
Lessons from Andrew Feldman
Andrew Feldman is the co-founder and CEO of Cerebras Systems, where he led the development of the Wafer-Scale Engine, the largest computer chip ever built. He built the company around "fearless engineering," a philosophy of taking on the fundamental physics and hardware problems the rest of the industry wrote off as impossible. This profile covers his perspective on hardware design, the physical limits of AI infrastructure, and what it actually takes to build a hardware company. — EE Times podcast, "Andrew Feldman on Brain-Sized AI" (2021).
Part 1: The Wafer-Scale Vision
- On Scale: "Until we came along, the largest chip was about 800 square millimeters and that's about the size of a postage stamp." Cerebras's is 46,000 square millimeters, "the size of a dinner plate." — Reference: EE Times podcast, "Andrew Feldman on Brain-Sized AI" (2021).
- On Initial Skepticism: When Cerebras chose wafer scale, "They told us we were out of our mind. It would never work. They listed reasons why it was impossible." In 2019 they proved it possible. — Reference: No Priors, "The Story Behind Cerebras' $63 Billion IPO" (2026).
- On Seeing It Work: When the wafer finally worked in the summer of 2019, in a makeshift Los Altos office, the team "couldn't speak for half an hour": nobody had been able to do it before. — Reference: No Priors, "The Story Behind Cerebras' $63 Billion IPO" (2026).
- On Physics: After solving cross-die connectivity and yield, the harder problems were physical: the wafer and circuit board expand at different rates under heat, and there was no off-the-shelf package or cooling for a part 50 to 60 times larger than any before. — Reference: ARK Invest, "Cerebras' Wafer Scale Engine AI Chip with CEO Andrew Feldman".
- On GPU Limitations: Cerebras was founded on the bet that AI would put new pressure on memory bandwidth and on the communication structure between processors, the weak points of clusters of small chips. — Reference: Twenty Minute VC, "The AI Chip Wars & The Plan to Break Nvidia's Dominance" (2025).
- On Inference: The GPU "was built for graphics processing." The late-2015 insight was that a clean-sheet design optimized only for AI could do better. — Reference: YouTube video: AI Crypto Minds interview with Andrew Feldman (2026).
- On studying past failures: Before betting on wafer scale, the team studied why earlier attempts died, meeting people from Trilogy, where Gene Amdahl's effort failed, and engineers who had built large military and satellite chips. — Reference: ServeTheHome, "Our Interview with Andrew Feldman CEO of Cerebras Systems" (2020).
- On Memory Bandwidth: On-chip memory "is really the memory that matters." When GPU users say they run out of memory, he argues they have really run out of memory bandwidth. — Reference: ARK Invest, "Cerebras' Wafer Scale Engine AI Chip with CEO Andrew Feldman".
Part 2: Fearless Engineering
- On Problem Selection: "We love solving problems that other people say can't be solved. We love hard problems, you know, we say if you know the answer when you start, it's not an interesting problem." — Reference: EE Times podcast, "Andrew Feldman on Brain-Sized AI" (2021).
- On Definition: Fearless engineering is the journey "to solve problems sort of without a safety net." — Reference: EE Times podcast, "Andrew Feldman on Brain-Sized AI" (2021).
- On Incrementalism: Guard the culture: "we would much rather fail in pursuit of the extraordinary than succeed in the ordinary." — Reference: No Priors, "The Story Behind Cerebras' $63 Billion IPO" (2026).
- On Technical Risk: As companies reach 1,000 to 3,000 people they tend to stop taking the risks they once took, drifting from fearless engineering toward whatever fits the next revision. He calls that "extraordinarily damaging." — Reference: No Priors, "The Story Behind Cerebras' $63 Billion IPO" (2026).
- On Challenging Assumptions: Wafer scale had defeated the industry for 70 years; even Gene Amdahl's Trilogy failed at it. Cerebras had to invent techniques to yield whole wafers as well as or better than makers of much smaller chips. — Reference: Twenty Minute VC, "The AI Chip Wars & The Plan to Break Nvidia's Dominance" (2025).
Part 3: Challenging the Status Quo
- On Competition: He calls himself "a professional David." Cerebras is his fifth startup, and he believes every sale won against Nvidia is brains beating muscle. — Reference: No Priors, "The Story Behind Cerebras' $63 Billion IPO" (2026).
- On Breaking Monopolies: "You're not going to get 15 or 20 times better than the GPU with a minor modification to their architecture." A radical improvement needs a different design. — Reference: No Priors, "The Story Behind Cerebras' $63 Billion IPO" (2026).
- On The AI Bubble: He rejects the bubble narrative: "We're not building ahead of demand. We're building behind demand," with a $25 billion backlog and data centers that can't be built fast enough. — Reference: YouTube video: Andrew Feldman on the future of data centres (May 2026).
- On Historical Context: Unlike railroads or the 1990s fiber-optic build-out, where infrastructure ran "way ahead of demand," AI infrastructure is behind demand. — Reference: YouTube video: Andrew Feldman on the future of data centres (May 2026).
