China Giving Away Frontier AI Models Poses a Problem For OpenAI, Anthropic

Major American AI companies like Anthropic, Google, and OpenAI are scrambling to figure out a coherent response to Chinese developer Moonshot’s Kimi K3 AI model. Not only is it almost as capable as the best Western-designed AI, but it’s an open-weight model, meaning anyone can run it on their own hardware without using Moonshot’s cloud platform.

Despite claims that Kimi K3 was trained using distillation techniques, Microsoft and others have begun trialing the AI model on their services. This eats into the value proposition of the exclusive contracts offered by major AI developers. Considering the industry was already facing a crisis of value as token costs spiked, the openness of Chinese models could pull the rug out from under the entire Western AI market.

Open weight models can run on any sufficiently powerful hardware (you still need top-tier GPUs for a near-3-trillion-parameter model, however). And the “open” nature means outside developers can analyze the AI and customize how it works. This could allow emerging AI developers to catch up quickly to what companies like Anthropic, Google, and OpenAI have spent tens of billions of dollars building and honing.

This might seem counterintuitive for China. If it can win the AI race, why open itself up to increased competition? Likely, because the move is designed to destabilize Western AI businesses as much as grow China’s. Although Kimi K3 is likely to generate a lot of buzz and attention, global AI users may shift away from the dominance of Anthropic and others.

A presentation slide generated by Kimi K3. (Everyone’s going for the Claude look, these days.)
Credit: Moonshot AI

This opens up the marketplace in ways it wasn’t before, making it more competitive and, arguably, unaffordable for many Western developers. It’s not going to put OpenAI out of business, but it absolutely threatens its long-term future and could make its profit goals for the next few years utterly impossible to achieve.

The increased adoption of Chinese models also makes it easier to develop domestic hardware to power them, especially if the US shuts down access for American companies. China-aligned countries and companies that want to run more affordable AI may turn to Chinese models now, but potentially hardware in the future if access to top chips remains restricted.

It’s not clear how this plays out long-term, but many US companies have called for model access not to be restricted, even as the White House considers it for national security reasons. Doing so might benefit the big players in the short term, but either way, Moonshot and Chinese developers in general have shown the major firms that the next few years won’t be the easy ride it’s been so far. Investment may dry up, and interest may go elsewhere.