The Story Behind Cerebras’ $63 Billion IPO with Founder and CEO Andrew Feldman
Andrew Feldman, co-founder and CEO of Cerebras, discusses his company's journey from building wafer-scale AI chips in the mid-2010s to achieving a $63 billion market cap post-IPO. He covers the technical and market challenges of being ahead of demand, the pivotal OpenAI and AWS partnerships, and his views on how speed in AI inference will unlock entirely new business models rather than merely improving existing ones.
Summary
Andrew Feldman joined No Priors to discuss Cerebras' remarkable journey from a contrarian hardware startup to a $63 billion public company. Cerebras builds AI computers optimized for AI workloads, and currently claims to be 15-20x faster than GPUs at inference across all model sizes and architectures. Founded in the mid-2010s, the company was built on the belief that a radically new workload (AI) would require a radically new architecture — leading them to build a wafer-scale chip the size of a dinner plate (46,000 square millimeters), compared to the postage-stamp-sized chips from competitors.
The technical journey was grueling. From mid-2017 to mid-2019, the team could not get the wafer-scale chip to work, burning roughly $8 million per month and facing board meetings every six weeks without a breakthrough. The team drew on iterative failure analysis until finally yielding a working chip in the summer of 2019 — a moment Feldman describes as leaving the team speechless for half an hour.
Despite proving the technology worked, Cerebras faced a prolonged period of market indifference. AI models were not yet smart enough to be widely useful, so inference speed was irrelevant. The company found early customers in supercomputing labs (Argonne, Lawrence Livermore, Sandia) and enterprise sectors like oil & gas and pharma. A pivotal moment came when sovereign entity G42 placed a billion-dollar order, allowing Cerebras to transform its supply chain, scale manufacturing, and battle-test its systems at real-world scale — ultimately positioning the company to meet demand when the market exploded around 2025.
The market inflection arrived when AI models became genuinely useful in everyday work. Feldman argues that once people use AI daily, speed becomes non-negotiable — drawing analogies to the zero market size for slow search or dial-up internet. This triggered explosive demand from companies like Cognition, Cursor, and Lovable, culminating in a $20+ billion deal with OpenAI (signed the night before Thanksgiving and finalized in 24 days) and a deployment agreement with AWS.
On the IPO decision, Feldman frames going public as exchanging professional VC investors for a broader investor class in exchange for reduced cost of capital and enhanced credibility. He notes that while a handful of companies (OpenAI, Anthropic, Databricks) have been able to raise public-market-scale capital privately, most companies benefit from the legitimacy and large-enterprise relationships that come with being public. Cerebras also had the unique positioning of being the only pure-play AI hardware company publicly traded.
Feldman reflects on lessons for founders: maintaining a fearless engineering culture as companies scale, avoiding the trap of settling on hiring to fill seats, holding yourself accountable to your original hypotheses when deciding whether to persist or quit, and loving the journey itself rather than being motivated purely by financial outcomes. He describes himself as a 'professional David' on his fifth startup, always competing against much larger incumbents.
Looking forward, Feldman draws an analogy to Netflix: speed did not just make DVD delivery more efficient — it enabled Netflix to become a movie studio. He believes fast AI inference will similarly unlock entirely new business models rather than simply improving existing ones, with the current wave of replacing coding, design, and SaaS tools being just the beginning of a deeper reorganization of how work is done.
Key Insights
- Feldman argues that being 15-20x faster than GPUs required a fundamentally different architecture (wafer-scale), not an incremental modification — radical improvement demands radical design departure.
- Cerebras was technically ready years before the market was, and Feldman claims the demand explosion only came in 2025 when AI models became smart enough that people used them daily, making speed non-negotiable.
- The G42 billion-dollar sovereign order was the critical bridge that allowed Cerebras to transform its supply chain and battle-test at scale before OpenAI and AWS deals materialized.
- Feldman describes the OpenAI deal — over $20 billion — as going from term sheet to signed master agreement in approximately 24 days, which he attributes to a broader market phenomenon of speed becoming possible when everyone is motivated by urgency.
- Feldman frames the IPO as Cerebras offering public markets something unique: the only pure-play AI hardware investment with 100% of revenue tied to the AI market, with no gaming, graphics, or PC revenue diluting the exposure.
- On when to quit, Feldman argues founders should define their hypotheses upfront and use trusted advisors to hold them accountable to those original conditions, specifically to guard against the 'slippery slope' of always testing 'one more thing.'
- Feldman claims AI coding tools have driven per-engineer token spend at Cerebras from near-zero to $25,000-$30,000 per month in eight months, but argues the productivity gains are highly uneven — a small number of engineers have gone from 10x to 100x while others are still figuring it out.
- Feldman draws a Netflix analogy to argue that fast AI will not merely improve existing workflows but will enable entirely new business models, just as broadband did not improve DVD delivery but instead enabled Netflix to become a studio.
Topics
Full transcript available for MurmurCast members
Sign Up to Access