A venture capitalist's reluctance to take a meeting with an unknown hardware startup in 2016 nearly cost his firm one of the most lucrative investments in AI history. Benchmark partner Eric Vishria almost didn't show up to hear Cerebras Systems pitch its vision for a revolutionary AI chip. But after the founders explained why graphics processors were fundamentally wrong for deep learning, Vishria's skepticism evaporated. That single insight led to a $25 million Series A investment that has now grown into a stake worth over $5 billion following Cerebras' blockbuster initial public offering (IPO).

Why Did a Top VC Almost Skip This Meeting?

Vishria had been a venture capitalist for only about 18 months when Cerebras founders approached Benchmark in 2016. The firm was known for being selective, and hardware investments were rare for them. "It was five founders and a deck, and it was our first hardware investment in 10 years," Vishria recalled. He was so unenthusiastic about the meeting that he even messaged his calendar assistant asking why she had booked it in the first place.

The turning point came on the third slide of the presentation. Co-founder and CEO Andrew Feldman laid out a simple but powerful argument: "GPUs actually suck for deep learning. They just happen to be 100 times better than CPUs." That observation sparked something in Vishria. "I was like, 'Oh, my God, of course. Like, why would a graphics processor be the right thing for AI?'" he said. The insight was so obvious once stated, yet nobody had seriously challenged the industry's reliance on graphics processors for AI training.

This was years before Google's 2017 Transformer research paper, which would eventually lay the groundwork for modern AI systems like ChatGPT. Cerebras was proposing something radical: a giant-sized chip specifically designed for AI training, one that the semiconductor industry wasn't yet equipped to manufacture. Vishria brought the opportunity to other Benchmark partners, but they admitted they lacked the hardware expertise to evaluate it properly. They insisted he recruit one of Benchmark's founding partners from the 1990s, Bruce Dunlevie, who actually understood chip design.

How Did Hardware Expertise Seal the Deal?

Dunlevie grilled Feldman about technical challenges like chip packaging and cooling methods. "Most of that meeting was like a dog watching TV for me," Vishria joked, because he understood so little of the technical discussion. But Dunlevie came away cautiously optimistic. He warned that what Cerebras was attempting would be extremely difficult, and others had tried and failed. However, he believed this team had the experience and capability to pull it off. His main concern was whether a market would actually exist for such a specialized chip.

Vishria's confidence rested on a simpler premise: if Cerebras could make AI faster, there would be a market for it. The team had already proven itself by selling a previous startup called 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 explained. "We hadn't just fallen off the back of a turnip truck. We were an experienced team".

Benchmark committed $25 million to Cerebras' Series A round in 2016, with Vishria joining the board. But the real test was about to begin.

What Challenges Did Cerebras Face During Eight Years of Development?

What followed was a grueling journey that tested Vishria's conviction repeatedly. Hardware development is notoriously difficult and capital-intensive. Cerebras faced a series of seemingly insurmountable obstacles:

• Thermal Management: Feldman and co-founder Sean Lie had to invent entirely new cooling methods to prevent a chip of that size from overheating when drawing significant power.
• Manufacturing Precision: The team had to develop a machine capable of drilling 40 screws into the wafer simultaneously without cracking it, a manufacturing challenge that didn't previously exist.
• Capital Requirements: Hardware development required continuous funding. When the company raised $500 million from investors, its chips were still in development, forcing another fundraising round during the brutal 2022 venture capital bear market.

During this period, Vishria repeatedly questioned the investment. "You don't have a lot of traction on the company yet, so yeah, that was where it got really tough," he recalled about the 2022 fundraising environment. The company had spent years developing technology with no revenue to show for it, and public investors were skeptical about backing a hardware company with massive losses.

When Did the Breakthrough Finally Arrive?

Around 18 months before the IPO, everything changed. Cerebras' chips, which were designed for training AI models and successfully manufactured by TSMC (the world's largest contract chip manufacturer), turned out to be even better for inference, the process of running already-trained AI models to generate responses. Just as this realization hit, the AI world became insatiably hungry for computing power. Suddenly, Cerebras had real customers and revenue.

The company initially attempted to go public in 2024, but the process stalled due to U.S. national security concerns. Its largest customer at the time was G42, an Abu Dhabi-based cloud provider, which triggered government scrutiny. Public investors also worried about the company's dependence on a single customer combined with its significant losses. The delay proved fortunate. By the time Cerebras finally went public in May 2026, the customer base had diversified dramatically. OpenAI and Amazon Web Services (AWS) had become major customers alongside G42. The company had also doubled its revenue and declared a profit in the previous year.

How Did Benchmark's Investment Pay Off?

Benchmark's early commitment of approximately $18 million in Series A shares, combined with roughly $250 million invested in later rounds, totaled about $270 million. At the IPO's opening price of $185 per share, Benchmark's 17.6 million shares were worth $3.3 billion. When the stock climbed above $300 on the first day of trading, the stake exceeded $5.3 billion.

The firm cannot sell shares until after a six-month lockup period expires, a standard restriction that prevents insiders from immediately cashing out after a company goes public. But the potential returns are staggering. Benchmark bought roughly 80 percent of its shares in early rounds for around $18 million, while purchasing the remainder at higher prices in later funding rounds.

"Persistence, ingenuity, but also adaptiveness," said Eric Vishria, describing what made the Cerebras team successful.

Eric Vishria, General Partner at Benchmark

Vishria credits the Cerebras founders with the success, but the investment also represents a significant win for Benchmark's willingness to venture outside its comfort zone. The firm had made its first major hardware investment in a decade, backed by a venture capitalist who had been in the business for less than two years at the time. As for Vishria's calendar assistant, the one he blamed for booking that fateful first meeting? "I think she'll do well, very well," Vishria laughed.