Cerebras IPO makes billions for Benchmark, but VC Eric Vishria almost didn't take the meeting

At a glance:

• Cerebras Systems' IPO was a breakout success, with early investor Benchmark holding 9.5% of the company in shares worth over $5 billion at first-day trading prices.
• Benchmark general partner Eric Vishria nearly passed on the meeting in 2016, calling it his first hardware investment in a decade and questioning why it was on his calendar.
• After 8.5 years of development struggles, national security scrutiny delays, and a brutal fundraising cycle, Cerebras is now profitable with OpenAI and AWS as major customers.

The meeting Eric Vishria almost skipped

Eric Vishria had been a venture capitalist for roughly 18 months when a pitch landed on his calendar from a then-unnamed AI chip startup. The founders — five of them — had a slide deck and a bold plan, but Benchmark, the famously selective VC firm, did not do hardware deals often. Vishria had already started kicking himself for agreeing to the meeting when he sat down to hear what Andrew Feldman and his team had to say.

"Why did I take this meeting?" he kept muttering, he later told TechCrunch. At one point he even messaged his assistant — the person who managed his calendar — and bugged her: "Why did you let me take this meeting?" But by the third slide, his irritation had given way to something else entirely. Feldman had laid out the argument that "GPUs actually suck for deep learning. They just happen to be 100 times better than CPUs." For Vishria, it was a light-bulb moment: "I was like, 'Oh, my God, of course. Like, why would a graphics processor be the right thing for AI?'"

A bet far outside Benchmark's comfort zone

Benchmark backed hardware companies so rarely that the firm did not have the internal expertise to evaluate the deal on its own. Vishria brought the pitch to founding partner Bruce Dunlevie, one of Benchmark's original founders from the 1990s who actually understood chip design. Dunlevie grilled Feldman on packaging, cooling, and manufacturing feasibility. "Most of that meeting was like a dog watching TV for me," Vishria joked, admitting he understood very little of the technical deep-dive.

Dunlevie warned that what Cerebras was attempting would be extraordinarily difficult — others had tried and failed. His chief concern was not the technology itself but whether a market existed for such a chip. Vishria, however, was convinced of one thing: if Cerebras could make AI faster, customers would come. The team's track record helped. Feldman and co-founder and CTO Sean Lie had previously sold their startup SeaMicro to AMD, which gave the pair credibility in the eyes of investors. "The advantage of having had a successful exit previously is it erases some of the uncertainty in the venture capitalists' minds," Feldman told TechCrunch.

Eight and a half years of hardware grind

What followed was an 8.5-year development marathon that tested the founders and their backers alike. Feldman and Lie had to invent entirely new cooling methods to prevent a wafer-scale chip from overheating under its own power draw. They also had to invent a machine capable of drilling 40 screws into a wafer simultaneously without cracking it — one of many manufacturing problems with no off-the-shelf solution.

The financial pressure was relentless. At one point, Cerebras had raised roughly half a billion dollars from a long list of investors, yet its chips were still in development. When the 2022 venture capital bear market hit, the company had to raise again under punishing conditions. "You don't have a lot of traction on the company yet, so yeah, that was where it got really tough," Vishria recalled. Throughout it all, he admits he repeatedly thought to himself, "What are we doing?"

The G42 roadblock and a delayed IPO

Cerebras first attempted to go public in 2024, but the offering stalled after U.S. government scrutiny over national security concerns linked to a large investment by G42, an Abu Dhabi-based cloud provider and the company's only major customer at the time. Public investors were also uneasy about Cerebras's heavy dependence on G42 and its substantial losses.

The delay, it turned out, was a blessing in disguise. In the roughly 18 months that followed, Cerebras' chips — designed primarily for AI training and manufactured by TSMC, the world's largest contract chip manufacturer — proved to be even better at inference, the task of running AI models to generate responses rather than training them from scratch. Just as that realization took hold, the broader AI industry's demand for compute surged to insatiable levels. OpenAI and AWS signed on as large customers, and Cerebras doubled its revenues and declared a profit.

The payoff

Benchmark's patience turned out to be historic. The firm initially bought approximately 80 percent of its eventual stake in early rounds for around $18 million, with the remainder acquired in later rounds at higher prices for roughly $250 million — bringing total investment to approximately $270 million, according to various disclosures and Vishria's confirmation to TechCrunch.

At the IPO's opening price of $185 per share, Benchmark held 17,602,983 shares worth $3.3 billion. When first-day trading pushed the price above $300, that stake's value exceeded $5.3 billion. A standard six-month lockup prevents insiders from selling immediately, but the returns are staggering by any measure. Vishria credits the Cerebras team for "persistence, ingenuity, but also adaptiveness." As for his assistant — the one he gave grief for scheduling that fateful first meeting — he laughed and said, "I think she'll do well, very well."

Why it matters

The Cerebras–Benchmark story is more than a big IPO headline. It illustrates how the most consequential bets in technology often look wrong at the outset. A general partner at one of Silicon Valley's most prestigious firms openly admitted he did not understand the technology, had to bring in a co-founder to vet the deal, and spent nearly a decade watching the company burn cash before the market caught up. In an era where AI infrastructure is one of the most fiercely contested sectors in tech, Cerebras' wafer-scale approach — once dismissed as a niche hardware gamble — now underpins some of the most demanding AI workloads on the planet.

The story also raises questions about the concentration risk that nearly derailed the IPO. G42's Abu Dhabi ties triggered a national security review that delayed the offering by more than a year, underscoring how geopolitical entanglements can complicate even the most promising tech IPOs. With OpenAI and AWS now in the customer mix, Cerebras has diversified away from that single-client dependency — but the episode remains a cautionary data point for any AI hardware company reliant on sovereign-wealth-backed customers.