Cerebras IPO makes Benchmark billions but VC Eric Vishria almost didn’t take the meeting

The Cerebras Systems IPO it hit big on Thursday, making billions for themselves, the founders, and its main investors.

Among the biggest winners is the Benchmark shareholder, who owns 9.5% of the company. One of the firm’s partners, Eric Vishria, has been a member of Cerebras’ board since 2016, the year the AI ​​chip maker was founded, after co-leading a $25 million Series A round.

But these billions only happened at Benchmark because Vishria met the startup almost against his will, he told TechCrunch.

“It was five startups and a train, and it was our first hardware investment in 10 years,” Vishria told TechCrunch about the first meeting. “I became a capitalist for 18 months.” (Prior to becoming a VC, Vishria sold his startup, RockMelt, to Yahoo for $60-$70 million in 2013.)

Benchmark is very selective in the companies it chooses, and it backs hardware companies so often that Vishria was kicking himself for giving Cerebras a chance.

“Why did I take this meeting?” He continued to mumble. At one point, he even texted his assistant, who manages his calendar, and pestered him: “Why did you let me take this meeting?” Vishria remembers.

But his pessimism faded with the third slide, when co-founder and CEO Andrew Feldman laid out big plans for Cerebras.

“The first slide is the title slide, the second slide is the team, and I was like, ‘Oh, this team is great.’ And the third slide is something about ‘GPUs absorb deep learning. They just happen to be 100 times better than CPUs.’ And as soon as he said that, the light bulb went off,” Vishria recalled. “I was like, ‘Oh, my God, of course. Like, why would an image processor be the right thing for AI?’

However, that was years ago Google’s famous Transformer paper – a 2017 study that laid the foundation for modern AI – which eventually led to ChatGPT. Cerebras was launching a new type of super chip, designed for AI training, that the processor world was not yet ready to produce.

Vishria was intrigued enough to consult with Benchmark’s colleagues, who quickly told him that they too were not familiar with the equipment. He said that if he wanted the deal, he would have to bring in one of Benchmark’s founders from the 1990s, who he understood.

Undeterred, Vishria arranged a meeting for Feldman to speak with co-founder Bruce Dunlevie, who advised the founder about chip placement and cooling and more.

“Most of the meeting was like a dog watching TV for me,” said Vishria jokingly, because she didn’t really understand. After the pitch, Dunlevie warned that what Cerebras was trying to do would be difficult. Some have tried and failed. But he thought the team had a shot. However, he worried that there would be no market for the chip.

Although Vishria didn’t understand the technology well, he was convinced that if Cerebras “could develop AI quickly” there would be a market for it, and the team had the chops to succeed, he said. He previously sold his startup, SeaMicro, to AMD.

“The benefit of getting out of the past, is removing some of the uncertainty from the minds of venture capitalists,” Cerebras CEO Andrew Feldman told TechCrunch. “We didn’t just fall into the back of the mustard truck. We were a team with a lot of experience.”

Hardware is hard

What followed was an 8.5 year hiatus as Cerebras battled challenge after challenge to develop its drug.

Feldman and Cerebras co-founder and CTO Sean Lie had to develop new cooling methods to keep a chip of its size from burning up while drawing power. They had to make a machine that could drill 40 screws at a time without breaking. And so on.

An investor at Benchmark repeatedly asked himself, “What are we doing?”

Also, the hardware is expensive. Where the company raised half a billion dollars from a long list of investorshis chips were still being made. It was supposed to be raised again in the 2022 VC market.

“You don’t have much in the company here, so yes, that’s where it got difficult,” Vishria recalls.

But about 18 months ago, everything changed. The Cerebras chips, designed for training and developed by TSMC, the world’s largest hardware manufacturer, were better at simulation – driving AI models to generate solutions, rather than training them first. As this realization hit, the AI ​​world grew with an insatiable thirst for such a model. It had a great customer base and money.

In lieu of another private tour, Cerebras tried to go public in 2024, only to find itself on the brink of a US government investigation into national security concerns triggered by large investments by its only major customer, Abu Dhabi cloud provider G42. Public investors also did not want to trust the G42 and suffered huge losses.

The delay was a blessing in disguise. Today, OpenAI and AWS are big customers, too. Macerebras increased revenue and announced a profit last year.

Vishria credits the Cerebras team with “perseverance, intelligence, and flexibility,” he says.

But this is a feather in the investor’s cap to find a winner so far in the usual comfort zone. Benchmark had 17,602,983 shares worth $3.3 billion at the IPO’s opening price of $185, and more than $5.3 billion if the first day of trading above $300 holds. It can’t sell shares until the six-month lock-up is over — a restriction that prevents insiders from selling after a company goes public.

The company bought about 80% of their shares initially for $18 million, various media reports and Vishria confirmed to TechCrunch. It bought the rest at bargain prices that cost about $250 million, Cerebras was revealed in his S-1.

All told, the venerable VC firm spent perhaps $270 million on the deal, which is worth several billion or more, depending on the stock’s performance.

Employees at a VC firm get bonuses when investments generate high profits – is that Vishria’s assistant, who sympathized with him in organizing that first meeting? He laughed and said, “I think he will do well, very well.”