Rahul Kar

Ask a room full of AI builders what's holding them back and they'll say chips. Maybe models. Maybe data. Rahul Kar listens to all of it, then points at the wall socket. The constraint, he argues, was never silicon. It was the watt.

Kar is the founder and CEO of Hammerhead AI, a Redwood City company that came out of stealth in November 2025 with a $10 million seed round led by Buoyant Ventures. The product is a piece of software with a deliberately aquatic name - ORCA, for Orchestrated RL Control Agents - and a contrarian premise: most data centers are running at 30 to 50 percent of the power they're already allocated. The other half sits there, paid for, stranded, doing nothing.

"All data centers that we have seen and have researched are severely underutilized," Kar says. His follow-up question is the entire business plan: "If there is slack power available within data centers, why do you need to go to the grid?"

It's a good question to be asking in 2025, a year when the bottleneck for artificial intelligence quietly stopped being compute and became electricity. Utilities are quoting multi-year waits for new interconnections. Operators are floating diesel generators and behind-the-meter gas. And Kar is walking in with a different message: you don't need more power. You need to actually use the power you have. Hammerhead's pitch is that ORCA can squeeze roughly 30 percent more usable capacity out of an existing facility - without a single new megawatt from the grid, and without breaking the service guarantees operators have already promised their tenants.

If that sounds like a trick of arithmetic, it isn't. It's a trick of orchestration, and Kar has spent more than a decade learning how to do it.

Before Hammerhead, Kar was the chief operating officer at AutoGrid, one of the original believers that software could turn a rigid electrical grid into something that bends. There, he scaled virtual power plants and grid-flexibility tools - the unglamorous machinery that lets a utility coordinate thousands of distributed energy resources into a single, responsive whole. The team didn't run a pilot and call it a day. They orchestrated roughly 8,000 megawatts of flexibility across more than 20 countries, and rode AutoGrid through to its acquisition by Schneider Electric.

That decade taught Kar a lesson he now repeats like a mantra: "The physical reality has always mattered." You can have the cleverest model in the world, but electrons still have to flow through copper, heat still has to go somewhere, and a transformer doesn't care about your roadmap. Intelligence without physics is a slide deck.

Along the way he also worked on technology strategy with the U.S. Department of Energy, helping shape how sustainable energy gets managed at scale. So when the AI boom arrived and everyone started panicking about power, Kar wasn't looking at a new problem. He was looking at his old problem wearing a new outfit. Grids and data centers, it turns out, suffer from the same disease: capacity that exists on paper but never gets fully used. He'd already built the cure once.

The founding moment he describes most warmly isn't the term sheet. It's the people. When respected colleagues from his AutoGrid days walked away from comfortable jobs to join him, he says, the idea stopped being his and became "a shared mission for a team." Co-founder Rajeev Singh signed on as CTO. Sadia Raveendran, who had managed AutoGrid's Schneider partnership from early investment all the way through the acquisition, took the go-to-market reins. The band, more or less, got back together - this time pointed at the hungriest electricity customer on Earth.