The Story Behind Cerebras’ $63 Billion IPO with Founder and CEO Andrew Feldman
Cerebras CEO Andrew Feldman discusses the company's journey from a contrarian bet on wafer-scale computing to a $63 billion public company. He details the technical breakthroughs, the challenge of being ahead of the market, and how the recent explosion in AI demand for fast inference validated their architecture, leading to a landmark $20 billion deal with OpenAI.
Cerebras was founded on the contrarian bet that a new AI workload would require a fundamentally new, non-derivative computer architecture. This led to the development of wafer-scale computing—building a single, dinner-plate-sized chip (46,000 square millimeters) instead of relying on clusters of small, postage-stamp-sized GPUs. This approach was initially met with skepticism, with many industry observers deeming it impossible. The company’s core mission is to build computers optimized to accelerate AI workloads, where they have achieved a significant performance advantage, particularly in inference.
The Wafer-Scale Bet and Technical Breakthrough
To achieve a radical 15-20x performance improvement over GPUs, a radically different design was necessary. The journey to prove the wafer-scale architecture was fraught with challenges. Andrew Feldman highlights a brutal two-year period between mid-2017 and mid-2019 where the team struggled to successfully build the chip while spending approximately $8 million per month. Each failure analysis, however, led to incremental improvements until the first wafer-scale chip began working in the summer of 2019. This breakthrough solved a problem that had stumped the computer industry for decades, including an effort by Gene Amdall, one of the pioneers of computing.
Navigating the Market Chasm
Despite the technical success, Cerebras faced a significant market challenge: they had built a blisteringly fast machine when AI was still a novelty, not an integral part of daily workflows. As Feldman notes, “How big is the market for slow search? It’s zero… That’s how big the market for slow inference will be.” But the market had to catch up.
The company navigated this period by following a path common to new computer architectures:
- Supercomputing: They first found traction with customers who prioritize speed above all else and are tolerant of immature software, such as Argon National Labs, Lawrence Livermore, and European supercomputing centers.
- Niche Industries: They expanded into sectors with a history of using large-scale compute, like oil and gas and pharmaceuticals.
- The Strategic Bridge: The turning point was a partnership with G42, a sovereign entity, which placed a billion-dollar order. This deal was transformative, allowing Cerebras to mature its supply chain and, crucially, battle-test its hardware and software at a scale that was impossible to replicate in an internal QA lab. This prepared them for the massive demand to come.
The Tipping Point: The Rise of Fast Inference
Around 2023-2024, AI models became “smart enough to be useful,” shifting from a novelty to a critical tool integrated into daily work. At this point, speed became paramount. The market for fast, responsive AI exploded, and Cerebras, having endured its time ahead of the curve, was perfectly positioned with the right product at the right time. This surge in demand validated their long-held architectural bets.
Strategic Milestones: The OpenAI Deal and IPO
The market shift culminated in a landmark deal with OpenAI, one of the largest in Silicon Valley’s history at over $20 billion. The deal came together with extraordinary speed, moving from initial conversations with Sam Altman in the summer to a signed term sheet the night before Thanksgiving and a finalized master agreement by December 24th—a timeline of roughly four weeks. This rapid execution highlights a new paradigm in the AI industry, where the “art of the possible has been expanded.”
Shortly after, Cerebras went public. Feldman outlines several reasons for this decision:
- Credibility: Public companies often have more legitimacy when doing business with other large enterprises.
- Access to Capital: It provides access to a different, broader class of investors.
- Market Differentiator: It offered the public markets a unique “AI pure play”—a company whose revenue is 100% derived from the AI market, with no legacy business in gaming, graphics, or PCs.
- Corporate Maturity: It marked a transition from “corporate adolescence to corporate adulthood.”
Leadership and Culture: The “Professional David”
Feldman emphasizes the importance of maintaining a “fearless engineering culture” that would rather “fail in pursuit of the extraordinary than succeed in the ordinary.” As the company scales past 800 employees, he focuses on recruiting and avoiding the temptation to simply “put a butt in a seat.”
He describes his role as a “professional David” who thrives on competing against Goliaths like Nvidia. This mindset is crucial for a startup taking on an established incumbent. He advises founders that they must “love the journey” and gain energy from solving problems others deem impossible. On the difficult question of when to persist versus quit, he advises leaders to hold themselves accountable to their own hypotheses and have trusted advisors who can remind them of the goalposts they set for themselves, pulling them “back off that slippery slope.”
The Future Enabled by Speed
Feldman believes the true impact of fast AI will not be in making existing business models incrementally better, but in creating entirely new ones. He draws an analogy to Netflix, which evolved from a DVD delivery service competing with Blockbuster into a global movie studio once internet speeds became fast enough.
“Right now we’re replacing things that everybody can see like coding, design… but once we start sort of fundamentally reorganizing around this, you’re going to see this sort of new business models and fundamental jumps in productivity.”