changes. That could be the cloud infrastructure providers like Amazon Web Services or Microsoft Azure, which host the computing power that AI models need. It could be the chipmakers like Nvidia, which make the processors that train and run these models. Or it could be the companies that use AI to improve their own products and services, from software to healthcare to finance.
The recent slide in Asian chip stocks on AI funding doubts and China competition shows how sensitive the sector is to these dynamics. And the Singapore central bank's warning about AI investment risk underscores that even regulators are watching closely.
Ultimately, the AI race is not just about who builds the smartest model. It's about who builds the most durable business around it. For investors, that distinction is everything.
