Cerebras IPO nets billions for Benchmark, despite almost missed meeting

The Cerebras Systems IPO delivered an enormous payday on Thursday, generating billions for the company, its founders, and its top backers. Among the biggest beneficiaries is Benchmark, which holds a 9.5% stake in the AI chip maker. General partner Eric Vishria, who joined the Cerebras board in 2016 after co-leading its $25 million Series A round, now oversees a position worth billions.

Yet that windfall almost didn’t happen. Vishria admitted to TechCrunch that he initially attended the founding meeting with deep reluctance. “It was five founders and a deck, and it was our first hardware investment in 10 years,” he recalled. “I had been a venture capitalist for like, 18 months.” At the time, he had just sold his social browser startup RockMelt to Yahoo for a reported $60 million to $70 million.

Benchmark is famously picky, and hardware deals are rare for the firm. Vishria recalls muttering to himself, “Why did I take this meeting?” He even texted his assistant, questioning why she had let him schedule it. But his frustration vanished by the third slide. Co-founder and CEO Andrew Feldman laid out Cerebras’ vision: “The first deck is the title slide. The second deck is the team. And I was like, ‘Oh, that team is really good.’ And the third slide is something along the lines of ‘GPUs actually suck for deep learning. They just happen to be 100 times better than CPUs.’ As soon as he said it, a light bulb went off,” Vishria said. “I was like, ‘Oh, my God, of course. Like, why would a graphics processor be the right thing for AI?’”

This pitch came years before Google’s landmark 2017 Transformer paper, which laid the foundation for modern AI and eventually led to ChatGPT. Cerebras was proposing a massive new chip designed specifically for AI training, a concept the semiconductor world wasn’t ready to manufacture. Vishria was intrigued enough to consult Benchmark partners, who admitted they lacked hardware expertise. They told him he’d need to bring in one of the firm’s original founders from the 1990s, someone who truly understood the space.

Vishria arranged a meeting where Feldman pitched to founding partner Bruce Dunlevie. Dunlevie grilled the CEO on chip packaging, cooling, and other technical details. “Most of that meeting was like a dog watching TV for me,” Vishria joked, given his limited understanding. After the pitch, Dunlevie warned that what Cerebras was attempting would be extremely difficult, noting that others had tried and failed. He believed this team had a shot but worried there might be no market for the chip.

Despite his own technical gaps, Vishria was convinced that if Cerebras “could make AI faster,” a market would emerge, and this experienced team had the ability to deliver. The founders 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. “We hadn’t just fallen off the back of a turnip truck. We were an experienced team.”

Hardware is hard, and the following 8.5 years were a grind. Feldman and co-founder and CTO Sean Lie had to invent new cooling methods to prevent the giant chip from overheating under power. They also built a machine capable of drilling 40 screws into the wafer simultaneously without cracking it. Throughout this ordeal, Vishria often thought, “What are we doing?”

Hardware is also expensive. After raising half a billion dollars from a long list of investors, Cerebras was still developing its chips. It had to raise more capital 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 recalled.

But about 18 months ago, everything changed. Cerebras’ chips, manufactured by TSMC, proved even better for inference , running AI models to generate responses , than for training. Just as that realization hit, the AI world became insatiably hungry for that kind of compute. Cerebras landed a major customer and started generating revenue.

Instead of another private round, the company attempted to go public in 2024 but got stuck in U. S. government scrutiny over national security concerns tied to a large investment from its only major customer, Abu Dhabi-based cloud provider G42. Public investors were also wary of its dependence on G42 and its hefty losses. That delay turned into a blessing. Today, OpenAI and AWS are large customers. Cerebras doubled its revenues and declared a profit last year.

Vishria credits the Cerebras team for “persistence, ingenuity, but also adaptiveness.” The deal is also a major feather in the investor’s cap for finding a winner so far outside Benchmark’s usual comfort zone. Benchmark owned 17,602,983 shares, worth $3.3 billion at the IPO’s opening price of $185 and over $5.3 billion if the first-day trading price of over $300 holds. The firm cannot sell shares until after a six-month lockup expires, a standard restriction preventing insider sales immediately after a public offering.

According to various disclosures and Vishria’s confirmation, Benchmark bought about 80% of those shares in early rounds for roughly $18 million. It purchased the remainder in later, pricier rounds costing around $250 million, as Cerebras disclosed in its S-1. In total, the venerable VC firm spent perhaps $270 million for a stake now worth multiple billions, depending on how the stock price holds.

VC firm employees receive bonuses when investments deliver big returns. As for Vishria’s assistant, the one he gave grief for okaying that first meeting? He laughed and said, “I think she’ll do well, very well.”