AI startup Snowcap raises $23M in funding to build superconducting chips that could outperform nvidia

Snowcap Compute, an AI startup working on next-gen computing chips built with superconductors, just raised $23 million in funding—and scored a high-profile addition to its board: former Intel CEO Pat Gelsinger. Cambium Capital and Vsquared Ventures also joined the funding round.

The California-based startup is building AI chips that run on superconducting technology, which allows electricity to flow with zero resistance. That could make them far more energy-efficient than today’s chips—if they can get it to work at scale.

Snowcap’s goal? Build computers that could outperform top-tier AI systems, all while using just a fraction of the electricity. That’s a big claim, but one rooted in real physics. Superconductors aren’t new—scientists have studied them for decades—but making practical chips from them has always hit a wall: they have to stay extremely cold to work. And that usually means a lot of power-hungry cooling systems.

But the AI boom has changed the equation. Chatbots and large language models are pushing chipmakers to their limits, and energy demands are starting to outpace what power grids can deliver. At a time when efficiency matters more than ever, revisiting superconductor tech suddenly makes a lot more sense.

Nvidia’s upcoming “Rubin Ultra” AI server, expected in 2027, is projected to draw around 600 kilowatts of power. Running it at full tilt for just one hour could use up two-thirds of the electricity an average U.S. household burns through in a month, Reuters reported.

That kind of math opens the door for alternatives—especially if the performance gains outweigh the cost of keeping chips cryogenically cool. Snowcap’s CEO, Michael Lafferty, says they’re on track to do exactly that. He claims the company’s chips will deliver 25x better performance-per-watt than today’s best processors, even after factoring in the cooling overhead.

“Power (efficiency) is nice, but performance sells,” said Lafferty. “So we’re pushing the performance level way up and pulling the power down at the same time.”

Lafferty isn’t new to futuristic hardware. Before Snowcap, he led advanced chip R&D at Cadence Design Systems. The founding team also includes Anna Herr and Quentin Herr, who previously worked on superconducting chips at Imed and defense contractor Northrop Grumman, along with former chip execs from Nvidia and Google.

The chips themselves won’t require exotic fabrication. They can be produced in standard factories. But the materials are a different story. To build these chips, Snowcap will use niobium titanium nitride—a specialty metal that relies heavily on suppliers in Brazil and Canada.

A basic version of Snowcap’s chip is expected by the end of 2026. Full systems will come later.

Gelsinger, who led the investment through Playground Global and is now joining Snowcap’s board, said the industry has reached a tipping point. AI is chewing through electricity, and power is quickly becoming a bottleneck for data centers.

“A lot of data centers today are just being limited by power availability,” he said.

Snowcap is still early, but the bet it’s making feels very 2025: smarter chips, less electricity, and a shot at outpacing the giants.