Exclusive: Mirendil inks $100M+ Google Cloud deal to scale self-improving AI

Mirendil, an AI lab whose co-founders came out of Anthropic, has locked in a multi-year compute deal with Google Cloud worth upwards of $100 million, according to CEO Benham Neyshabur. For context, that figure is roughly half of what the startup raised in seed funding back in late June, when it was valued at $1 billion. This isn't just a cloud bill. It's a bet on a specific, ambitious idea: AI that keeps getting smarter on its own.

The arrangement gives Mirendil access to Google's TPUs, Nvidia GPUs, and managed training clusters, all aimed at building what's called self-improving or recursively self-improving AI. Neyshabur describes it as pointing a problem at a system and letting it keep improving over time, the way a human researcher accumulates expertise in a field. He specifically cites Alzheimer's research as the kind of open-ended, ambitious target this technology could chip away at continuously rather than in one-off bursts.

Co-founder Harsh Mehta frames the technical challenge less as raw computing horsepower and more as matchmaking. Different workloads suit different chips, he says, and getting that pairing right is what actually lowers costs, both for Mirendil and for whoever ends up using its systems down the line. Google's own AI infrastructure chief, Amin Vahdat, echoed that framing in a statement, arguing that progress now hinges on orchestrating whole systems of intelligence rather than squeezing more performance out of individual chips.

What Mirendil is chasing isn't small. The company wants its AI to eventually take on the workload of an entire frontier AI lab, automating chunks of scientific research across medicine, biology, and materials science. It's not alone in circling this idea. Anthropic has explored recursive self-improvement, and newer startups like Recursive Superintelligence and Ricursive Intelligence have popped up around the same goal. For Google, the payoff is a partner working the frontier of a technology it can eventually pitch to its enterprise customers, on top of another example of a cloud giant locking down a hungry startup with serious infrastructure commitments.