Eric Vishria was annoyed. The Benchmark partner had been a venture capitalist for barely 18 months, and here was his calendar assistant scheduling him for yet another meeting with a hardware startup. He didn’t do hardware. Benchmark didn’t do hardware. Why was he taking this meeting?
He kept muttering the question to himself, even shooting his assistant a frustrated message. The answer, it turned out, would be worth billions of dollars.
According to reporting from TechCrunch, Cerebras Systems’ IPO on Thursday was a runaway success, and Benchmark’s stake in the AI chip company is now worth somewhere north of $5 billion. The VC firm owns 9.5% of the company, with Vishria having sat on the board since 2016. But that stake almost never happened because Vishria nearly talked himself out of that first pitch meeting before it even began.
The Third Slide Changed Everything
The meeting itself sounded unremarkable: five founders and a deck. But somewhere around slide three, Vishria’s irritation evaporated. Cerebras co-founder and CEO Andrew Feldman was explaining something that clicked instantly in Vishria’s head. GPUs sucked for deep learning. Yes, they were 100 times better than CPUs. But why would a graphics processor be the right tool for AI at all?
“As soon as he said it, a light bulb went off,” Vishria told TechCrunch. “I was like, ‘Oh, my God, of course. Like, why would a graphics processor be the right thing for AI?’”
It was a simple insight that cut through years of industry assumptions. But Vishria still didn’t understand the hardware side well enough to lead the deal alone. Benchmark’s other partners told him they didn’t know enough about chip manufacturing either. If he wanted to pursue this, he’d need to bring in Bruce Dunlevie, one of the firm’s founding partners from the 1990s, who actually understood the space.
Dunlevie grilled Feldman about chip packaging, cooling systems, and the brutal manufacturing challenges ahead. He warned that others had tried and failed. But he also thought this particular team might actually pull it off. They had previously sold SeaMicro to AMD, which meant they weren’t inexperienced operators gambling with venture capital money.
The risk was still enormous. Dunlevie worried there might not even be a market for a chip like this. Vishria couldn’t fully comprehend the technical details, but he could understand a simple idea: if Cerebras could make AI faster, there would definitely be a market. And this team had the track record to try.
The Long Grind
What followed was 8.5 years of brutal, relentless difficulty. This wasn’t a software company that could iterate quickly and cheaply. This was hardware. Every problem required inventing new solutions that didn’t exist yet.
Feldman and co-founder Sean Lie had to develop entirely new cooling methods to prevent a massive chip from burning itself to death under power. They had to invent machinery that could simultaneously drill 40 screws into a silicon wafer without shattering it. The list went on.
Hardware is also crushingly expensive. By the time Cerebras had raised half a billion dollars from a crowded investor list, it still didn’t have a finished product. It needed to raise more money. Then came the 2022 VC bear market, when raising capital for an unprofitable hardware company felt nearly impossible.
“You don’t have a lot of traction on the company yet, so yeah, that was where it got really tough,” Vishria recalled to TechCrunch. Venture capitalists are paid to be optimistic, but even the most devoted believer has moments of doubt. Vishria had plenty of them. What were they actually doing here?
The Moment It Clicked
About 18 months ago, everything shifted. Cerebras’ chips had been designed specifically for training AI models. But it turned out they were even better at inference, the process of running those trained models to generate responses. The timing was absurd. Just as this realization hit, the entire AI world exploded with demand for exactly this kind of computing power.
Cerebras went from burning through capital with minimal traction to having major customers and real revenue. The company tried to go public in 2024, but got tangled up in U.S. national security reviews. A massive investment from G42, an Abu Dhabi-based cloud provider and Cerebras’ primary customer at the time, spooked regulators. Public investors were also nervous about the concentration risk.
The delay proved fortunate. By the time Cerebras actually went public, OpenAI and AWS were major customers too. Revenue had doubled. The company declared a profit.
The Numbers
Benchmark bought roughly 80% of its current stake in the early rounds for about $18 million, according to various disclosures and Vishria’s own comments. The remaining 20% cost around $250 million at later, pricier rounds. Total Technology investment: approximately $270 million.
At the IPO’s opening price of $185 per share, Benchmark’s 17.6 million shares were worth $3.3 billion. By the end of the first day of trading, as the stock climbed above $300, that stake was worth over $5.3 billion. The firm can’t sell any of those shares for six months due to standard lockup restrictions, but when that window opens, the return will be staggering.
It’s the kind of outcome that makes venture capital tick, and it’s particularly remarkable coming from a firm that made its name by being ruthlessly selective about which Business investments it even considers. Benchmark is the firm that fires founders and makes major strategic bets, often concentrating capital in a small number of companies rather than diversifying across dozens of middle-tier prospects.
This win came because Vishria’s assistant forced him to take a meeting he didn’t want to take. When asked if that assistant would be getting a nice bonus, Vishria laughed and said, “I think she’ll do well, very well.”
The real lesson here isn’t that perseverance and ingenuity always pay off. It’s that sometimes the best opportunities come disguised as frustrations, and that saying yes to something outside your comfort zone, at precisely the right moment, can be worth billions. The question is whether anyone recognizes it when it’s sitting across from them on slide one.
