How Anthropic and Google split on the next US-China deal

This article summarizes a discussion at the 2026 World Economic Forum between Anthropic CEO Dario Amodei and Google DeepMind CEO Demis Hassabis.

Anthropic CEO Dario Amodei / Photo credit: Mijansk786 / Shutterstock

Dario Amodei, CEO of Anthropic, has a warning for government leaders. He believes the debate over selling advanced AI chips to China is more than competing on business. It is about national survival, like the biggest international problems of the last century.

AI will soon improve itself

For years, AI progress was limited by computer hardware. Now, Amodei believes the main slowdown is software, a challenge that AI itself can solve. He explains that AI will be able to write software like humans in the near future. This will create a cycle of rapid improvement.

Amodei explains, “We might be six to 12 months away from when the model is doing most, maybe all, of what [software engineers] do, end to end. And then it’s a question of how fast that loop closes… I would guess that this goes faster than people imagine.”

This cycle of improvement has limits. While AI can write code, it cannot build the computer centers it runs on. Demis Hassabis, CEO of Google DeepMind, points out that these real-world delays, from making computer chips to building data centers, will slow things down.

Hassabis says, “Not every part of that loop is something that can be sped up by AI, right? There’s, like, chips, there’s the manufacturing of chips, there’s training time for the model… It’s very hard for me to see how it could take longer than a few years.”

The limits of problem-solving AI

Today’s AI is good at solving clear problems better than people can. Hassabis argues, however, that the last step to AGI [Artificial General Intelligence] is learning to ask the right questions, not just find answers. This part is harder to copy than just the ability to analyze things.

Hassabis notes, “I also think there are some missing capabilities at the moment… not just solving existing problems but actually coming up with the question in the first place or coming up with the theory or the hypothesis. I think that’s much, much harder.”

Real-world tests decide what is true
Even a highly creative AI would still face real-world limits. New ideas only matter once they are tested, and in many scientific fields, that process remains slow and dependent on physical experiments.

Hassabis continues, “Some areas of natural science are much harder to do… You won’t necessarily know if the chemical compound you’ve built or this prediction about physics is correct. You may have to test it experimentally, and that will all take longer.”

Why AI’s economic impact may arrive quickly

As AI improves rapidly, it is expected to displace a growing number of jobs. Unlike past technological shifts that mainly affected manual labour, today’s AI threatens cognitive and office-based roles, raising concerns about the scale and speed of the transition.

Amodei projects, “[I said] half of entry-level white collar jobs could be gone within the next one to five years… my worry is as this exponential keeps compounding… it will overwhelm our ability to adapt.”

The people making AI are the first to feel its effects
This is not a problem for the future. AI makes work so much faster that it is already changing jobs at the companies that make it. The people making this technology are seeing how it changes work.

What purpose look like in a world with less work

The West’s main advantage is controlling computer chips

A race with no rules makes the technology more dangerous