By Stephen Nellis
SAN FRANCISCO, Dec 2 (Reuters) – Axiado, a Silicon Valley-based startup that is making a chip designed to save space and power in artificial intelligence servers, said on Tuesday it has raised $100 million in fresh capital.
Inside AI data centers, thousands of powerful servers must work together in concert to train models. Doing so requires coordination over an internal network, and all servers contain a handful of chips whose job it is to handle those instructions and manage the server’s circuit board while keeping it secure from hackers.
Those board management chips tend to use older technologies, and Axiado is working to consolidate all of them into a single, smaller chip. At a time when major chip firms such as Nvidia and Advanced Micro Devices are working to pack as many of their graphics processing units into servers as possible and data centers are trying to make room for liquid cooling equipment, saving space counts.
“Real estate is really valuable there,” Andrew Homan, managing partner at Maverick Silicon, which led the funding round, told Reuters. “If you can open up footprint, that’s a huge value-add to any of the big incumbents that are building these systems.”
But Axiado’s chip also employs AI on its own. The chip can learn what normal behavior from a server looks like when a given software program is running on the server.
Using that knowledge, the Axiado chip can check for any divergences from normal behavior that might represent a cybersecurity threat. Axiado’s chip can also learn to control the server’s cooling system, dialing it up and down, saving up to 50% of the energy used in cooling.
“Our AI engine learns the behavior. Based on the behavior itself, I can tell that particular load will take a certain level of capacity of the GPU, so you don’t need to run if you had a full scale,” Gopi Sirineni, founder and CEO of Axiado, told Reuters.
(Reporting by Stephen Nellis in San Francisco; Editing by Lincoln Feast.)
Comments