Teahose.
SIGN IN
NEW HERE — WHAT TEAHOSE DOES
We read the entire AI & tech firehose — so you don't have to.
PODPodcastsAll-In, No Priors, Acquired…
NEWNewslettersStratechery, Newcomer…
PAPPapersPhysical AI research
PHProduct Huntdaily launches
VCInvestor ScoutSequoia, a16z, Benchmark…
CLAUDE DISTILLS →
7 reads, 30 sec each — free, 6 AM ET.
+ a live graph of the companies, people & themes underneath.
HOME/SOURCERY/Cerebras CEO: The Truth About IP…
POD
// EPISODE
SOURCERY

Cerebras CEO: The Truth About IPO Wealth, "We Made 1,000 Millionaires"

DATE November 21, 2026SOURCE SOURCERYPARTICIPANTS ANDREW FELDMAN, MOLLY O'SHEA
// KEY TAKEAWAYS6 ITEMS
  1. 01Speed has become the buying criterion in AI inference
  2. 02Hyperscaler validation for Cerebras in six months
  3. 03AI's value delivery is at the very beginning
  4. 04AI as a job creator and small-business enabler
  5. 05AI accelerating basic research and drug discovery
  6. 06Upgrades are the point of maximum fragility; infrastructure is getting more reliable

1. Key Themes

Speed has become the buying criterion in AI inference

Feldman argues the market has flipped from "why does it need to be fast?" to "faster." Cerebras claims 10–15x faster AI compute, and he ties that to a surge in demand and a string of hyperscaler wins.

"I think a year and a half ago, some people were like, well, why do we need it to be fast? And now they're like, I use it every day. It needs to be fast. Faster! Faster!" — Andrew Feldman 00:08:41

"we build the fastest AI compute on earth. Not by a little bit, but 10 or 15x. And so, as the market recognizes how important speed is, the demand for our products is off the chart." — Andrew Feldman 00:08:52

Hyperscaler validation for Cerebras in six months

Beyond OpenAI as an existing customer, Cerebras added the three major cloud providers recently, which signals that fast inference is being embedded into mainstream cloud offerings.

"As you know, we have OpenAI as a customer, but in the last six months, we won AWS and Microsoft and Oracle. Pretty good six months." — Andrew Feldman 00:09:16

"step by step, everyone's like, whoa, this fast inference thing. That's a big deal. Yeah. And we need product for it. We need solutions for it. And we need to embed it in our offerings." — Andrew Feldman 00:09:28

AI's value delivery is at the very beginning

Feldman distinguishes between the heavy deployment of hardware so far and the delivery of actual value, which he thinks is just starting.

"We've talked a lot. There's been a lot of deployment of gear. But in terms of delivering value, we're just at the front end of this." — Andrew Feldman 00:01:18

"I think those who suggest that the impact is going to be little are way off base." — Andrew Feldman 00:02:18

AI as a job creator and small-business enabler

Rather than a job destroyer, Feldman frames AI and cloud infrastructure as engines of construction jobs in the short term and small-business creation over the long term.

"For a year, two years, three years, they create a huge number of high paying construction jobs. But they also generate small business jobs after via the AI. It's the cloud and AI are small business job creators" — Andrew Feldman 00:02:33

"instead, it's empowered you and other creators of various types by the tens of thousands to engage and participate in the economy in a way that was otherwise foreclosed." — Andrew Feldman 00:03:07

AI accelerating basic research and drug discovery

He sees AI compressing the most time-consuming parts of research, which flows from basic into applied research and then into industry.

"AI is really good at that. And I think it is really good at some of the most time-consuming parts of research. And this will sort of launch us forward across the board in basic research, and basic research concatenates into applied research and then into industry." — Andrew Feldman 00:07:06

Upgrades are the point of maximum fragility; infrastructure is getting more reliable

On the recent outages, Feldman says working systems are weakest at the moment of change, and that failures in cloud infrastructure are so rare that they feel outsized.

"I think a working system is at its weakest when you try and change it." — Andrew Feldman 00:03:59

"So I think some of these things are so successful and so reliable that when they fail, we sort of lose our mind." — Andrew Feldman 00:05:34

Managing the psychology of sudden wealth

Cerebras's IPO created over 1,000 millionaires. Feldman's guidance is to avoid reacting to daily stock volatility and to avoid rushing decisions.

"first we made more than a thousand millionaires. That's something I'm really proud of. I think that's our job as leaders is to make our guys wealthy." — Andrew Feldman 00:10:39

"The daily fluctuations are not what they should watch. What we should watch is our sales and our growth, whether we continue to deliver engineering products that are extraordinary and that we're proud of." — Andrew Feldman 00:10:39

Radical transparency as a retention and culture tool

Learned from Mark Leslie, Feldman's open-book approach to his team produced very low attrition and decades-long loyalty.

