Editor's Note: On May 14th, Cerebras officially debuted on the Nasdaq with the stock code CBRS. The closing price on the first day surged by about 68% from the IPO price, making it one of the most watched AI hardware IPOs since 2026.
This article, written by Cerebras' early investor Steve Vassallo, looks back on his 19-year partnership with Andrew Feldman from SeaMicro to Cerebras. While on the surface it tells the story of a venture capital journey from a term sheet to an IPO, it actually chronicles how a cutting-edge hardware company, in an era when consensus was against it, bet on the fundamental reconstruction of AI computing architecture: from chip-level designs, memory bandwidth bottlenecks, to power delivery, heat dissipation, and electrical continuity, the challenges Cerebras faced were not just singular technical hurdles, but a reinvention of an entire modern computing system.
Most noteworthy is not that Cerebras ultimately created a wafer-scale chip 58 times larger than traditional chips, but that this company chose a direction contrary to industry inertia from the outset: as GPUs became the default answer for AI training, it sought to redefine "what truly is a computer made for AI." This required both technical foresight, enduring capital, 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 compute revolution is not just about stacking more GPUs but may also stem from a reimagining of the computing architecture itself.
The following is the original article:
On April 1, 2016, Friday, I sent an email to Andrew Feldman, informing him that I would hop over his backyard fence and personally deliver the term sheet for our investment in Cerebras into his hands.
It was April Fools' Day, but I wasn't joking.
Strictly speaking, this was not a standard procedure for a venture capital firm. However, by that time, I had known Andrew for nine years and had been discussing the next company with him for almost two years. I couldn't let the deal slip away just because of a clause that was still being revised on a Saturday afternoon.
The first time I met Andrew was in October 2007. At that time, he and Gary Lauterbach had just founded SeaMicro. I did not invest in that round, but I had a great connection with them, especially appreciating their first-principles thinking approach. Since then, I have been following them.
Truly valuable relationships require time to mature. Valuable companies are the same. Today, from the outside, Cerebras is a company that has been around for ten years and is about to go public. But in my view, this is a relationship that has lasted for nineteen years, finally reaching the moment of ringing the bell.
Deep Relationship, and Unreasonable Ambition
When AMD acquired SeaMicro in 2012, I had a feeling that Andrew would not stay in a large company for too long. He had a strong sense of perseverance and a rebellious spirit. By early 2014, he had begun looking for opportunities to leave, and we started meeting frequently to discuss what the next steps might be.
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.
Regarding the first issue, many smart people I knew also had different opinions. After AlexNet appeared in 2012, some corners of the research community had already begun to achieve almost magical results using convolutional neural networks. But in the broader software industry, AI still hovered between marketing buzzwords and research projects.
The second issue, the hardware issue, had hardly been seriously raised. GPUs had become the default choice for neural network training mainly because researchers found that, compared to CPUs, they were "not as bad." Building a new computing system specifically for AI workloads meant challenging the mainstream architecture that researchers around the world were using at that time.
However, Andrew, Gary, and their co-founders Sean, Michael, JP saw a different direction. They each had decades of experience in the chip and system fields: Gary's background dated back to the 1980s with pioneering work in data flow and out-of-order execution; Sean focused on advanced server architecture; Michael was in charge of software and compilers; JP delved into hardware engineering. They were an extremely rare group of individuals: individually outstanding, but together, their capabilities exhibited a multiplier effect. They could envision an entirely new computer.
They believe that if AI truly unleashes its potential, the resulting market size will far exceed the sum of all existing forms of computation.
They also saw the essence of the GPU: it was originally a chip designed for graphics processing, only temporarily promoted to an AI training tool in the new battlefield. It is indeed better at parallel processing than a CPU, but if one were to design a chip from scratch for AI workloads, no one would design an architecture like the GPU. What truly limits the capability of neural networks is not raw computational power, but memory bandwidth. This means that the chip they want to create will optimize not for matrix multiplication in isolated cores, but for how data flows efficiently throughout the entire computational structure.
Within the company, investing in Cerebras was far from a consensus decision. Several of my partners had witnessed firsthand that the previous round of semiconductor investments had almost only resulted in losses, and they expressed their concerns very frankly. But in the end, we as a team reached a consensus. On a weekend in April 2016, we explicitly told Andrew: we want to be the first investors to give him a term sheet.
Several weeks later, Andrew, Gary, Sean, Michael, and JP moved into our EIR office space on the second floor of 250 Middlefield. To this day, I still have the floor plan drawn by the office manager at that time. On that drawing, Cerebras sat next to a founder of Foundation, with only a few doors separating them from Bhavin Shah, who would later found Moveworks. It was a floor very suitable for the growth of startup companies.
Knowing Which Rules to Bend and Which Rules to Break
Prior to Cerebras, the largest chip in computing history was approximately 840 square millimeters, about the size of a postage stamp. The chip made by Cerebras, however, has an area of 46,000 square millimeters, 58 times larger than the former.
Opting for a wafer-scale chip also meant choosing all the downstream design challenges that came with it. In the nearly 80 years of computing history, no one has truly accomplished this feat before. It also means that no one has ever systematically addressed these issues: how to power such a huge chip? How to cool it? How to maintain electrical continuity among tens of thousands of connection points?
In order to achieve wafer-scale computing, Cerebras had to almost reinvent every aspect of modern computing simultaneously: semiconductor, system, data structure, software, and algorithm. Each direction could have been a startup on its own. Andrew and his team chose to tackle the most challenging technical problems first. Through their intense and almost tireless efforts, these problems were tackled one by one.
