Funding & Finance

Take the Meeting

Eric Vishria was texting his assistant during the pitch. “Why did you let me take this meeting?” He had

By Conor Healy

May 21, 2026

Eric Vishria was texting his assistant during the pitch. “Why did you let me take this meeting?” He had been a venture capitalist for 18 months. Benchmark hadn’t backed a hardware startup in a decade. The company in front of him was five founders and a slide deck, pitching a new kind of chip for a technology that barely existed yet.

The Cerebras IPO closed this week at a $24 billion valuation. Benchmark’s stake, bought mostly for $18 million in early rounds, is now worth over $5 billion. Vishria has been on the board for ten years.

He almost said no.

The Third Slide

The first slide was the title. The second was the team. Vishria noticed they were good, but he was still annoyed to be there. Then came the third slide.

“GPUs actually suck for deep learning,” said Andrew Feldman, the CEO. “They just happen to be 100 times better than CPUs.”

Vishria’s irritation vanished. “Why would a graphics processor be the right thing for AI?” The question had never occurred to him. Suddenly it seemed obvious.

Cerebras was proposing something nobody else was attempting: a chip designed specifically for AI, built from scratch rather than adapted from graphics processors. The wafer-scale chip would be the largest ever manufactured, covering an entire silicon wafer instead of being cut into smaller pieces. A single Cerebras chip contains 4 trillion transistors. An Nvidia H100, the industry standard, contains 80 billion.

This was 2016. Google’s Transformer paper, the research that led to ChatGPT, was still a year away. The AI boom was theoretical. Feldman was betting on a future that didn’t exist yet.

The Team

Feldman wasn’t a newcomer. He and his co-founder Sean Lie had previously built SeaMicro, a company that redesigned server architecture from first principles. AMD acquired it for $334 million in 2012.

“The advantage of having had a successful exit previously is it erases some of the uncertainty in the venture capitalists’ minds,” Feldman told TechCrunch after the IPO. “We hadn’t just fallen off the back of a turnip truck. We were an experienced team.”

Vishria brought the deal to his partners at Benchmark. They told him they didn’t know enough about hardware. If he wanted this, he needed to bring in one of the original Benchmark founders from the 1990s.

Bruce Dunlevie, a founding partner who understood chips, agreed to hear the pitch. He grilled Feldman on packaging, cooling, manufacturing. “Most of that meeting was like a dog watching TV for me,” Vishria later said. He understood almost none of it.

Dunlevie’s verdict: what Cerebras was attempting would be extremely hard. Others had tried and failed. But this team had a shot.

Vishria still didn’t fully understand the technology. But he understood one thing: if Cerebras could make AI faster, there would be a market for it.

Benchmark led the $25 million Series A.

Eight and a Half Years

What followed was grind. Cerebras had to invent new cooling methods to prevent chips of that size from burning and build machines that could drill 40 screws into a wafer simultaneously without cracking it. They raised half a billion dollars before their chips were even finished.

In 2022, they raised again in a VC bear market with minimal traction. “That was where it got really tough,” Vishria recalls.

Then, around 18 months ago, everything changed. The chips Cerebras had designed for training turned out to be even better for inference, the process of running AI models to generate responses. Just as that realisation hit, the AI industry’s demand for inference compute exploded. Nvidia couldn’t ship GPUs fast enough. Hyperscalers were desperate for alternatives.

In 2024, Cerebras tried to go public. The IPO got stuck in government scrutiny over a large investment from Abu Dhabi-based G42, their biggest customer. G42 had ties to the Chinese military that made US regulators nervous. Public investors didn’t like the customer concentration or the losses.

The delay turned out to be a gift. Cerebras used the time to diversify. By 2026, OpenAI and AWS were customers too. Revenue doubled. The company turned profitable. When they finally went public this week, the concerns that had stalled them were gone.

Cerebras opened at $185 per share. The stock closed its first day above $300.

The Anti-Portfolio

Venture capital is full of meetings that almost didn’t happen. David Cowan of Bessemer Venture Partners famously refused to meet two Stanford students who had rented a garage from his friend Susan Wojcicki. Their names were Larry Page and Sergey Brin. Google is now worth $2 trillion. Bessemer keeps a public “Anti-Portfolio” of their biggest misses as a reminder to stay humble.

Brad Feld of Foundry Group took a call from Fitbit’s CEO during a Colorado snowstorm in 2010. He was so distracted by power outages that his only goal was to get off the phone. He passed. Nine months later, other investors convinced him to look again. He invested. Fitbit went public in 2015. Feld’s stake was worth $1.6 billion.

Seven venture capitalists met with Airbnb founder Brian Chesky in 2008. He was asking for $150,000 in exchange for 10% of the company. Two didn’t bother to reply. The other five said no. “We couldn’t wrap our heads around air mattresses on living room floors as the next hotel room,” Fred Wilson of Union Square Ventures later wrote. That 10% stake would be worth $8 billion today.

Chris Sacca passed on Airbnb because he thought someone would get raped or murdered and the blood would be on his hands. “Airbnb is a $100 billion company I’m not an investor in,” he has said.

The pattern is consistent: the best investments often look wrong at first glance. They fall outside the investor’s expertise. The market doesn’t exist yet. The idea sounds ridiculous.

Vishria’s instinct was to skip the Cerebras meeting. Hardware wasn’t their thing. He had been a VC for less than two years. The smart move was to pass.

He took the meeting anyway.

The Return

Benchmark owns 9.5% of Cerebras. The firm spent roughly $18 million in early rounds and another $250 million in later rounds. Total investment: around $270 million.

At the opening price, their stake was worth $3.3 billion. At the first day’s close, over $5 billion.

Vishria gives credit to the Cerebras team. “Persistence, ingenuity, but also adaptiveness,” he says.

But the investment almost didn’t happen. A junior VC texted his assistant in frustration during the pitch. Hardware wasn’t their thing. He didn’t understand the technology. The smart move was to pass.

He took the meeting anyway. And about that assistant, the one Vishria blamed for scheduling it in the first place? VC employees get bonuses when investments deliver returns like this. “I think she’ll do well,” Vishria said. “Very well.”

The Cerebras IPO is now the biggest tech listing of 2026. It started with a meeting that almost didn’t happen.