"while we were private, I would answer any question, any question on our all hands... I asked the question about how much I made, what my salary was, how much stock I had." — Andrew Feldman 00:19:17

"We have really, really low attrition. We have people who've been with me now for 25 years. We have some of the, been with me over four companies, five companies." — Andrew Feldman 00:21:06

The founder's learning algorithm

Founders are defined by their tolerance for the unknown, and success depends on how they learn and how they find mentors who fit their style.

"What is your learning algorithm? How do you deal with uncertainty, right?" — Andrew Feldman 00:18:15


2. Contrarian Perspectives

Employees should sell some stock right after the IPO, even in the company they believe in

Most founders say "stay the course." Feldman tells employees that diversification is rational and is not a vote against the company.

"I think there's no reason for you to have 99% of your wealth in one stock. Right? Professional investors don't do it. And Mr. Employee, there's no reason for you to do it either." — Andrew Feldman 00:12:31

"It's not a statement on the company, it's a statement on your situation as such that it's unwise to have 99% of your wealth or 95% of your wealth in a single company." — Andrew Feldman 00:12:59

Do nothing with a windfall for months

In a culture that rewards action, he argues the right move after a liquidity event is often inaction, parking cash and deciding over 90 days to six months.

"if you get a big cash gain, you don't have to do anything. Right? There's no rush. Right? You can put it in a money market and you can breathe." — Andrew Feldman 00:13:24

"I'm going to make this decision over a 90-day period or over a six-month period, and that will be fine." — Andrew Feldman 00:14:54

The recent infrastructure and airline chaos is not the "new normal"

While others read systemic fragility into the outages, Feldman argues the opposite: these are rare outliers that show how reliable the system usually is.

"I don't think people should think of, of the tough last week as sort of the new normal. I, I, I think that, that that's just a tough week or two." — Andrew Feldman 00:22:18

"Well, I mean, when we used it for 10 years, it never had a problem. Right. So they had a once in 10 year failure." — Andrew Feldman 00:05:34

Open source and closed source models both are needed, and both will improve

Rather than predicting a winner, Feldman says the contest itself is healthy.

"I think the open source models will continue to get better and better. Mm-hmm. I, I think the battle between open source and closed source will continue to heat up. And I think we need both." — Andrew Feldman 00:22:45

Most "hidden" corporate information is boring, so share it all

Counter to the instinct to control information, Feldman argues secrecy creates fear while transparency reveals mundanity.

"most of the stuff people fear is so dead boring that if they actually learned it, they would never think about it again." — Andrew Feldman 00:19:47


3. Companies Identified

Cerebras

Description: AI compute company building wafer-scale systems for fast AI inference; recently IPO'd. Led by Andrew Feldman. Why mentioned: Feldman claims the fastest AI compute on earth, with demand "off the chart" and a hallmark IPO.

"because we build the fastest AI compute on earth. Not by a little bit, but 10 or 15x." — Andrew Feldman 00:08:52 "you did have one of the hallmark IPOs of the year... Cerebras' IPO was amazing." — Molly O'Shea 00:09:28

OpenAI

Description: Maker of ChatGPT; Cerebras customer; named as a possible IPO next year. Why mentioned: Existing Cerebras customer, and ChatGPT as an AI tutor.

"As you know, we have OpenAI as a customer" — Andrew Feldman 00:09:16 "And now ChatGPT can do it. And it can do it really well." — Andrew Feldman 00:01:47

AWS (Amazon Web Services)

Description: Largest cloud provider. Why mentioned: Recently won as a Cerebras customer.

"in the last six months, we won AWS and Microsoft and Oracle." — Andrew Feldman 00:09:16

Microsoft

Description: Cloud and software giant. Why mentioned: Recent Cerebras customer win; also cited for a rare outage.

"in the last six months, we won AWS and Microsoft and Oracle." — Andrew Feldman 00:09:16 "Now, we have had a failure at Microsoft recently." — Andrew Feldman 00:05:34

Oracle

Description: Enterprise software and cloud infrastructure provider. Why mentioned: Recent Cerebras customer win.

"in the last six months, we won AWS and Microsoft and Oracle." — Andrew Feldman 00:09:16

SpaceX

Description: Space launch and satellite company. Why mentioned: Cited as part of the current IPO wave.

"we're seeing that with, you know, SpaceX, Anthropical, IPO soon and, you know, maybe OpenAI next year." — Molly O'Shea 00:09:28

Anthropic

Description: AI lab behind Claude. Why mentioned: Mentioned by Molly as an upcoming IPO candidate.

