When Cerebras Systems went public on Thursday, the AI chip maker delivered a multibillion-dollar windfall for its investors. Among the biggest winners is Benchmark Capital, which owns 9.5% of the company — a stake worth over $5.3 billion at the first day’s closing price. But that payout almost never happened, because the venture capitalist who led the deal nearly skipped the first meeting.
A Reluctant Start
Eric Vishria, a general partner at Benchmark, was less than enthusiastic when Cerebras co-founder and CEO Andrew Feldman first came knocking in 2016. “Why did I take this meeting?” Vishria recalled muttering to himself, according to an interview with TechCrunch. He even messaged his assistant to ask why she had let the meeting on his calendar. Benchmark had not made a hardware investment in a decade, and Vishria had been a VC for only 18 months at the time.
Also read: AI Pioneer Richard Socher Launches Recursive Superintelligence With $650M to Build Self-Improving AI
But by the third slide of Feldman’s pitch, Vishria’s attitude changed. Feldman argued that GPUs were fundamentally ill-suited for deep learning — they just happened to be 100 times better than CPUs. “A light bulb went off,” Vishria said. The problem was that Vishria lacked the hardware expertise to evaluate the deal on his own.
Bringing in the Old Guard
Benchmark’s other partners told Vishria that if he wanted to pursue the investment, he would need to involve one of the firm’s founding partners from the 1990s who understood chip design. Vishria arranged a second meeting where Feldman pitched to Bruce Dunlevie, who grilled the founder on chip packaging and cooling. “Most of that meeting was like a dog watching TV for me,” Vishria joked. Dunlevie warned that what Cerebras was attempting would be extremely difficult — others had tried and failed — but he believed the team had a real shot. His main concern was whether a market for such a chip would exist.
Also read: Clawdmeter: A tiny desktop dashboard that visualizes your Claude Code token usage in real time
Vishria was convinced that if Cerebras could make AI faster, demand would follow. The team had a track record: they had previously sold SeaMicro to AMD. “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 Grind
Building a wafer-scale chip proved to be as hard as Dunlevie predicted. Cerebras had to invent new cooling methods to prevent the chip from burning itself out. It had to design a machine that could drill 40 screws into the wafer simultaneously without cracking it. The company raised half a billion dollars before its chips were fully developed, and had to raise again during the 2022 VC bear market. “You don’t have a lot of traction on the company yet, so yeah, that was where it got really tough,” Vishria said.
The breakthrough came about 18 months ago, when Cerebras discovered that its chips — originally designed for training AI models — were even better for inference, the process of running models to generate responses. Just as that realization hit, demand for inference compute exploded. Cerebras landed a major customer in Abu Dhabi-based cloud provider G42 and began generating meaningful revenue.
A Delayed IPO That Paid Off
Cerebras initially attempted to go public in 2024, but the offering was delayed by U.S. government scrutiny over national security concerns tied to G42’s investment. Public investors were also wary of the company’s dependence on a single customer and its large losses. The delay proved beneficial. By the time Cerebras finally listed, it had added OpenAI and AWS as major customers, doubled its revenue, and declared a profit.
Benchmark owned 17,602,983 shares at the IPO. At the opening price of $185, the stake was worth $3.3 billion. By the end of the first trading day, with the stock above $300, it was worth over $5.3 billion. The firm cannot sell until a six-month lockup expires, a standard restriction for insiders. Benchmark spent approximately $270 million to acquire its stake across multiple funding rounds, according to regulatory filings and Vishria’s confirmation.
What This Means for Venture Capital
The Cerebras story is a reminder that the biggest venture bets often come from deals that almost didn’t happen. Vishria’s willingness to step outside Benchmark’s comfort zone — and to rely on the judgment of more experienced partners — produced one of the firm’s most lucrative investments. It also highlights the importance of persistence in hardware investing, where product cycles are measured in years and technical risk is high.
As for Vishria’s assistant, the one who scheduled that first meeting? “I think she’ll do well, very well,” Vishria said with a laugh.
Conclusion
Cerebras’ IPO is a landmark event for the AI chip industry and for venture capital. It demonstrates that even in an era of software-first investing, bold hardware bets can generate extraordinary returns — provided investors have the patience and conviction to see them through. The story also underscores the value of experienced teams, adaptive strategy, and a willingness to learn from partners with deeper expertise.
FAQs
Q1: How much did Benchmark invest in Cerebras?
Benchmark spent approximately $270 million to acquire its 9.5% stake across multiple funding rounds, from the Series A to later rounds.
Q2: Why did Cerebras delay its IPO?
The IPO was delayed in 2024 due to U.S. government scrutiny over national security concerns tied to a large investment from G42, an Abu Dhabi-based cloud provider that was Cerebras’ only major customer at the time.
Q3: What makes Cerebras’ chip different from other AI chips?
Cerebras builds a wafer-scale chip — a single, massive processor that is much larger than traditional chips. It was originally designed for AI training but turned out to be even more effective for inference, the process of running AI models to generate responses.
