Snowcap Compute is a new company dedicated to developing super-guided AI computing chips. The company was founded to create a computer with the best performance than the current best AI system, while consuming only a small part of its power. The company announced Monday (6/23) that it received US$23 million (about NT$680 million) of funds invested by Playground Global, Cambium Capital and Vsquared Ventures.
Superconductors are materials that allow current to flow without resistance, and Snowcap is not the first company to use superconductor materials to make new chips. The scientific community had a deep understanding of superconductors as early as the 1990s and proposed the idea of making computer chips with it. However, since this type of chip must operate in ultra-low temperature environments, and the cooling equipment itself consumes a lot of power, this has become a natural fatal injury that has been difficult to be practical for decades.
With GenAI driving the huge global demand for computing power, and traditional chips have gradually reached the limit of performance per watt, this undoubtedly provides a machine for superconducting chips to return to the stage. More importantly, the unlimited demand for power by AI data centers has caused unprecedented pressure on the Internet.
Based on the "Rubin Ultra" that Nvidia plans to launch in 2027, in the future, the server of this chip will be used and the estimated power consumption will be about 600 kilowatts. In other words, the power consumed by a single "Rubin Ultra" server at full speed is equivalent to about two-thirds of the average power used by a US household throughout the whole month.
Super-conducting chips have 25 times more efficient than the strongest chips available, and the first basic chip will be launched by the end of 2026With this trend change, Snowcap executive Michael Lafferty believes that as long as the efficiency is improved, and the power of some data centers is distributed to low-temperature cooling equipment, the company's super-guided chip advantages will be revealed.
He is confident that even if the energy consumption required for cooling is considered, its wafers can still be about 25 times more efficient per watt than the strongest wafers available. Further, although energy efficiency is very important, the real selling point lies in higher efficiency. Because of this, the company is committed to significantly boosting high efficiency standards and at the same time reducing power consumption.
In manufacturing, the chip can be produced in standard chip factories, but special metal called niobium titanium nitride needs to be added, which mainly comes from Brazil and Canada. Snowcap plans to launch its first base chip by the end of 2026, but the complete system is expected to be found later.
Snowcap teams all come from the semiconductor/superconductor industry, and Lafferty was responsible for the development of forward-looking chips in Cadence Design Systems. In addition, it includes front-chip high-levels from Nvidia and Alphabet, as well as two scientists with deep research skills in the hyper-conducting chip field, Anna Herr and Quentin Herr, who have also served in chip research agency Imec and defense company Northrop Grumman. As for Playground Global General Partners (GP), Pat Gelsinger, also a former Intel executive, will also join the Snowcap board.
Snowcap Compute raises $23 million for superconducting AI chips