The Story Behind Cerebras’ $63 Billion

1. Key Themes

The "Speed Doesn't Matter Until It Does" Trap in Hardware

Cerebras built a product that was dramatically faster than competitors years before the market cared. The lesson is that being right too early is nearly indistinguishable from being wrong — and surviving that gap requires deliberate bridge-building.

"We were fast when AI was a novelty. And when it's a novelty, nobody cares that you're fast because it's not being used." 00:03:39

"We solved it and we solved this sort of the hardest problem in the computer industry and nobody cared. Nobody. It was like, you know, the first gen we might have sold a dozen." 00:08:42

Radical Architecture Differentiation Is the Only Path to Radical Performance

Cerebras made the contrarian bet that wafer-scale chips — a chip the size of a dinner plate versus a postage stamp — were necessary to achieve 15-20x improvement over GPUs. Incremental modification of existing architectures was never going to get there. The industry called them crazy, and then they proved it worked.

"To be radically better, right? You can't build something that is a similar architecture, right? You're not going to get 15 or 20 times better than the GPU with a minor modification to their architecture." 00:03:14

"From the beginning, we chose wafer scale, which means we build a 46,000 square millimeter chip, a chip the size of a dinner plate, whereas everybody else is building chips the size of postage stamps. They told us we were out of our mind. It would never work." 00:03:39

Speed Unlocks Entirely New Business Models, Not Just Incremental Improvements

Andrew Feldman's most forward-looking theme is that fast AI inference won't simply make existing workflows a bit more efficient — it will reorganize work entirely and enable business models that don't yet exist, analogous to Netflix going from DVD delivery to movie studio when internet speeds increased.

"When I think about what speed does, it doesn't make the existing business models a little better. Netflix used to deliver DVDs and envelopes and they thought their competition was Blockbuster. And when the internet got fast, they became a movie studio. That's what happens with speed." 00:28:19

"Right now we're replacing things that everybody can see, like coding, design, the SaaS tools. But once we start sort of fundamentally reorganizing around this, you're going to see new business models and fundamental jumps in productivity." 00:29:32

2. Contrarian Perspectives

The Market for Slow AI Inference Will Be Zero

Most people think of speed as a nice-to-have feature. Feldman argues it is existential — and history proves this. Once something becomes essential in daily work, slowness eliminates the entire addressable market. This reframes inference speed not as a competitive differentiator but as a category-defining threshold.

"How big is the market for slow search? It's zero. How big is the market for dial-up internet? It's zero. That's how big the market for slow inference will be." 00:04:09

New Workloads Never Go to the Incumbents

The contrarian insight here is that incumbent chip makers — Intel, AMD — never win when new compute workloads emerge. This is not obvious because incumbents have capital, distribution, and talent. But architecture lock-in and organizational inertia make them structurally incapable of winning new paradigms.

"When graphics emerged, you got the discrete GPU and you got NVIDIA. And when the mobile compute hit, you got ARM. And it was interesting that not Intel, not AMD, not all sorts of people who you would have thought have been really well positioned to win in that business, they all got no share." 00:06:26

Going Public Is Just Exchanging One Class of Investor for Another — Nothing More Mystical

The conventional wisdom is that going public is a major strategic milestone or liquidity event. Feldman strips away the mythology and frames it simply as replacing specialized tech investors (VCs) with generalist public investors, thereby reducing cost of capital — with added regulatory burden as the tradeoff.

"Going public is exchanging some professional investors, venture capitalists who specialize in technology investing for a different class of investors. And in so doing, reducing your cost of capital a little bit, right? This is really what's happening. Suddenly, we go from pros like you to my dad." 00:20:21

The Art of the Possible Has Been Dramatically Underestimated

Conventional wisdom says complex deals take months, data centers take years, and acquisitions of large companies take quarters. Feldman argues those were "truncated aspirations," not physical constraints. When pushed by market demand, all of these timelines compress dramatically.

"You can't do a deal like this in 24 days. But if you work on it every day for eight or 10 hours a day, you can. And I think the art of the possible has been expanded by this push in a way I'd never have expected." 00:25:38

The Best Engineers Will Go from 10x to 100x, But AI Coding Is Not Universal

The popular narrative is that AI coding tools make everyone dramatically more productive. Feldman pushes back: the uplift is extremely uneven and concentrated among a small number of engineers who have specifically restructured how they think and work. Most people are "limping along."

"There are some people who have sort of the perfect mindset for it. They're running eight or ten agents, 7x24. They've moved their coding style to being one in which they govern agents... They've gone from being sort of 10x guys to being 100x guys. I think the rest of us, myself included, we're sort of limping along." 00:13:17

3. Companies Identified

Cerebras Systems

AI hardware company building wafer-scale chips (46,000 mm² vs. standard postage-stamp-sized chips) optimized for AI inference. Currently 15-20x faster than GPUs at inference. Recently went public at ~$63B market cap, signed a $20B+ deal with OpenAI, and an AWS deployment agreement.

"We're the fastest at inference, not by a little bit, but by a lot. 15, 18, 20X faster than GPUs." 00:01:28

G42

UAE-based AI and technology holding company led by CEO Peng and Chairman Sheikh Tahnoun. Became Cerebras' strategic bridge customer, placing a billion-dollar order that allowed Cerebras to transform its supply chain, scale manufacturing, and battle-test clusters at scale before winning OpenAI and AWS.

"We won a sovereign, G42. And they became a strategic partner and close friends. And they placed a billion dollar order on us. And with that, we were able to sort of transform the company." 00:09:49

OpenAI

Sam Altman's AI research company. Signed a $20B+ deal with Cerebras in late 2024 after recognizing that inference speed had become critical to keeping up with user demand.

