十年押注Cerebras：「晶圆级AI芯片」如何登上纳斯达克

Original Title: Reflections on a decade with Cerebras

Original Author: Steve Vassallo

Original Translation: Peggy, BlockBeats

Editor's Note: On May 14, Cerebras officially listed on Nasdaq under the ticker CBRS, closing its first day up approximately 68% from its IPO price, making it one of the most closely watched AI hardware IPOs since 2026.

This article is written by Steve Vassallo, an early investor in Cerebras. It recounts his partnership with Andrew Feldman spanning nineteen years, from SeaMicro to Cerebras. On the surface, it tells the venture capital story of a journey from term sheet to IPO, but it actually documents how a frontier hardware company, during a period when consensus was unfavorable, bet on a fundamental restructuring of AI computing architecture: from wafer-scale chips and memory bandwidth bottlenecks to a series of engineering challenges including power delivery, cooling, and electrical continuity. Cerebras faced not just individual technical challenges, but the reinvention of an entire modern computing system.

What's 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 opposite to industry inertia from the very beginning: when GPUs became the default answer for AI training, it sought to redefine "what a computer built for AI should be." This required not only technical judgment and capital patience, but also a long-term, non-transactional relationship of trust between investors and the founding team.

For today's AI hardware competition, Cerebras's significance lies in reminding the market that the computing revolution isn't just about stacking more GPUs; it could also come from reimagining the computing architecture itself.

Below is the original text:

On Friday, April 1, 2016, I sent an email to Andrew Feldman telling him I would climb over his backyard fence and hand-deliver the term sheet for our investment in Cerebras.

It was April Fools' Day, but I wasn't joking.

Strictly speaking, this wasn't standard operating procedure for a venture capital firm. But by then, I had known Andrew for nine years and had been discussing his next company with him for nearly two. I wasn't about to miss this deal over a few clauses being revised on a Saturday afternoon.

I first met Andrew in October 2007. At the time, he and Gary Lauterbach had just founded SeaMicro. I didn't invest in that round, but we hit it off immediately, and I particularly admired their way of thinking from first principles. I kept an eye on them ever since.

Truly valuable relationships take time to mature. The same goes for truly valuable companies. Today, from the outside, Cerebras is a ten-year-old company about to go public. But to me, it's a nineteen-year-long relationship finally reaching the moment of ringing the bell.

August 2019, Andrew and I at the Hot Chips conference on Stanford campus. That time, Cerebras launched the first-generation Wafer-Scale Engine.

Deep Relationships and Unreasonable Ambition

When AMD acquired SeaMicro in 2012, I had a feeling Andrew wouldn't stay long in a big company. He has a strong sense of defiance and a rebellious spirit. By early 2014, he was already looking for an exit, and we began meeting frequently to discuss what he might do next.

At that time, two things were far from being consensus: first, that AI would actually become useful; second, that GPUs were not the ideal computing architecture for AI.

On the first point, many smart people I knew disagreed. After AlexNet emerged in 2012, some corners of the research community were already achieving seemingly magical results with convolutional neural networks. But in the broader software industry, AI was still somewhere between a marketing buzzword and a research project.

The second point, the hardware problem, had barely been seriously raised. GPUs had become the default choice for neural network training, mainly because researchers had accidentally discovered they were "less bad" than CPUs. Building a new computing system specifically for AI workloads meant challenging the mainstream architecture used by researchers worldwide.

But Andrew, Gary, and their co-founders Sean, Michael, and JP saw a different direction. They each had decades of experience in chips and systems: Gary's background came from pioneering work on dataflow and out-of-order execution in the 1980s; Sean focused on advanced server architecture; Michael handled software and compilers; and JP specialized in hardware engineering. They were a rare group: individually, each was outstanding; together, their capabilities had a multiplier effect. They could envision an entirely new kind of computer.

They believed that if AI truly unleashed its potential, the resulting market would far exceed the sum of all existing computing forms.

They also saw the nature of GPUs: originally a chip designed for graphics processing, temporarily promoted as an AI training tool on a new battlefield. It was indeed better than CPUs for parallel processing, but if designed from scratch for AI workloads, no one would design an architecture like a GPU. The real bottleneck limiting neural network capability wasn't raw computing power but memory bandwidth. This meant the chip they needed to create shouldn't optimize matrix multiplication within isolated cores, but rather how data flows efficiently through the entire computing structure.

Within the firm, investing in Cerebras was far from a consensus decision. Several of my partners had witnessed the previous round of semiconductor investments resulting almost entirely in losses, and they expressed their concerns very candidly. But ultimately, we reached an agreement as a team. That weekend in April 2016, we explicitly told Andrew: we wanted to be the first investor to give him a term sheet.

A few weeks later, Andrew, Gary, Sean, Michael, and JP moved into our EIR office space on the second floor at 250 Middlefield. I still have the floor plan drawn by the office manager at the time. On that plan, Cerebras sat next to a founder of Foundation, just a few doors away from Bhavin Shah, who later founded Moveworks. It was a floor conducive to startup growth.

Cerebras's first headquarters, on the second floor of our old office at 250 Middlefield.

Knowing Which Rules to Bend and Which to Break

Before Cerebras, the largest chip in computing history was about 840 square millimeters, roughly the size of a postage stamp. The chip Cerebras made was 46,000 square millimeters, 58 times larger.

Choosing a wafer-scale chip meant accepting all the downstream design challenges that came with it. In nearly 80 years of computing history, no one had ever truly succeeded in doing this. It also meant no one had ever systematically solved these problems: How do you power such a massive chip? How do you cool it? How do you maintain electrical continuity across tens of thousands of connection points?

