Can AI Solve the Navier-Stokes Millennium Problem? Inside DeepMind’s Secret Project

The Navier-Stokes equations describe the motion of air, liquids, and essentially any flow — everything that surrounds us. But there is one small catch: no general analytical solution is known. We do not fully understand how the solutions to these equations behave. And that matters not only for mathematics, but for physics as well — for instance, to finally understand the nature of turbulence. So much so that the Clay Mathematics Institute is offering $1,000,000 for a solution to the existence and smoothness problem for the Navier-Stokes equations. It is one of the seven Millennium Prize Problems. This is not just a math puzzle — it is a question about the predictability of the physical world.

The question has remained unanswered for nearly 200 years, since the time of Henri Navier and George Stokes. In 2014, a team led by Thomas Hou at Caltech found the first hint of a “singularity” — a theoretical point at which a solution loses smoothness. But that work involved an inviscid model: the Euler equations, not Navier-Stokes.

The Secret Operation Behind the Navier-Stokes Millennium Problem

For the past three years, 20 engineers from Google DeepMind and five leading mathematicians — headed by Javier Gómez Serrano — have been running a confidential project: Operation Navier-Stokes.

Serrano is a professor at Brown University. Together with co-authors including Tristan Buckmaster and two geophysicists from Princeton, the team is betting that traditional mathematics cannot crack this problem — but AI can.

How AlphaEvolve Is Attacking the Millennium Problem

The tool at the center of the effort is AlphaEvolve, a DeepMind system that does not merely optimize or brute-force solutions. It understands problems. Over four months it was trained on 50 equations, and the results were striking:

• In 75% of cases it reproduced the best known human solutions.
• In 20% of cases it found solutions that were even better.

A breakthrough could arrive within the next one to two years. Demis Hassabis, CEO of DeepMind, has hinted as much: “We are close to solving one of the Millennium Problems.”

What Solving Navier-Stokes Would Mean for Science

If the equations are solved, science will gain a theoretical guarantee of fluid predictability — or, conversely, proof that even mathematics cannot always predict when a calm sea will become a hurricane.

“I believe AI will change the world — and I want to believe it will change it for the better,” says Gómez Serrano.

We are watching this unfold in real time. A solution that would reshape physics. Mathematics. And perhaps the way we understand reality itself.