Cerebras Systems’ Billion-Dollar IPO: How an Unlikely Meeting Led to Massive Returns for Benchmark

Cerebras Systems’ IPO: A Billion-Dollar Triumph Rooted in an Unlikely Meeting

Cerebras Systems’ recent initial public offering (IPO) has emerged as a monumental success, generating billions for the company, its founders, and its major investors. Among the most significant beneficiaries is Benchmark, a venture capital firm that holds a 9.5% stake in Cerebras. Eric Vishria, a general partner at Benchmark, has been a member of Cerebras’ board since 2016, the year the AI chipmaker was founded, having co-led its $25 million Series A funding round.

Interestingly, this lucrative partnership almost didn’t materialize. Vishria, who had been a venture capitalist for approximately 18 months at the time, was initially hesitant to meet with the Cerebras team. Reflecting on that period, he recalled, It was five founders and a deck, and it was our first hardware investment in 10 years. Prior to his venture capital career, Vishria co-founded the social browser startup RockMelt, which Yahoo acquired for a reported $60 to $70 million in 2013.

Benchmark is known for its selective investment approach, particularly in hardware companies. Vishria admitted to questioning his decision to take the meeting, even messaging his assistant to ask, Why did you let me take this meeting? However, his skepticism dissipated by the third slide of the presentation, where co-founder and CEO Andrew Feldman articulated Cerebras’ ambitious vision.

Feldman highlighted a critical insight: GPUs actually suck for deep learning. They just happen to be 100 times better than CPUs. This revelation struck a chord with Vishria, who realized the potential for a purpose-built AI processor. At that time, years before Google’s 2017 Transformer paper laid the groundwork for modern AI, Cerebras was proposing a revolutionary, large-scale chip designed specifically for AI training—a concept that was ahead of its time.

Despite his initial reservations, Vishria’s interest was piqued. He consulted with his Benchmark partners, who acknowledged their limited expertise in hardware. They suggested involving Bruce Dunlevie, one of Benchmark’s founding partners with a background in hardware investments. Dunlevie met with Feldman and delved into technical aspects such as chip packaging and cooling. While he recognized the challenges ahead, Dunlevie believed the Cerebras team had a viable chance of success, though he expressed concerns about the market demand for such a chip.

Vishria, despite not fully grasping the technical details, was convinced that accelerating AI processing would create a substantial market. He was also confident in the team’s capabilities, given their previous success in selling SeaMicro to AMD. Feldman noted, The advantage of having had a successful exit previously is it erases some of the uncertainty in the venture capitalists’ minds.

The Challenges of Hardware Development

The journey that followed was marked by perseverance and innovation. Over 8.5 years, Cerebras faced numerous obstacles in developing its product. Feldman and co-founder and CTO Sean Lie had to devise new cooling methods to prevent their large-scale chip from overheating. They also invented machinery capable of drilling 40 screws into the wafer simultaneously without causing damage. Throughout this period, Vishria often questioned the endeavor, pondering, What are we doing?

Hardware development is notoriously capital-intensive. Cerebras raised half a billion dollars from a diverse group of investors while its chips were still in development. The company faced additional fundraising challenges during the 2022 venture capital downturn. Vishria recalled, You don’t have a lot of traction on the company yet, so yeah, that was where it got really tough.

Approximately 18 months ago, Cerebras experienced a pivotal shift. Their chips, initially designed for AI training and successfully manufactured by TSMC, proved to be even more efficient for inference—running AI models to generate responses. This realization coincided with a surge in demand for such computing capabilities. The company secured significant customers and revenue streams.

In 2024, Cerebras attempted to go public but encountered scrutiny from the U.S. government over national security concerns related to a substantial investment from its primary customer, Abu Dhabi-based cloud provider G42. Public investors were also wary of the company’s dependence on G42 and its significant losses.

This delay turned out to be advantageous. Today, OpenAI and AWS are among Cerebras’ major customers. The company doubled its revenues and reported a profit last year. Vishria attributes this success to the Cerebras team’s persistence, ingenuity, but also adaptiveness.

A Rewarding Investment

For Benchmark, this investment represents a significant achievement, especially considering it ventured outside its typical investment areas. At the IPO’s opening price of $185 per share, Benchmark’s 17,602,983 shares were valued at $3.3 billion. If the stock’s first-day trading price of over $300 holds, this stake could be worth more than $5.3 billion. Due to a standard six-month lockup period, Benchmark cannot sell its shares immediately.

Benchmark acquired approximately 80% of these shares in early funding rounds for around $18 million. The remaining shares were purchased in later rounds at higher valuations, costing around $250 million. In total, the firm invested approximately $270 million for a stake now worth multiple billions, depending on the stock’s performance.

This remarkable outcome underscores the importance of openness to unexpected opportunities and the potential rewards of venturing beyond conventional investment strategies.