Google DeepMind's AlphaEvolve Discovers New Math, Recovers 0.7% of Google's Global Compute

Google DeepMind's AlphaEvolve Discovers New Math, Recovers 0.7% of Google's Global Compute

While OpenAI dominates headlines, Google DeepMind has been quietly deploying something far more consequential: AlphaEvolve, a Gemini-powered coding agent that pairs large language models with evolutionary algorithms, has been used to push the boundaries of complexity theory, discovering new mathematical structures that improve state-of-the-art results on long-standing open problems. The same system has already been quietly deployed inside Google's infrastructure for over a year, recovering 0.7% of Google's worldwide computing resources continuously and speeding up a key kernel in Gemini's architecture by 23%.

Let that sink in: 0.7% of Google's compute is an enormous number. At Google's scale, that's likely measured in thousands of GPUs.

What This Really Means

AlphaEvolve is a Gemini-powered coding agent for designing advanced algorithms showing incredible promise for application across many areas in computing and math and could be transformative across many more areas such as material science, drug discovery and energy. For example, AlphaEvolve enhanced the efficiency of Google's data centers, chip design and AI training processes — including training the large language models underlying AlphaEvolve itself.

This is recursive optimization: the AI that trains AI is improving the hardware that trains AI. It's not AGI, but it's the closest thing we've seen to self-improving systems already embedded in production infrastructure.

The Competitive Angle

While OpenAI talks about autonomy, DeepMind is actually building it—at the infrastructure level, where it matters. DeepMind's founder and CEO Demis Hassabis called it "the engine room" of Google's AI efforts, adding that changes had been made to enable the tech giant to rapidly roll out AI products amid a "ferocious competitive environment."

My take: Google's advantage isn't in models or hype. It's in systematic infrastructure optimization at inhuman speed. Expect AlphaEvolve derivatives across every major tech company within 18 months.