By Stephen Nellis
(Reuters) – Enfabrica, a Silicon Valley chip startup working on networking chips for artificial intelligence data centers, said on Tuesday that it has raised $125 million in venture capital, with Nvidia joining as a strategic investor.
Founded by executives from Broadcom and Alphabet’s Google, Enfabrica is part of a broader trend of data centers being completely re-designed to create generative AI technologies similar to ChatGPT. Chips from Nvidia, the world’s most valuable semiconductor firm, are at the heart of that transition.
But Nvidia’s graphics processing unit (GPU) chips face a problem: They sometimes sit idle because the networks connecting them cannot feed them data fast enough.
Enfabrica has designed a network chip that aims to connect the different pieces of a data center together in new ways to solve that problem. The Enfabrica chip creates a network that looks like a hub and spokes, allowing the Nvidia GPUs doing the data crunching to draw data from multiple different places without hitting speed bumps.
Enfabrica co-founder and Chief Executive Rochan Sankar said that leads to much more efficient use of GPUs so that the same amount of computing work can be done with about half as many chips because they are kept busy the whole time. In the technology industry, such a feat is seen as a positive because busier chips are more cost effective.
“It’s no secret to Nvidia or anybody else out there that in order for AI computing to become truly ubiquitous, the cost curve has to come down,” Sankar said. “The key here is that we enable those GPUs to be better utilized.”
The Series B venture funding round was led by Atreides Management, whose founder Gavin Baker, a veteran of Fidelity Investments, is taking a seat on Enfabrica’s board. Enfabrica added new backers that included IAG Capital Partners, Liberty Global Ventures, Valor Equity Partners, Infinitum Partners and Alumni Ventures. Earlier investor Sutter Hill Ventures also joined the round.
(Reporting by Stephen Nellis in San Francisco; Editing by Stephen Coates)
