Cerebras IPO: It Takes a Village
In 2017, I signed an offer to join Cerebras Systems (Nasdaq: CBRS) as the first new-grad hire in the hardware systems org. Five years later, the company looked very different. We had grown from 50 employees to 400, and we were shipping some of the most ambitious AI systems in the world.
Despite this explosive growth, Cerebras had its first round of layoffs soon after I departed. It was a difficult time for employees, alumni, and investors who faced the shadowy uncertainty of a company with an unknown future. Many investors wrote the company off. Many employees left without exercising their ESOPs. This wasn’t unique to Cerebras. Most AI chip startups were closing up shop, unable to compete against NVIDIA’s distribution moat. But it was particularly frustrating for early investors looking to prove hardware naysayers wrong. Among those investors was Eclipse.
In 2022, I joined the Eclipse team. At the time, Eclipse was commanded by the legendary Pierre Lamond and his co-founder Lior Susan. As the valley’s most notorious semiconductor founder-investor, Pierre sat on the board of Cerebras and was Andrew Feldman’s prized confidant. Pierre was a hero to me and many others in the semiconductor industry, and I was fortunate to spend time with him while at the firm.
When I joined Eclipse, I offered a different perspective as a former Cerebras employee. I had a lot of opinions and I delivered many of these, unsolicited, in the form of long documents, decks, and water-cooler keynotes to my colleagues. One such deck was circulated to everyone at the firm entitled “Cerebras Pre-Mortem & Strategy Proposal”. In it were 50 detailed slides outlining a sobering picture of the state of the company. It was February 2023, and Cerebras was burning cash, executing RIFs, and losing support from all directions.
The picture in that deck was unique. Investors saw only a top-down view at board meetings. Meanwhile, I developed a litany of opinions from a bottoms-up view as a former employee. Everything from engineering challenges, to communication bottlenecks, to unit economics and product problems. The deck was circulated among key board members. People paid attention, and Eclipse intensified its involvement with the company.
Over the course of the following weeks, we scrutinized every communication, every decision to come out of Cerebras. Lior called in famed investors and leaders at the company one-by-one into Eclipse’s Palo Alto office. He’d gather information, then we’d debrief afterwards. It was a quintessential example of hands-on investing.
On April 28th 2023 it became official: G42 was to sign a historic deal with Cerebras to become the company’s largest customer by a wide margin. Lior had made an introduction to G42 and organized talks leading up to the deal. Behind the scenes, I was leading talks of a datacenter joint-venture that we’d incubate to make G42’s Condor Galaxy available to the public. Cerebras was now in a very different position, and for the first time in years, the future no longer felt uncertain.
At close of business that day, I handed Lior a hat and wand and said, “Lior, you’re a wizard.”
4/28/23 at the Eclipse Palo Alto office
Cerebras now had the runway and traction to try another product path: low-latency inference. Employees returned to the company, market demand for compute spiked, and general AI usage shifted to the API layers, enabling Cerebras and others to tackle more tractable parts of the inference stack.
This past week, Cerebras had one of the largest IPOs in history.
History is way messier than people realize. It takes a village to build a company. I consider myself fortunate to have played a part in this story and to have seen what real conviction and support for a startup looks like.
Congratulations to the Cerebras team and all my former colleagues. And congratulations to Eclipse and the investors who stuck with Cerebras through the difficult times. It’s been a thrilling journey.
Cerebras circa 2018 in the original Los Altos office