- On Data Sovereignty: The US should help allies build sovereign AI on American infrastructure, so they work in the US ecosystem rather than China's. — Reference: YouTube video: AI Crypto Minds interview with Andrew Feldman (2026).
Part 4: Entrepreneurial Hustle and Work Ethic
- On Work-Life Balance: "This notion that somehow you can achieve greatness, you can build something extraordinary by working 38 hours a week and having a work-life balance that is mind-boggling to me." — Reference: Business Insider, "Cerebras CEO Andrew Feldman says you can't 'build something extraordinary' working 38 hours a week" (2025).
- On Commitment: Challenging a trillion-dollar incumbent takes an "every waking minute" level of commitment. — Reference: Business Insider, "Cerebras CEO Andrew Feldman says you can't 'build something extraordinary' working 38 hours a week" (2025).
- On Elite Performance: Greatness takes an elite athlete's devotion; he points to Jerry Rice running routes the day after winning a Super Bowl. — Reference: Business Insider, "Cerebras CEO Andrew Feldman says you can't 'build something extraordinary' working 38 hours a week" (2025).
- On Hiring: With many openings it is "so easy to settle" and just fill a seat, which he calls "death." He spends part of every day talking to candidates. — Reference: No Priors, "The Story Behind Cerebras' $63 Billion IPO" (2026).
- On Stamina: From mid-2017 to mid-2019 Cerebras couldn't build the chip while spending about $8 million a month, reporting failure at board meetings every six weeks and improving a little with each failure analysis. — Reference: No Priors, "The Story Behind Cerebras' $63 Billion IPO" (2026).
- On Intensity: "Nobody becomes an engineer to do things a little better than the last one." The work attracts people who want to build differently. — Reference: Business Insider, "Cerebras CEO Andrew Feldman says you can't 'build something extraordinary' working 38 hours a week" (2025).
- On Ambition: You have to love the journey: doing something this hard "for the money is a horrible thing" when there are easier ways to get rich. — Reference: No Priors, "The Story Behind Cerebras' $63 Billion IPO" (2026).
Part 5: Infrastructure and Physical Bottlenecks
- On The Bottleneck: Data centers are now the bottleneck: the industry "can't build data centers fast enough to keep up with demand." — Reference: YouTube video: Andrew Feldman on the future of data centres (May 2026).
- On Grid Mismatch: The US has power, but not where data centers can be built near good fiber, and local regulations make access to it hard. — Reference: Twenty Minute VC, "The AI Chip Wars & The Plan to Break Nvidia's Dominance" (2025).
Part 6: Corporate Responsibility and Community
- On Reputation: He says the AI industry has done a poor job of selling data centers to the communities that host them. — Reference: Business Insider, "Cerebras CEO Says Tech Has Done a Terrible Job Selling Data Centers" (2026).
- On Thoughtful Building: "These can be clean, they can make jobs, they can be good for communities… We can do this thoughtfully." — Reference: Business Insider, "Cerebras CEO Says Tech Has Done a Terrible Job Selling Data Centers" (2026).
- On Giving Back: With heavy equipment already on site, builders could add a football field or a school for the town: "We can be good neighbors at very, very low cost." — Reference: Business Insider, "Cerebras CEO Says Tech Has Done a Terrible Job Selling Data Centers" (2026).
- On Energy Efficiency: "AI uses a lot of power. And that means we have an obligation to produce amazing results." He adds that Cerebras chips use a fraction of GPUs' power. — Reference: YouTube video: AI Crypto Minds interview with Andrew Feldman (2026).
Part 8: Lessons from SeaMicro and Past Ventures
- On Identifying Inefficiencies: In 2007 he co-founded SeaMicro to build computers optimized for low power. — Reference: YouTube video: AI Crypto Minds interview with Andrew Feldman (2026).
- On building infrastructure for others: In the 1990s his teams built fast switches without knowing Skype or WhatsApp would follow: "We put our brick in the wall." His grandmother went from $4.95-a-minute calls to Skyping her great-grandchildren for free. — Reference: ServeTheHome, "Our Interview with Andrew Feldman CEO of Cerebras Systems" (2020).
- On aiming for the history books: When the SeaMicro founders regrouped in 2015, they wrote two things on a whiteboard: they wanted to work together, and to build something that would land them in the Computer History Museum. — Reference: Unite.AI, "Andrew Feldman, Co-founder & CEO of Cerebras Systems" interview (2023).
- On Validation: SeaMicro was sold to AMD in 2012 for a little over $357 million, and its core team went on to found Cerebras. — Reference: ARK Invest, "Cerebras' Wafer Scale Engine AI Chip with CEO Andrew Feldman".
- On getting good advice: "The valley is awash in capital. Wise counsel and advice are much rarer." Meet great advisors, "set your ego at the door, and you listen," and admit when you're unsure you're on the right path. — Reference: ServeTheHome, "Our Interview with Andrew Feldman CEO of Cerebras Systems" (2020).
- On why experience mattered: In chips, $60–100 million goes out before a single unit ships, so credibility counts: "It would have been very, very hard if this had been your first time." — Reference: ServeTheHome, "Our Interview with Andrew Feldman CEO of Cerebras Systems" (2020).