Original Title: Reflections on a decade with Cerebras
Original Author: Steve Vassallo
Original Translator: Peggy, BlockBeats
Editor's Note: On May 14, Cerebras officially landed on NASDAQ with the stock code CBRS, closing the first day at about 68% higher than the issuance price, becoming one of the most watched AI hardware IPOs since 2026.
This article is written by Cerebras early investor Steve Vassallo, reflecting on his nearly two-decade collaboration with Andrew Feldman from SeaMicro to Cerebras. On the surface, it tells a venture capital story from term sheet to IPO, but in essence, it documents how a cutting-edge hardware company bet on the fundamental reconstruction of AI computing architecture during a time when consensus was not optimistic: from wafer-scale chips, memory bandwidth bottlenecks, to a series of engineering challenges such as power supply, heat dissipation, and electrical continuity, what Cerebras faced was not just a single technical challenge, but a complete reinvention of modern computing systems.
What is most noteworthy is not that Cerebras ultimately created a wafer-scale chip 58 times larger than traditional chips, but that the company chose a direction contrary to industry inertia from the very beginning: while GPUs became the default answer for AI training, it tried to redefine "what a computer born for AI really is." This required not only technical judgment, but also capital patience, and most importantly, a long-term, non-transactional trust relationship between investors and the founding team.
For today's AI hardware competition, Cerebras serves as a reminder to the market that the computing power revolution is not just about stacking more GPUs, but also about reimagining the computing architecture itself.
Here is the original text:
On April 1, 2016, Friday, I sent an email to Andrew Feldman, telling him that I would climb over the fence of his backyard and personally deliver the term sheet for our investment in Cerebras to him.
That day was April Fools' Day, but I was not joking.
Strictly speaking, this was not a standard operation for a venture capital firm. But by then, I had known Andrew for nine years and had been discussing the next company with him for nearly two years. I could not let this deal slip away because of a clause sentence I was still revising on a Saturday afternoon.
I first met Andrew in October 2007. At that time, he and Gary Lauterbach had just founded SeaMicro. I did not invest in that round, but I hit it off with them, especially appreciating their way of thinking from first principles. Since then, I have been following them closely.
Really valuable relationships take time to settle. Truly valuable companies do too. Today, from the outside, Cerebras is a company founded ten years ago, about to go public. But to me, this is a relationship that has lasted nineteen years and has finally reached the moment of ringing the bell.
In August 2019, Andrew and I at the Hot Chips conference held at Stanford. That time, Cerebras released the first generation of the Wafer-Scale Engine.
Deep Connections and Unreasonable Ambitions
When AMD acquired SeaMicro in 2012, I had a premonition: Andrew would not stay in a big company for very long. He possesses a strong will to win and a rebellious spirit. By early 2014, he had begun to seek opportunities to leave, and we started meeting frequently to discuss what could be done next.
At that time, two things were far from consensus: first, AI would indeed become useful; second, GPUs were not the most suitable computing architecture for AI.
On the first issue, many intelligent people I knew had differing opinions. After AlexNet appeared in 2012, some corners of the research community had begun to achieve almost magical results using convolutional neural networks. However, in the broader software industry, AI still lingered between a marketing buzzword and a research project.
The second issue, that is, the hardware problem, had hardly been seriously raised. GPUs had already become the default choice for training neural networks, mainly because researchers had discovered that they were "not as bad" as CPUs. Creating a new computing system specifically designed for AI workloads meant challenging the mainstream architecture used by researchers worldwide at that time.
But Andrew, Gary, and their co-founders Sean, Michael, and JP saw a different direction. They had amassed decades of experience in chip and system domains: Gary's background stemmed from pioneering work in data flow and out-of-order execution in the 1980s; Sean focused on advanced server architecture; Michael was responsible for software and compilers; and JP was deep into hardware engineering. They were an exceptionally rare group of individuals: each of them was outstanding on their own; combined, their capabilities manifested a multiplicative effect. They could envision an entirely new computer.
