Jul 13, 2026 · 33m · sourcery
Andrew Feldman on Building a Chip 58x Larger Than Nvidia's · Sourcery with Molly O'Shea
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of Sourcery, host Molly O'Shea interviews Andrew Feldman, Co-Founder and CEO of Cerebras Systems, at the RAISE Summit in Paris. Feldman discusses Cerebras's $20 billion deal with OpenAI, high-speed AI inference, post-IPO leadership lessons, and his optimistic vision for AI's long-term societal benefits.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Molly holds 28% of the talking time here. How this is scored →
Molly as informed peer 3.6 Guest teaching 2.9 Guest disagreement 1.1 Molly pushing back 0.3
speaking balance: gold is Molly, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 20:59 Calling circular ecosystem token deals drug pushing
Feldman uses sharp, aggressive rhetoric to call predatory token subsidies and market locking tactics the work of drug pushers and urges startups to remain independent.
Hardest push from Molly ▶ 14:47 Challenging the near-term timeline for space computeWhen Molly suggests data centers are being deployed in space, Feldman firmly checks the hype by asserting it is at least five years away and talking about it precedes reality.
Biggest teaching moment ▶ 15:55 Explaining the breakdown of traditional operating system abstractionFeldman educates Molly on why historical chip architecture could afford OS separation and why modern AI scale strictly mandates simultaneous hardware-software co-design.
Molly holds their own ▶ 23:00 Synthesizing Alex Karp's sovereign AI stanceMolly demonstrates industry literacy by bringing up Alex Karp's CNBC comments regarding sovereign AI and IP ownership, prompting a substantive discussion on stack strategy.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Molly as informed peer | Guest teaching | Guest disagreement | Molly pushing back | Why |
|---|---|---|---|---|---|---|
| Welcome from the RAISE Summit in Paris | 3 | 1 | 0 | 0 | Molly opens the interview with light banter at the RAISE Summit, noting the industry trend that hardware and semiconductors are back in fashion. Feldman agrees enthusiastically and reflects on the Cambrian explosion of architectures and the history of Versailles. | |
| The OpenAI Deal and the Critical Role of Inference | 3 | 3 | 1 | 0 | Molly asks about Feldman's stage appearance with OpenAI and his social media post about having a chip on his shoulder. Feldman explains the distinction between training and inference and why Cerebras is 20x faster in inference workloads. | |
| Wealth Creation for Employees and Staying Balanced | 3 | 2 | 1 | 1 | Molly asks how Feldman maintains an even mindset given massive employee wealth creation and AI market markups. Feldman gently clarifies that over a thousand employees became millionaires at IPO and outlines his philosophy on prioritizing building over chasing money. | |
| Sponsor Segment: Brex Financial Platform | 4 | 3 | 0 | 0 | After an ad break, Molly cites Feldman's interview with Harry Stebbings about data centers falling behind current AI demand. Feldman elaborates on how data center construction moves at the slow pace of real estate while compute demand has accelerated dramatically. | |
| Innovations in Data Center Stack and Space Compute Realities | 4 | 4 | 2 | 1 | Molly brings up Tony Kim's perspectives on data center stacks and mentions putting data centers in space. Feldman pushes back on near-term space compute hype, grounding the timeline at more than five years out while explaining why wafer-scale chips are uniquely suited for space. | |
| Hardware-Software Co-Design in Frontier AI | 4 | 4 | 1 | 0 | Molly asks how chip design has evolved toward co-design and what misconceptions persist. Feldman explains the historical separation created by OS layers and highlights the difficult engineering compromises required when designing hardware and software simultaneously. | |
| Deciphering Industry Deals and Market Concentration Risks | 4 | 4 | 3 | 1 | Molly inquires about high-profile capacity deals and circular GPU financing in the ecosystem. Feldman is candidly critical of Nvidia using balance-sheet power and free cloud credits like drug pushers to lock in startups and stifle competition. | |
