Accelerating Discovery: Andrew Feldman, Co-Founder and CEO, Cerebras Systems

At the end of 2015, Dr. Christopher D. Manning, Professor of Computer Science and Linguistics at Stanford University, said, “Deep learning waves have lapped at the shores of computational linguistics for several years now, but 2015 seems like the year when the full force of the tsunami hit the major Natural Language Processing (NLP) conferences.” A bold statement, especially since Manning essentially acknowledged established methods of an entire field of research had been superseded by a new discovery. However, his assertion was appropriate considering the meteoric rise of deep learning over several years as the result of massive improvements over the current approaches towards difficult problems in artificial intelligence (AI), enormous investments from tech giants, and a new focus and growth in research publications.

Around this time — what is now largely considered the beginning of the modern deep learning era — Andrew Feldman, Gary Lauterbach, Michael James, Sean Lie and Jean-Philippe Fricker, former colleagues at SeaMicro, came together once more to build a new class of computer to accelerate artificial intelligence and deep learning work by orders of magnitude beyond the current state of the art. Their mission this time around? To solve the problem of AI compute by building a full AI accelerator solution, from chip to system to software. And so, Cerebras Systems, the pioneer in high performance AI compute, was born. Over the years, the Cerebras team successfully built the largest chip ever made — one that is nearly 60x larger than the previous largest chip — and has accelerated deep learning, enabling major companies around the world to find answers to critical societal issues in less time.

We recently sat down with Co-Founder and CEO, Andrew Feldman, to learn more about what keeps the 5x entrepreneur coming back for more, the impact Cerebras continues to have on medical research, energy and environmental areas and beyond, his biggest tip for aspiring entrepreneurs, and more.

Eclipse: Tell us a little about yourself and your background.

Andrew: I was born and raised in Palo Alto, California. My parents were professors at Stanford — they came here in the mid-sixties to get their PhDs and never left. I went to Stanford a couple of times, tried to get a PhD at Berkeley, but didn’t have the stomach for it. I have a couple of Master’s and I’ve always been an entrepreneur. Even at a young age, I had a little company that was making surf and ski t-shirts and board shorts. My whole life, I’ve had little projects I’ve been tinkering on.

Eclipse: You said you’ve always been an entrepreneur — what interests you about entrepreneurship? What’s the pull there?

Andrew: That’s a really good question. I think the chance to build something is what’s attractive to me. I’m not a technical visionary, so I build things with groups of people. And so, gathering people together, fueling their work with funding — and sugar and pizza — is really exciting to me. I’m a bit of a change junkie — I love to see change and after weeks or months if there’s no change, frustration wells up inside of me. With good people working hard, you ought to be able to see important change over periods of time. When you look back over six weeks, 12 weeks, three months, a year, you should be awed at the amount of progress you’ve made. That’s really, really rewarding to me.

Eclipse: Talk us through your “aha” moment that led you to start building Cerebras.

Andrew: You know, I don’t think ideas form like children from Zeus’ head, sort of fully formed. When the five founders got together for the first time, we wrote on the whiteboard that we wanted to work together again, as we had all worked together at our previous company. We also wrote that we wanted to do something important — we wanted to move an industry. We weren’t doing this to make money. We weren’t doing this for a hundred reasons. We were doing this to see if we could pursue an idea that landed us in the Computer History Museum and if there was to be a book on 21st century compute, there’d be a chapter on us, not a footnote. That was the goal from the get-go: To solve a huge problem that moved an industry and transformed a society.

Eclipse: Tell us more about Cerebras and what you are building.

Andrew: We built a new type of computer that is optimized for artificial intelligence work. AI, as you know, is one of the most important types of work that computers do and it poses some very unusual challenges to the processor. As a result, some of the AI work gets delayed. It will take weeks or months to test ideas with a typical computer. We invented a new computer that can do that work in hours or days. You can think of it as sort of an AI supercomputer. We’re a system company, so we designed the chip, the processor, and the system — all of which power and feed the processor with I/O and with software that allows users to write in the language of AI, which is PyTorch and TensorFlow.

