Phaidra raises $50M+ Series B to build AI agents that autonomously run data center infrastructure

Summary

Phaidra has raised just over $50 million in a Series B to build AI agents that autonomously operate and optimize the physical infrastructure inside large-scale AI data centers — what Nvidia calls "AI factories."

Founder Jim (a mechanical engineer who joined Google in 2010) traces the idea directly to AlphaGo. After helping design Google's early large cooling systems and then leading its data center energy efficiency programs, he cold-emailed Mustafa Suleyman, co-founder of DeepMind, in 2016 with a pitch: apply reinforcement learning agents to control large-scale infrastructure the way AlphaGo learned to play Go. The experiment worked. He spent three years at DeepMind leading a vertical called DeepMind Energy, applying reinforcement learning to commercial building HVAC and similar systems. Phaidra grew out of that work, co-founded with Veta, one of the original AlphaGo engineers.

The core product

Phaidra's agents ingest hundreds of thousands of sensor readings in real time across a data center's mechanical systems — pumps, chillers, liquid cooling CDUs — and issue AI-generated command signals to local control systems for automatic execution. The positioning is "AI-enabled supervisory control": a global view of the entire facility, acting like, in Jim's words, "a general on the battlefield."

The agents handle software-level orchestration. Physical tasks — turning wrenches, taking a chiller offline for cleaning — still require humans, and Jim expects robots to eventually fill some of that gap.

Why now, and why hard to replicate

AI factories at gigawatt scale are orders of magnitude more complex than the 30-megawatt modular data centers that were considered large in 2010. That complexity creates a real control problem, and a real labor problem. The data center operations workforce is aging out, taking institutional knowledge with it. Phaidra frames its agents as "virtual plant operators" — permanent members of the operations staff who never go offline.

The integration layer is non-trivial. Legacy automation vendors like Schneider Electric and Siemens operate within walled gardens. Protocols like Modbus, OPC-UA, and MQTT exist, but third-party access to the data and analytics flowing off mission-critical systems is deliberately restricted. A large part of Phaidra's work is building integrations across these SCADA and building management systems — domain knowledge that, as Jim puts it, you can't pick up from a Google search.

The Series B positions Phaidra to scale that integration work and build out agents purpose-built for the AI factory buildout underway across the industry.