A new player just emerged from stealth with a solution to one of AI's most expensive headaches. Niv-AI announced today it's raised $12 million in seed funding to help data centers measure and manage the wild power surges that happen when GPUs shift between workloads. As AI training clusters scale into the hundreds of thousands of chips, those power spikes are becoming a serious constraint on how much computing power can actually fit in a building.
Niv-AI is betting that the next big AI infrastructure problem isn't about getting more GPUs, it's about actually using the ones you have. The startup emerged from stealth today with $12 million in seed funding to tackle what sounds like a mundane issue but is quietly driving data center operators crazy: GPU power surges.
Here's the problem. When Nvidia H100s or similar chips switch between different AI workloads, they don't sip power steadily. They gulp it in massive, unpredictable surges that can spike 50% or more above baseline in milliseconds. Data centers design their electrical infrastructure around peak loads, not averages, which means those surges directly limit how many GPUs you can cram into a facility before you hit the building's power ceiling.
For hyperscalers and cloud providers racing to build out AI capacity, that's not just an engineering headache, it's millions of dollars in stranded infrastructure. You might have physical rack space and cooling capacity for another thousand GPUs, but your electrical panels say no. Meta and Microsoft are building entire data center campuses to support their AI ambitions, and every watt counts when you're deploying hardware at that scale.
Niv-AI's approach centers on real-time measurement and intelligent power management. The company hasn't disclosed full technical details yet, but the pitch is straightforward: if you can predict and smooth out those power spikes, you can fit more compute into the same four walls. That matters enormously in a market where GPU allocation is the new currency and data center space is the constraint.
The $12 million seed round signals that investors see genuine value in infrastructure tooling beyond just the headline-grabbing foundation model companies. While and chase AGI, there's a whole ecosystem of picks-and-shovels companies solving the practical problems of running AI at scale. Power management sits right at that intersection of critical need and underserved market.