To achieve wafer-scale computing, Cerebras had to reinvent almost every aspect of modern computing simultaneously: semiconductors, systems, data structures, software, and algorithms. Each direction alone could have been a startup. Andrew and his team chose to tackle the toughest technical problems head-on. With their intense, almost tireless efforts, they pushed through these problems one by one.

Every six to eight weeks, we held a board meeting. They would update us on their attempts since the last meeting: a new system design variant, a new power supply scheme, or a thermal management adjustment. Through repeated face-offs with systemic challenges from every angle, they developed a hard-earned ability to articulate clearly. They would explain where they thought things went wrong and what they planned to try next.

We would ask questions, then delve deeper with the team, mobilizing the people, resources, and connections needed to help them find new breakthroughs. Six to eight weeks later, when we met again, the story would repeat on another technical problem: another frontier to explore. Each solution revealed the next problem that had to be solved.

Their first prototype wafer smoked on its first power-up. The team called it a "thermal event"—the term you use when you don't want to scare the board or the landlord by calling it a fire.

I was constantly calculating power consumption per square millimeter, partly out of curiosity, partly because the numbers seemed too high to be real. So we brought in engineers from Exponent, a failure analysis firm that used to be literally called Failure Analysis. They confirmed that the power numbers were indeed as bold as they looked and helped us think through solutions that didn't require defying the second law of thermodynamics. After all, that was a law Andrew was smart enough not to argue with.

Engineering discipline lies in knowing which rules can be broken, which can be bent, and which must be respected. Andrew and his team have a battle-tested judgment for this distinction. They know when they are challenging convention—which is what they intend to do—and when they are challenging the laws of physics—which is what they don't do.

When you're building frontier technology, failure is inevitable. The only way to get through failure is with discipline, persistence, and most importantly, trust: trust in the mission, trust in each other, and trust in the fact that after the first prototype self-destructs, you'll still be back in the lab the next morning to start the next iteration.

This kind of work has no transactional version. It only has a long-term version: staying in the room through incomplete solutions and patient explanations, so that when it finally succeeds, you are 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 they had designed themselves run for the first time. To an outsider, it seemed to do nothing interesting. According to Andrew, it was probably as boring as watching paint dry. But this time was different: never before had a bucket of this particular "paint" actually dried. They stood there watching for 30 minutes, then went back to work.

Who You Build With Matters

Some people choose problems based on what they know they can solve. Andrew chooses problems based on what he believes is worth solving. Incremental iterations don't excite him; he wants a 1000x leap. From day one, he wanted to build Cerebras into a generational, one-of-a-kind company.

This drive comes partly from his personality. Andrew describes it as a kind of "affliction" of a computer architect—being plagued by an idea for decades. But to me, it's more broadly an "affliction" of a founder. He looks at a problem and first asks himself: Can I create something that leads to a step-change improvement? Then he asks: If I succeed, will anyone care? If the answer to both is yes, he'll devote the next ten years of his life to it.

Another part of this drive comes from his upbringing. Andrew grew up surrounded by geniuses, as naturally as most kids grow up watching TV. His father was a pioneering professor of evolutionary biology who played doubles tennis every Sunday with the same six people. Among those six, three later won Nobel Prizes, and one won a Fields Medal.

According to Andrew, these giants would patiently explain their work in physics, math, and molecular biology in language a child could understand. He formed a deep impression of what true intelligence looks like, while also learning, as his mother said, that being smart doesn't mean you have to be an asshole.

I later realized this is one of Andrew's core traits, as important as his rebellious ambition and his almost phototropic instinct for problems truly worth solving. He firmly believes that the most exceptional people he has encountered are also remarkably kind.

This belief shaped how his team came together to accomplish incredibly difficult things. The first 30 people Cerebras hired had all worked with him before; some had been following him since 1996. Today, Cerebras has about 700 employees, roughly 100 of whom have followed him through multiple companies.

August 2022, the Cerebras founding team 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 mutually exclusive. Andrew is fiercely determined to win. He likes to say he's a professional David, taking on Goliath. Goliath is slow and always bracing for a frontal attack, leaving room for every other tactic. David's advantage lies in appearing in ways and places Goliath cannot.

At SeaMicro, Andrew's largest channel partner in Japan was NetOne. NetOne's primary supplier was Cisco, which would entertain partners with private jets and yachts worth more than most homes in Palo Alto. Andrew's budget was much more modest, so he invited NetOne's CEO to a backyard BBQ. The CEO later told him he had done business with Cisco for decades but had never been invited to anyone's home. This seemingly small, deeply human gesture—something Goliath would never think to do—cemented their relationship.

From the First Term Sheet to the IPO

This morning, Andrew rang the opening bell at Nasdaq. I stood by his side. It's been ten years since it all began in our 250 Middlefield office, and 2,600 miles away.

Today, there are still rare founders doing what Andrew did back then: sketching on a whiteboard at 3 AM, wrestling with unsolved technical problems. They too carry a strong sense of defiance and a rebellious spirit. They are trying to find a partner truly willing to fight alongside them: someone who will dive into the problem with them when the first prototype won't power up, and who will stay until it finally runs.

These are the founders I want to support: those who choose problems worth solving, imagine solutions 1000x better than the status quo, and persistently refine and persevere through the inevitable challenges along the way.

For founders like Andrew, Gary, Sean, Michael, and JP, I would climb over a backyard fence on a Saturday afternoon to hand-deliver a term sheet to them.

Original Link