"He said, we've been trying so hard just to keep up with demand. And we now see the importance of fast inference. That produced a set of trials and some testing that was done. And we were so much faster than the competition." 00:23:41

Cursor

AI-powered coding tool. Cited as an example of a company experiencing extraordinary growth enabled by fast AI inference, demonstrating the demand wave that Cerebras is riding.

"The guys at Cursor and Cognition, you see sort of growth we've never seen before." 00:26:00

Cognition (Devin)

AI software engineering company. Specifically cited for the quality of experience running on Cerebras hardware, and as an example of unprecedented growth in the AI tools space.

"Devin on Cerebras is a really magical experience. It's cool. Coding on Cerebras is like high performance at massive speed is really special." 00:26:25

4. People Identified

Andrew Feldman

Co-founder and CEO of Cerebras Systems. Serial entrepreneur on his fifth startup, with a background in computer architecture. Spent 8+ years building toward a market that didn't yet exist, surviving a two-year period of burning $8M/month without a working product, then scaling to a $20B+ backlog.

"I'm a professional David. This is my fifth startup. I compete against Goliath. That is what I do for a living." 00:17:29

Sam Altman

CEO of OpenAI. Cited for his role in initiating the OpenAI-Cerebras partnership and for the remarkable speed at which his organization executes large deals.

"I spoke to Sam in sort of middle of the summer in 25. And he said for the first time, he said, we've been trying so hard just to keep up with demand. And we now see the importance of fast inference." 00:23:41

Gene Amdahl

Legendary computer scientist, considered one of the founding fathers of modern computing. Referenced as the cautionary example — even he failed to build a wafer-scale product, which illustrates the magnitude of what Cerebras accomplished.

"Gene Amdahl, sort of one of the fathers of our field, one of the guys on Mount Rushmore of compute, failed miserably to do it." 00:07:14

Peng (CEO, G42) and Sheikh Tahnoun (Chairman, G42)

Leadership of UAE-based G42. Identified as exceptional strategic partners who took an early billion-dollar bet on Cerebras and worked collaboratively through the product development phase.

"This is Peng, who's CEO, G42, and his chairman, Sheikh Tahnoun. And we couldn't ask for better partners." 00:10:46

5. Operating Insights

Use Sovereign/Government Customers as a Strategic Bridge Across the Chasm

Feldman describes a deliberate sequencing strategy: win supercomputing labs first (they love speed and tolerate immature software), then oil & gas and pharma, then land one large sovereign order to fund the operational transformation needed to serve hyperscalers. This is a repeatable playbook for deep tech hardware companies.

"Often you begin in the supercomputer world because those guys love speed and they don't care if your software is immature... But then historically there's this giant chasm because none of them provide the volume to get to mainstream. And we won a sovereign, G42... They placed a billion dollar order on us. And with that, we were able to sort of transform the company." 00:09:20

Protect "Fearless Culture" as the Company Scales as Aggressively as Any Other Business Priority

Feldman identifies cultural risk — the drift from fearless engineering to playing it safe — as the primary threat as a company grows from hundreds to thousands of employees. He treats cultural preservation as an operational discipline, not a soft concept.

"One of the malaise of companies as they get to 1,000 to 2,000, 3,000 people is they stop taking the type of risks that they were taking before. You move from being a fearless engineering culture to sort of being, what can we get in in the time frame of the next rev? And I think that's extraordinarily damaging." 00:14:44

Hiring Standards Must Be Defended Actively — Never Settle to Fill Seats

Feldman spends a meaningful portion of every day personally involved in recruiting, and frames the temptation to lower hiring bars as "death" to the company. At 800+ people, he explicitly calls this out as a daily risk.

"Recruiting, right? You have so many openings and it's so easy to settle. And it's so easy to just try and put a butt in the seat. Yeah, pretty good. Let's get that butt in the seat. I mean, that is death." 00:15:14

6. Overlooked Insights

The Compiler Took 10 Years — Software Stack Timeline Is a Vastly Underestimated Moat in AI Hardware

This was mentioned almost in passing, but it's strategically enormous. Feldman's co-founder told him it would take 10 years to build a compiler. Feldman thought 5. It took 10. This means any new AI hardware entrant today faces a decade-long software stack build — and Cerebras now sits on a completed, battle-tested compiler that represents an almost insurmountable head start for competitors.

"One of my co-founders said, Andrew, it's going to take about 10 years to build a compiler. I said, no, that's crazy. That's big company talk. We can do it in five. It takes about 10 years. It takes a long time to build a compiler. It is an extraordinarily difficult piece of software. And now we've got a good software stack." 00:12:04

This is not just a technical footnote — it's a durable competitive moat hiding in plain sight. Any investor evaluating AI chip competitors should pressure-test their software stack timeline, not just their silicon roadmap.

Per-Engineer Token Spend Is a Leading Indicator of AI Adoption Velocity

Feldman casually dropped a data point that functions as a real-time productivity gauge: internal token spend per engineer went from effectively zero to $25,000-$30,000 in eight months. This is an operational metric that almost no companies are tracking or disclosing publicly — and it reveals both the pace of AI tool adoption and which companies are genuinely integrating AI versus paying lip service to it.

"Eight months ago, we weren't spending $1,000 an engineer on tokens. And we're probably at $25,000 or $30,000 right now. And it's ripping." 00:12:48

For investors, asking portfolio companies for this metric — token spend per engineer per month — could be a surprisingly high-signal indicator of true AI integration depth and future productivity leverage.