| Sponsor Segment: MongoDB and AssemblyAI | 4 | 3 | 1 | 0 | Following sponsor reads, Molly references Alex Karp's remarks on sovereign AI and full-stack ownership. Feldman offers a nuanced alternative view, arguing that companies should maintain choice and protect proprietary data rather than trying to own every single layer. | |
| Positive Externalities: Cancer, Autonomous Driving, and Education | 4 | 3 | 2 | 0 | Molly prompts Feldman on positive AI externalities, mentioning peptides and Brian Armstrong's New Limit. Feldman pushes back against one-sided AI doomerism, arguing that personalized AI tutoring, cancer elimination, and autonomous driving will drastically reduce preventable deaths. | |
| Mentors, Co-Founders, and Executive Leadership | 3 | 2 | 0 | 0 | Molly asks Feldman about his mentors and sources of executive inspiration. Feldman reflects warmly on board member Pierre Lamond, Veritas CEO Mark Leslie, and his long-time co-founders at Cerebras before closing out the interview. |
Statements from this episode (14)
Feldman: Cerebras to provide OpenAI over $20B in hardware
“We'll be doing more than twenty billion dollars of hardware with, for them over the next several years.”
Feldman: Cerebras AI inference is 20x faster than competitors
“And we're the fastest, not by a little bit, but by 20 x.”
Feldman: Post-IPO inbounds surged to roughly 80 daily non-work requests
“If you get 80 emails a day of different people asking you for something. That's not work. That's just sort of this range of people who want something of one form or another to meet you, your time, for you to present, for you to donate to their cause, for you t…”
Feldman: Cerebras IPO created about 1,000 employee millionaires
“At IPO, it was more. It was about a thousand.”
Feldman: Real estate timelines make data center capacity a major industry limitation
“Data centers had historically moved at sort of the speed of real estate, right? Somebody would decide to build a building and two years later after permits and they poured concrete and the building would go up and two years ago nobody cared about AI. All right…”
Feldman: Cerebras chip is 58x larger but fits standard data center racks
“While we build this sort of super big chip, the chip that's like 58 times larger than any other chip, it goes in a server. It goes in a metal enclosure that's about the size of a fridge for a dorm room, a little fridge, and we put a couple of those in a standa…”
Feldman: OpenAI Partnership Informs Cerebras's Chip Designs for Frontier AI
“One of the advantages of our relationship with open AI is we get to see exactly where the frontier is going and we get a chance to roll that into our designs.”
Feldman: Google TPU Co-Design with Gemini and DeepMind Is a Major Advantage
“One of the advantages Google has is that their TPU can be designed in collaboration with the team building Gemini or the team of DeepMind, and so they can inform their choices back and forth, and that's an enormously powerful thing that is surprisingly relativ…”
Feldman: xAI leased GPUs to Anthropic because Grok lacked usage
“They had available capacity because the Grok model wasn't used very much. So they had these GPUs that were sitting around and that's a bad idea. And so they sold a whole block of them or released a whole block of them to Anthropic.”
Feldman: Nvidia uses investments to pressure clouds and builders against rival chips
“There are ways for NVIDIA to use their balance sheet to limit competition, right? That if they invest in, in a NeoCloud, right? The NeoCloud is less likely to use a non-NVIDIA chip. If they invest in a model builder there's pressure not to use other people's c…”
Feldman warns enterprises and nations against depending entirely on Nvidia
“You shouldn't be dependent on, on NVIDIA. You shouldn't be dependent on one model maker. I think that that rarely works out well, right? What you'd like to be as a nation or as a big company is you'd like to have choices. And I don't know if you need to own al…”
Feldman: Ending cancer deaths for next generation achievable in 25 years
“Look, I think that we have a chance for our children or the next generation, not only to not die from cancer, but to not know anybody who died from cancer. I think that is a real achievable goal in 25 years.”