They believed that if AI truly unleashed its potential, the resulting market size would far exceed the sum of all existing computing forms.
They also recognized the essence of GPUs: it was originally a chip designed for graphics processing, temporarily elevated to an AI training tool on a new battlefield. It undeniably performed better than CPUs in parallel processing, but no one would design an architecture like a GPU from scratch specifically for AI workloads. The real limitation of neural network capabilities was not raw computing power, but memory bandwidth. This meant that the chip they aimed to create would not focus on isolated core matrix multiplication, but rather on how data could flow efficiently throughout the entire computing structure.
Internally, investing in Cerebras was far from a consensus decision. Several of my partners had witnessed the previous round of semiconductor investments lead to almost nothing but losses, and they candidly expressed their concerns. However, in the end, we reached an agreement as a team. That weekend in April 2016, we clearly told Andrew: we wanted to be the first investors to give him a term sheet.
Weeks later, Andrew, Gary, Sean, Michael, and JP moved into our EIR office space on the second floor of 250 Middlefield. I still retain the floor plan drawn by the office manager at that time. On that plan, Cerebras is situated next to a founder from Foundation, just a few doors down from Bhavin Shah, who later founded Moveworks. It was a conducive floor for startup growth.
Cerebras' first headquarters was on the second floor of our old office at 250 Middlefield.
Knowing Which Rules Can Be Bent and Which Must Be Broken
Before Cerebras, the largest chip in computing history was about 840 square millimeters, roughly the size of a postage stamp. The chip made by Cerebras, however, spans 46,000 square millimeters, which is 58 times larger than the former.
Choosing a wafer-scale chip also meant choosing all the accompanying downstream design challenges. In nearly 80 years of computing history, no one has ever truly succeeded in doing this. This also meant that no one had systematically solved these problems: how to power such a massive chip? How to cool it? How to maintain electrical continuity among tens of thousands of connection points?
To achieve wafer-scale computing, Cerebras had to practically reinvent every aspect of modern computing simultaneously: semiconductors, systems, data structures, software, and algorithms. Each direction could itself have formed a startup. Andrew and his team chose to tackle the most complex technical issues first. With their intense and almost tireless efforts, these challenges were pushed forward one after another.
Every six to eight weeks, we would hold a board meeting. They would introduce attempts made since the last meeting: a new system design variant, a new power supply solution, or a thermal management adjustment. By repeatedly confronting systemic challenges from various angles, they developed a hard-won clarity in expression. They would explain where they thought the problems lay and what they planned to try next.
We would pose questions, then delve deep with the team, mobilizing the necessary people, resources, and relationships to help them find new breakthroughs. Six to eight weeks later, when we met again, the story would repeat on another technical issue: yet another frontier boundary to explore. Each solution would expose the next problem that needed solving.
Their first prototype wafer smoked when powered on for the first time. The team referred to it as a "thermal event"—the term usually employed when you don’t want to alarm the board or landlord about a fire.
I was continually calculating the power consumption per square millimeter, partly out of curiosity, and partly because those numbers looked too absurd to be real. So, we brought in engineers from Exponent. This company is a failure analysis firm, and its former name happened to be Failure Analysis. They confirmed that those power consumption figures were indeed as bold as they seemed, and helped us think through a series of solutions that did not require challenging the second law of thermodynamics. After all, that is a law Andrew is smart enough not to argue against.
The discipline of engineering lies in knowing which rules can be broken, which can be bent, and which must be respected. Andrew and his team had a practical judgment on this distinction. They knew when they were challenging convention—this was what they intended to do; and they also knew when they were challenging the laws of physics—something they were not aiming to do.
When you are building frontier technology, failure is inevitable. The only way to navigate through failures is through discipline, perseverance, and most importantly, trust: trust in the mission, trust in each other, and trust in the idea that when the first prototype self-destructs, the next morning you will still return to the lab to start the next round of iteration.
