Foundation Capital's Steve Vassallo: Cerebras' first investor on five-startup-in-one engineering risk and the Solana connection

Steve Vassallo on Cerebras: Five startups in one

Steve Vassallo, a general partner at Foundation Capital, was Cerebras' first institutional investor, writing the initial check in 2016 when the company was betting that AI workloads would eventually demand purpose-built silicon. The thesis was workload-driven: Vassallo and his partners watched ML compute demand spike steeply across their portfolio and concluded that a new class of processor, purpose-built for AI training, was inevitable — the same pattern that produced GPUs for graphics and mobile chips for low-power computing.

What he didn't fully advertise at the time was how many problems Cerebras was actually solving simultaneously. Vassallo describes it as "five startups in one": how do you yield a semiconductor the size of a dinner plate? How do you power it, cool it, maintain continuity across thousands of connections, integrate it into a data center, and then rack 64 of them together? Each problem was hard enough alone. Stacked, the risks become combinatorial. He recalls leaving board meetings genuinely uncertain whether the team would find its way through a fundamental thermodynamics challenge — and not a negotiating problem, he's clear, but a physics one. As he puts it, Andrew Feldman is a strong negotiator, but even he can't negotiate with the second law of thermodynamics.

The pivot to inference came roughly seven years in, when co-founder Sean flagged from the boardroom that inference was exploding. Vassallo frames this as the intended arc for hard-tech investing: start focused enough to win, then rotate toward the much larger opportunity. The earlier risk he worried about wasn't being too narrow — it was the opposite, getting siloed into traditional high-performance computing use cases that, while real markets, weren't growing anywhere near the rate of inference and reasoning workloads.

“We worried a little bit about being in a niche that was not interesting enough to build a really nodal company. There were like five startups worth of hard problems for us to go after. How do you yield a semiconductor that's the size of a dinner plate? How do you power it? How do you cool it? The risks are combinatorial — even more dangerous. Anatoly, co-founder of Solana, chose to work with us partly because we were investors in Cerebras — he said, 'You guys take hard problems seriously.'”

Going public

Vassallo's advice to founders entering public markets comes down to three things. First, accept that share price volatility is largely disconnected from daily execution — especially in an environment where a new model drop can move markets within days — and make sure engineering teams understand that. Second, build the operational discipline that public life demands: earnings cadences, investor communications, and the ability to talk about "the business of the business" rather than just the technology. Third, and most urgently, don't let the quarterly cycle kill long-horizon thinking. Vassallo calls that quarterly mindset "one of the most toxic ways to kill a company built around innovation."

Robotics

Vassallo studies the robotics space with the scar tissue of someone who studied embedded systems and spent five years at IDEO designing products for Apple, Cisco, and others. He's skeptical of the humanoid form factor for most applications. A human body is poorly suited to moving pallets around a factory floor. His frame is that the highest compliment a robotic system can earn is when people stop calling it a robot — when it becomes a forklift, or a washing machine, and the technology disappears into the application. He's more interested in focused, application-specific automation than general-purpose humanoids, and applies the same lens he used on Cerebras: find the workload, find where compute is spiking, and look for purpose-built solutions that start small enough to win.