How Elon Musk Sees China's AI

The Economist just released its 85-minute interview with Elon Musk — his first long one since the SpaceX IPO. About ten minutes of it is on China's AI, and he doesn't hedge.

The line that matters: China will probably be the leader in AI. Not "might be." Not "worth watching." It's the conclusion he reaches after working through the constraints.

He starts by conceding ground

Asked about Kimi K3, released last week and reportedly almost as good as Fable, he says realistically Fable is still clearly the smartest model — but adds that the Kimi model is getting quite close. There are, he says, several very competent Chinese AI companies.

The actual argument

Then comes the deduction. Chinese AI companies got here with relatively little compute. So: what if they had a lot? "There's a good chance that they would be the leaders. And at some point they probably will have a lot of compute. So they will be the leaders."

Notice the structure. He turns "China leads" from a position into a deduction — the premise is compute, and compute is just a matter of time.

The constraints: electricity and chips

He reduces AI to two things: electricity and AI chips.

Then the number most people haven't internalized: China already produces more electricity than the United States, Europe and India combined. And it's still climbing — his estimate is China eventually reaches four times US electricity production, roughly proportional to population.

The asymmetry

The most interesting part is an asymmetry. Outside China, the binding constraint is electricity — AI chips are now being manufactured faster than new power comes online, and the real bottleneck is power and cooling, not silicon. (He also debunks a popular claim in passing: AI's water usage is negligible. It's the electrical demand that's high.)

Inside China, the constraint is chips, because the US banned exports of the latest AI chips. China is, in his words, somewhat chip starved.

But that asymmetry is being flattened by two things: China is getting much more efficient with the chips it has — Kimi K3's efficiency being the example — and, more consequentially, he says China is closer than most people realize to solving the lithography problem, which would let it manufacture AI chips at very large volumes.

On export controls

He's blunt. The US can pass a law stopping American companies from using Chinese models, but it can't stop the rest of the world. And: "I don't think that helps. That won't stop China from being the leader in AI."

Robots

He adds a piece people often miss — robotics. China has very good robot companies. He mentions their robot sporting events, including boxing matches, and a clip he saw of a robot that kept fighting after its head was knocked off.

Safety runs through China

His conclusion on safety is the part worth sitting with. He has proposed that leading labs get a week of private cross-testing before frontier releases — via API, not handing over the model. Talking about China, he widens it himself: that mechanism should probably include the Chinese companies doing frontier models too. And when it comes to enforcement, only two governments actually have the power to act — the US and China.

Takeaways

Two things I'm taking from this. First, don't treat "Chinese models are cheap" as the whole story; doing near-frontier work on far less compute is itself hard power. Second, electricity is the underrated variable. While everyone argues about chip bans, the real bottleneck may be the grid — which is why compute and token supply chains will be some of the most valuable real estate of the next few years.

Source: The Economist, "The full-length interview with Elon Musk," 29 July 2026.