The venture capital arm of Google-parent company Alphabet is leading a $56 million funding round in SambaNova Systems, a start-up building computer processors and software for artificial intelligence and data analytics.
It's the first time Alphabet's venture arm, GV, has invested in an AI hardware company.
The move comes as Google CEO Sundar Pichai and other execs frequently insist that Google is an "AI-first" company. Google sells AI services for developers and counts on it to enable special features in its consumer products, while Waymo relies on AI for autonomous driving. Alphabet called out AI in the business overview section of its 2016 and 2017 annual reports.
But GV, formerly known as Google Ventures, held out on backing a start-up building hardware for AI for the past several years, general partner Dave Munichiello told CNBC.
"I would say by our standards, they're right on time because a bunch of other companies we've looked at were too early," he said. Those start-ups built products that proved to be not quite right for the time when they became available, he said.
GV has a unique insight into why existing products from larger companies won't necessarily be just right for AI in the future, Munichiello said. Google employs graphics cards from Nvidia to train AI models, and Google itself has built a custom AI chip, the tensor processing unit. Intel, another Google partner, has developed silicon for AI as well, following its 2016 acquisition of AI chip start-up Nervana.
But the talent at SambaNova, which has more than 50 employees after being founded in November, comes from different places. CEO Rodrigo Liang ran a team of nearly 1,000 chip designers at Oracle. Others came from Stanford University, including professors Christopher Re and Kunle Olukotun. Liang and Olukotun worked together at chip company Afara Websystems, which Sun acquired in 2002.
Olukotun said that SambaNova's focus area goes beyond training AI models: Its infrastructure will be able to accommodate other types of calculations, including speedy queries of databases.
"All sorts of approaches have been tried, but they're all facing the fundamental limit that is Moore's Law is slowing down," Olukotun said. "To get more performance going forward, we need to think about a much more efficient way of doing computations."
The start-up doesn't have a first-generation chip available yet, but it has spoken with potential customers, as well as public cloud providers, Liang said.
Walden International led the round alongside GV. Atlantic Bridge Ventures and Redline Capital also participated in the round.