Snowcap raises $23m for super cooled computing
US startup Snowcap Compute has raised $23m to roll out a superconducting compute platform that runs digital chips at ultra-low temperatures.
The approach is aimed at reducing the power consumption of traditional datacentre chips such as GPUs and AI accelerators using Josephson junctions rather than transistors.
The $23 million seed round was led by Playground Global, with ex-Intel CEO Pat Gelsinger joining the board of directors as chair.
“Snowcap represents a rare opportunity to realize dramatic performance and power efficiency gains,” said Gelsinger who is General Partner at Playground Global. “Reimagining a post-CMOS world from the ground up with the most capable and experienced team in superconducting technology is exactly the kind of breakthrough that Playground was built to enable. The implications for AI, quantum and HPC are both thrilling and profound.”
The company, based in Palo Alto, says it has solved the key engineering challenges that have prevented superconducting technology from reaching broad commercialization in the past, including scaling, fab compatibility, EDA challenges, and system architecture issues.
The platform allows digital chip designs to operate at 4.5 K using Josephson junctions instead of transistors. Chips can be built on 300mm silicon wafers with a standard fab process without exotic materials and Snowcap will work with customers to produce test chips on the platform at the end of 2026. These chips will have to go through significant validation testing with the shift to the Josephson junction architecture.
Reports put the resulting power efficiency improvement at a factor of 25, which would allow a chip such as the next generation GPUs that are expected to use 1200W to run at just 48W.
Running at 4.5 K allows higher speeds and dramatically lower losses, but requires significant power for the cooling system. Snowcap aims to tap into the 4K dilution fridge technology being developed by quantum computing startup PsiQuantum, which is also backed by Playground. This can operate with a power consumption of around 100W, rather than the higher power need for quantum computing systems that run at even lower temperatures under 1 K.
The combination of a 50W GPU and 100W rack-based cooling system would provide a dramatic power consumption advantage for AI datacentre operators.
“We’re building compute systems for the edge of what’s physically possible,” said Michael Lafferty, CEO of Snowcap. “Superconducting logic lets us push beyond the limits of existing CMOS technology, achieving orders-of-magnitude gains in processing speed and efficiency. That performance is essential for the future of AI and quantum computing.”
The founding team also includes Chief Science Officer Anna Herr, and Chief Technology Officer Quentin Herr, who had key research roles at imec and Northrop Grumman. Brian Kelleher, former SVP of GPU engineering at Nvidia, and Phil Carmack, former VP of silicon engineering at Google, join as advisors.
European semiconductor investment firm Vsquared Ventures and Cambium Capital, which specializes in compute and deep tech investments also joined the round.
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
