An SEC filing posted late Wednesday shows that a stealthy Silicon Valley AI chip company is most of the way through raising a big round of outside funding.
Follow Crunchbase News on Twitter
Groq, a company with a very spartan website and some intellectual heft behind it, is at least $52.27 million into a targeted $60 million fundraise, according to the filing. At least 16 investors have contributed to the round so far.
In this initial filing, there’s no change in the listed board members from its last funding filing, dated December 2017. Groq’s board members include:
- Chamath Palihapitiya, avid amateur poker player and current CEO of Social Capital.
- Jonathan Ross, a former Google hardware engineer.
- Douglas Wightman, a former Google software engineer.
Ross and Wightman were foundational members of Google’s TPU team. The Tensor Processing Unit is a custom application-specific integrated circuit (ASIC) chip designed to run the TensorFlow algorithm. Google uses the TPU to accelerate machine learning in products including translation, photo management, and search.
ASIC chips are optimized for efficiency over flexibility. They’re designed to do one set of computations very quickly and efficiently. In the TPU’s case, it’s crunching the multivariate math of machine learning.
Back in March 2017, Social Capital’s Chamath Palihapitiya appeared on Squawk Box and told his CNBC hosts that he’d spent 1.5 years recruiting the engineers behind the TPU at Google. “But,” he said, “we were able to take them out. And we have eight of the 10 original people that built that chip building the next generation chip now [at Groq].”
About a month after the cable news appearance, news broke that Palihapitiya led a $10 million funding round in Groq.[irp posts=”15097″ name=”Artificial Intelligence Continues Its Fundraising Tear In 2018″]
And apart from a couple of stories about the company hiring a former Xilinix executive as its chief operating officer, there hasn’t been much press coverage about the company.
The only publicly-facing announcements from the company are in the form of simple, numbers-driven graphs posted to the company’s blog (which, for now, is also its homepage).
A graphic posted in November 2017 says the company plans to ship its first machine learning product, a “single chip,” in 2018. The graphic says Groq’s first chip can perform 400 trillion operations (tera-operations) per second. As far as efficiency goes, Groq’s silicon performs 8 trillion operations per watt of electricity used.
The most recent update to the company’s blog, dated January 2018, promises an “open platform” and “performance without lock-in.”
There are stealthy startups, and then there are stealthy startups. Although the silicon business isn’t exactly an open book, other AI chip companies have been slightly more forthcoming. With tens of millions in fresh capital and plans to debut its tech sometime in the next four months, Groq should start making some noise. After all, that same graphic promises that its first product is coming this year.
There are only so many months left.
Illustration: Li-Anne Dias