"we're seeing that with, you know, SpaceX, Anthropical, IPO soon" — Molly O'Shea 00:09:28

GlaxoSmithKline

Description: Leading pharmaceutical company. Why mentioned: Example of a drug maker using AI to develop drugs.

"GlaxoSmithKline, you know, a leading drug maker, their use of AI is extremely interesting in helping them develop drugs in the future." — Andrew Feldman 00:07:34

Veritas

Description: Enterprise storage and data management software company. Why mentioned: Led by Mark Leslie, Feldman's mentor.

"Mark was CEO of Veritas for many years." — Andrew Feldman 00:15:58

AI drug-discovery startups (unnamed)

Description: Startups searching for small molecules in new ways. Why mentioned: Feldman sees them as doing work "that couldn't have been done by four 24-year-olds."

"there are startups that are looking for small molecules, looking for drugs in new and innovative ways that couldn't have been done by four 24-year-olds." — Andrew Feldman 00:08:00


4. People Identified

Mark Leslie

Description: Former longtime CEO of Veritas; mentor to Feldman. Why mentioned: Taught Feldman radical transparency with employees.

"Mark was CEO of Veritas for many years. Mm-hmm. He was an extraordinary CEO and is generally wise." — Andrew Feldman 00:15:58 "Mark Leslie really pushed me to be extremely open with my team and completely transparent." — Andrew Feldman 00:19:17

Pierre Lamond

Description: Legendary venture capitalist, now in his mid-90s; mentor to Feldman. Why mentioned: Continues to hold Feldman accountable on pace and execution.

"Pierre LeMond, who is one of the great venture capitalists in our industry, now at 95 or 96, still calls me and asks me if I'm driving hard enough" — Andrew Feldman 00:16:21 "I learned an enormous amount from him about how to be a CEO and how to be a board member." — Andrew Feldman 00:16:21

Alexander the Great

Description: Ancient king and general. Why mentioned: Example of the tutor model of education that AI now democratizes.

"this is the way Alexander the Great was taught. Right? This is the way the uber wealthy have taught their kids forever." — Andrew Feldman 00:01:47

Gavin

Description: Conference host who will question Feldman on stage (surname not given). Why mentioned: Feldman will speak on whatever he asks.

"Whatever, whatever Gavin asks me." — Andrew Feldman 00:08:18


5. Operating Insights

Pick a "learning algorithm" and a mentor who fits your style

Founders face constant first-time experiences, from board meetings to investment banks to hiring finance leads. Feldman's tactic is to deliberately find mentors and not copy them but extract learning methods.

"finding your way to learn, right? What is your learning algorithm?" — Andrew Feldman 00:18:15

"not do it their way, but their own way... encouraged me to find my own way and my own sort of philosophy on being a leader" — Andrew Feldman 00:16:50

Open-door all-hands to defuse rumor and fear

Answering any question, including compensation and equity, makes the unknown mundane. This builds inclusion and lowers attrition.

"I think by sharing, you remove some of the veil that's over a lot of stuff. And it turns out that stuff is mostly ordinary." — Andrew Feldman 00:20:38

Coach employees on the signal versus the stock price

After going public, redirect the team's attention to controllable inputs: sales, growth, product quality, customer happiness, and hiring.

"You've got to think about what your job is every day, which is how do we build extraordinary product? How do we make our customers happy? How do we hire great people?" — Andrew Feldman 00:11:41

Treat upgrades as the highest-risk moment

Whether you're swapping copper for fiber, reconfiguring routers, or upgrading software, plan for peak fragility during the change.

"you take a working thing that's under control and you tear it down. And that is when a system's at its weakest." — Andrew Feldman 00:04:08

Make wealth creation an explicit leadership job

Feldman frames employee wealth building as a leader's responsibility, not a side effect.

"I think that's our job as leaders is to make our guys wealthy." — Andrew Feldman 00:10:39


6. Overlooked Insights

The speed thesis is a real shift in buyer behavior, not marketing

Feldman's casual remark that sentiment flipped within about 18 months implies fast inference is becoming a procurement requirement. A string of hyperscaler wins within six months (AWS, Microsoft, Oracle) is strong corroborating evidence that large buyers are rewriting roadmaps around latency.

"The first thing is, that's cool. The second thing is, can I get it faster?" — Andrew Feldman 00:08:41

"step by step, everyone's like, whoa, this fast inference thing. That's a big deal." — Andrew Feldman 00:09:28

His retention record is a hidden compounding asset

In passing, Feldman says employees have followed him across four or five companies over 25 years. That kind of loyalty means a repeat-founder team with embedded trust, which is a durable advantage in executing hard hardware programs.

"We have people who've been with me now for 25 years. We have some of the, been with me over four companies, five companies." — Andrew Feldman 00:21:06