This kind of work has no transactional version. It only has a long-term version: staying in the room amid incomplete solutions and patient explanations. Thus, when it finally succeeds, you will be there to witness it firsthand.
That moment occurred in August 2019. Andrew, Sean, and their team stood in the lab, watching a brand-new computer they had designed run for the first time. To outsiders, it appeared to be doing nothing interesting. According to Andrew, that scene was probably as dull as watching paint dry. But this time was different: prior to this, no bucket of such "paint" had ever truly dried. They stood there watching for 30 minutes, then went back to work.
Who You Build With Is Crucial
Some people choose problems based on what they know can be resolved. Andrew’s criterion for choosing problems is whether he believes they are worth solving. Incremental iterations do not excite him; he yearns for a 1000-fold leap. From day one, he has wanted to position Cerebras as a generational, one-of-a-kind company.
This drive partly comes from his personality. Andrew describes it as a "condition" of a computer architect—being captivated by some idea for decades. But to me, it is more broadly a "condition” of a founder. When he looks at a problem, he first asks himself: can I create something that causes a leap forward? Then he asks: if I succeed, will anyone care? If the answers to both questions are yes, he will invest the next decade of his life into it.
The other part of this drive comes from his upbringing. Andrew grew up surrounded by geniuses, as naturally as most children watch television. His father was a pioneering professor of evolutionary biology who would rotate playing doubles tennis with six people every Sunday. Among those six, three later won Nobel Prizes, and one received a Fields Medal.
According to Andrew, these giants would patiently explain their work in physics, mathematics, and molecular biology in terms a child could understand. This left him with a profound impression of what true intelligence looks like and, as his mother would say, understanding that being smart does not mean you have to be a jerk.
I later came to realize that this is one of Andrew's core traits, as important as his rebellious ambition and his light-seeking instinct for genuinely worthwhile problems. He firmly believes that the most exceptional people he has encountered tend to be exceptionally kind as well.
This belief has shaped how his team has come together to accomplish extraordinarily difficult tasks. The first 30 people hired by Cerebras had all worked with him before; some have been with him since 1996. Today, Cerebras has about 700 employees, of whom about 100 have followed him across multiple companies.
In August 2022, the founding team of Cerebras at the Computer History Museum. From left to right: Sean Lie, Gary Lauterbach, Michael James, JP Fricker, and Andrew Feldman.
Importantly, kindness and competitiveness are not contradictory. Andrew is extremely eager to win. He likes to say that he is a professional version of David, fighting against Goliath. Goliath is slow and always on guard against direct attacks, which leaves room for all other tactics. David's advantage lies in appearing in ways and places where Goliath cannot.
At SeaMicro, Andrew’s largest channel partner in Japan was NetOne. NetOne's main supplier was Cisco, which would entertain partners with private jets and yachts, the value of those assets often exceeding that of most houses in Palo Alto. Andrew's budget was much more modest, so he invited the CEO of NetOne to his backyard for a barbecue. Later, that CEO told him that after decades of doing business with Cisco, he had never been invited to anyone's home. This seemingly small yet incredibly human gesture—something a Goliath would never think to do—solidified their relationship.
From the First Term Sheet to IPO
This morning, Andrew rang the opening bell at NASDAQ. I stood beside him. It has been ten years since everything started in our office at 250 Middlefield, and 2600 miles away.
Today, there are still some rare founders doing what Andrew did back then: drawing diagrams on whiteboards at 3 AM, wrestling with unresolved technical challenges. They share the same strong will to win and possess a rebellious spirit. They are looking for a true partner willing to fight alongside them: someone who will dive in to solve problems when the first prototype fails to power on; and who will stay until it finally runs.
This is exactly the kind of founders I want to support: those who choose worthwhile problems, envision solutions that are a thousand times better than the status quo, and persist through inevitable challenges along the way.
For founders like Andrew, Gary, Sean, Michael, and JP, I would be willing to climb over a backyard fence on a Saturday afternoon to personally deliver the term sheet to them.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
