Google DeepMind: AlphaEvolve available through Google Cloud, first industrial results

Google DeepMind published on May 7, 2026, the first detailed report on the industrial impact of AlphaEvolve, a coding agent powered by Gemini models, and simultaneously opened its commercial availability through Google Cloud. AlphaEvolve, first introduced in May 2025, is now available to organizations in finance, logistics, scientific simulations, and other sectors.

What are the concrete results in science and infrastructure?

In quantum physics, AlphaEvolve reduced errors in quantum circuits by 10× on Google’s Willow processor. In electrical grids, it increased the share of solvable AC Optimal Power Flow problems from 14% to 88%, with direct implications for grid stability. In genomics, it improved DeepConsensus DNA sequencing by 30% in error correction, and in natural disaster prediction it increased accuracy by 5% across 20 risk categories. In mathematics, it solved Erdős problems and improved bounds for the Travelling Salesman Problem and Ramsey numbers.

How does AlphaEvolve affect Google’s own infrastructure?

Within Google, AlphaEvolve optimized TPU chip design, reduced writes to the Spanner cache by 20%, and reduced compiler memory consumption by 9%. These results show that the agent is used as a tool for internal system optimization, not just as a research project. DeepMind states that AlphaEvolve is already in production across multiple Google systems and that the accumulated savings represent a significant share of infrastructure costs.

What does the first group of commercial customers say?

Klarna doubled the training speed of its own transformer model with AlphaEvolve’s help. Logistics company FM Logistic achieved 10.4% better routing efficiency, and marketing group WPP improved the accuracy of its marketing models by 10%. Substrate reported multiple speedups in semiconductor simulation, while Schrödinger achieved 4× faster training and execution of molecular force field models. This is the first group of customers from diverse industries to integrate AlphaEvolve into production processes.

The opening of AlphaEvolve through Google Cloud marks a transition from the research phase into the market category of autonomous optimization agents. DeepMind has not announced a pricing model or SLA details for the Cloud version.