Every six to eight weeks, we would have a board meeting. They would walk us through the attempts since the last meeting: a new system design variant, a new power delivery scheme, or a thermal management adjustment. Due to repeatedly confronting systemic challenges from various angles, they developed a hard-earned clarity of expression. They would explain where they believed the issue was and what they were planning to try next.
We would ask questions and then dive deep with the team, mobilizing the necessary people, resources, and relationships to help them find a new breakthrough. Six to eight weeks later, when we met again, the story would replay on another technical problem: yet another frontier to explore. Each solution would expose the next problem that needed to be solved.
Their first prototype wafer smoked on its first power-up. The team referred to it as a "thermal event"—a term usually used when you don't want to scare the board or the landlord, typically used to refer to a fire.
I was constantly calculating the power consumption per square millimeter, partly out of curiosity and partly because those numbers seemed unbelievably high. 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 numbers were indeed as bold as they appeared and helped us think through a series of solutions that did not challenge the second law of thermodynamics. After all, that was a law Andrew was smart enough not to argue with.
The discipline of an engineer lies in knowing which rules can be broken, which rules can be bent, and which rules must be obeyed. Andrew and his team have a well-honed judgment on this distinction through practical experience. They know when they are challenging conventions—which is what they set out to do—and also know when they are challenging physical laws—which is not what they intend to do.
When you are building cutting-edge technology, failure is inevitable. The only way through failure is discipline, perseverance, and most importantly, trust: trust in the mission, trust in each other, and trust in one thing—that after the first prototype self-immolates, you will still show up in the lab the next morning for the next round of iteration.
This work does not have a transactional version. It only has a long-term version: to stay in the room in those still incomplete solutions and patiently explanations. This way, when it finally succeeds, you will be there to witness it firsthand.
That moment came in August 2019. Andrew, Sean, and their team stood in the lab, watching a brand-new computer designed by their own hands boot up for the first time. To the layperson, it didn't appear to be doing anything interesting on the surface. According to Andrew, the scene was probably as boring as watching paint dry. But what made this time different was that, prior to this, no bucket of "paint" had truly dried. They stood there together for 30 minutes, then went back to work.
Building with Whom Is Key
Some people choose problems based on what they know they can solve. Andrew chose problems based on what he believed was worth solving. Incremental iteration did not excite him; he wanted a 1000x leap. From day one, he aimed to build Cerebras into an intergenerational, one-of-a-kind company.
This drive, in part, comes from his nature. Andrew describes it as a "pathology" of a computer architect—trapped by an idea for decades. But in my view, this is more broadly a founder's "pathology". He looks at a problem, asking himself first: Can I create something that would make a quantum leap improvement? Then he asks: If I succeed, will anyone care? If both answers are yes, he devotes the next decade of his life to it.
Another part of this drive comes from his upbringing. Andrew grew up in an environment surrounded by geniuses, as naturally as most kids grew up watching TV. His father was a pioneering evolutionary biology professor, who played doubles tennis every Sunday with a rotation of six people. Among these six, three later won Nobel Prizes, and one won a Fields Medal.
As Andrew tells it, these giants patiently explained their work in physics, math, and molecular biology in terms a child could understand. This left a deep impression on him: what true intelligence looks like; while also understanding, as his mother put it, that being smart doesn't mean you have to be an asshole.
Later on, I gradually realized that this was one of Andrew's most core qualities, as important as his rebellious ambition and his nearly instinctive sense of what problems are truly worth solving. He firmly believes that the most exceptional people he has encountered are also exceptionally kind.
This belief shaped how his team came together to accomplish incredibly difficult things. The first 30 people Cerebras recruited had all worked with him before; some had been with him since 1996. Today, Cerebras has around 700 employees, with about 100 who have followed him through multiple companies.
Importantly, kindness and competitiveness are not contradictory. Andrew is fiercely competitive. He likes to say he is a professional version of David, taking on Goliath. Goliath moves slowly and is always guarding against a frontal attack, leaving room for all other plays. David's advantage is to appear where and how Goliath cannot.
During his time at SeaMicro, Andrew's largest channel partner in Japan was NetOne. NetOne's primary supplier was Cisco, and Cisco would entertain partners with private planes and yachts, assets worth more than most houses in Palo Alto. Andrew's budget was much more modest, so he invited NetOne's CEO for a backyard barbecue. Later, the CEO told him that despite doing business with Cisco for decades, he had never been invited to anyone's home. This seemingly small but very personal gesture—an action a Goliath would never think to take—cemented their relationship.
From First Term Sheet to IPO
This morning, Andrew rang the opening bell at Nasdaq. I stood by his side. It's been a decade since everything started in our 250 Middlefield office, 2600 miles away.
Today, there are still some rare founders doing what Andrew did back then: drawing on whiteboards at 3 am, grappling with unsolved technical challenges. They carry the same strong sense of resilience and rebellious spirit. They are trying to find a true partner willing to stand shoulder to shoulder: when the first prototype won't power on, someone who will dive in with them to solve the problem; and who will stay until it finally boots up.
This is exactly the kind of founder I want to support: those who choose a problem worth solving, envision a solution that is 1000 times better than the status quo, and persist through the inevitable challenges on the journey.
For founders like Andrew, Gary, Sean, Michael, and JP, I would be willing to hop over a backyard fence on a Saturday afternoon and hand-deliver the term sheet to them.
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