This technological breakthrough has helped major companies around the world find answers in less time. For example, Argonne National Laboratory found our system 300 times faster than their existing compute infrastructure, enabling them to rapidly conduct cancer research. In a different project, they were able to model Covid-19 in a way that had never been modeled before. At GlaxoSmithKline, they are performing epigenetic research 160 times faster with our system than their existing computing infrastructure. AstraZeneca is doing work in days that had previously taken weeks. The Pittsburgh Supercomputing Center has created a cloud where researchers from around the country can jump on and test ideas and experiments against our system. TotalEnergies, the French energy giant, found our system more than a hundred times faster for a key element in oil exploration.
Eclipse: Wow, talk about impact! What would you say excites you the most about the potential of what you’re building?

Andrew: Well, our first customer used our system to identify cancer drug interactions and their impact on tumor size in ways that have never been done before. That’s why you’re an entrepreneur, right? We have customers trying to gain purchase on the problems of drug design, therapeutic development, the impact of drugs on cancer, efficient ways to harvest clean energy, more efficient oil exploration, and so on. These aren’t small problems in society — these are fundamental problems. Artificial intelligence gives purchase on some of these problems and our system allows AI to quickly and easily do work that previously took a long time.

Eclipse: What do you wish you knew before you started your entrepreneurial journey?

Andrew: This is my fifth big VC-backed project — we sold three companies and took one public. I think this type of deep tech that we specialize in can be a pressure test on your soul. It’s amazing how many things can go wrong before lunch alone and yet, it’ll still turn out to be a good day. What nobody tells you is the internal fortitude needed to build an organization of 500-600 people from scratch, the technical and organizational challenges you will face in deep tech, and the logistical challenges.

Eclipse: How do you summon that mental fortitude to deal with those ups and downs?

Andrew: Have a good partner, have a family. That’s tongue in cheek, but my wife has been with me for three decades now and I’m fortunate to have a supportive family, friends, mentors, and other CEOs I can go to and just say, “This week kicked my butt.” They usually laugh and say, “You get some of those sometimes.” But I think, mostly, it’s a perverse belief that each week is three steps forward, one step back, and a kick in the stomach. As long as your stomach is strong enough, at the end of a year, you’re a hundred steps forward.

Eclipse: What would you say your greatest fear as an entrepreneur is?

Andrew: My greatest fear is to succeed at the ordinary — to do work that isn’t interesting or that isn’t important. To loaf when you could run, to look back over a several year period and say, “Sh*t, I didn’t do my best work. I could have done better. I could have built something cooler.” I think boredom — essentially problems we know the answer to when we start — is a fear that drives many of us. Those problems aren’t interesting at all. We like problems that are so hard, we talk about it as doing fearless engineering. We like problems that are hard and when you solve them, you feel like you’ve gone somewhere — you’ve made progress. In the Valley, interesting problems are far more rare than profitable problems. It’s easier to make money than to find extremely interesting projects.

Eclipse: What is the biggest tip you’d give aspiring entrepreneurs just starting out in their journey?

Andrew: I think there is a duality. You have to be passionate and believe, but at the same time, you have to listen very carefully to what the market is telling you. And, you have to understand when they tell you that you’re going to be wrong. Everybody told us we couldn’t build the chip, but we went out and built a part that was 56 times larger than anything anybody has ever done. Our chip has 2.5 trillion transistors more than any other part made in history. A lot of people told us it couldn’t be done, but we understood exactly where that was coming from because we worked really hard to examine previous failures to produce big chips. We studied these attempts, we interviewed people, we talked to vendors and people who had been involved in these projects. It’s not enough to discount them and say, “Oh, they’re naysayers.” You have to really think about and try to learn from their negativity. You’ve got to be able to be both passionate and have drive behind your belief, while listening carefully to the market. Listen to what people say, but don’t take it at face value — try to understand why they believe the idea won’t work, the problems they think are going to bite you in the backside, and why you believe those problems won’t bite you. Finding that balance between passionate prosecution of your vision and listening and gathering meaningful feedback from the market is a very hard balance to find.

Eclipse: What has been the most exciting moment as an entrepreneur and/or specifically as a Co-Founder of Cerebras?

Andrew: The first time our part worked, we had been working at it for three years. We solved the problem and the founders stood in the lab watching something — that for 70 years and the entire history of the computer industry — nobody had been able to do. There we were in a sh*tty little lab in Los Altos and 85 of us were taking in that we had done something that Intel couldn’t do with 100,000 people. And Nvidia couldn’t do it with 40,000 people. Every other company in history couldn’t accomplish what we had just accomplished. We just stood there in awe. It was awesome.