Cerebras Systems’ Andrew Feldman on whether AI can keep scaling at TechCrunch Disrupt 2026
TechCrunch TechCrunch Events ● Covered by 2 sources
Cerebras boss Andrew Feldman is bringing his AI-scaling pitch to TechCrunch Disrupt 2026. The hook: bigger models now need more power, data centers, and chips than ever.
Based on reporting by TechCrunch, TechCrunch Events — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
AI models keep getting better, but the bill keeps getting bigger too. More compute. More electricity. More cooling. More factories. That pressure is the subject Andrew Feldman plans to tackle at TechCrunch Disrupt 2026, where the Cerebras Systems CEO and co-founder will talk about “Can AI Keep Scaling?” on the Disrupt Stage.
Cerebras has spent years arguing that the industry does not have to build AI around conventional chips. Feldman co-founded the company in 2015 after earlier stints building infrastructure companies, including SeaMicro, the energy-efficient microserver startup that AMD bought in 2012. He also held leadership roles at Force10 Networks and Riverstone Networks. At Cerebras, the bet was wafer-scale computing: building a processor on the wafer itself instead of slicing the wafer into separate chips.
That idea looked unconventional for a long time. Now it sits right in the middle of the AI boom. Cerebras says it raised $5.5 billion in its May IPO, signed a multiyear agreement with OpenAI to deploy 750 megawatts of Cerebras systems from 2026 through 2028, and introduced CS-4 in August as its latest wafer-scale AI infrastructure.
The company is also talking in infrastructure terms, not just chip terms. In August, Cerebras said it had more than 600 megawatts of data center capacity live or under contract for delivery by the end of 2027, and that manufacturing capacity was going up more than tenfold during 2026. It also plans to bring its first European data center capacity online this year and reach 200 megawatts there by the end of 2027.
That is the real point of Feldman’s session. Faster processors alone do not solve the scaling problem if the power, cooling, and data center buildout cannot keep up. At Disrupt, he’ll argue from a decade of betting on a different hardware path, just as the rest of the industry runs harder into the physical limits behind AI growth. The event runs October 13-15 at Moscone West in San Francisco and includes 200+ sessions, more than 10,000 attendees, 250+ speakers, and 300+ startups.
My take — AI-written commentary, not fact-checked reporting
This is the part of AI nobody can market away: physics still gets a vote. The louder the hype gets, the more the real moat looks like power, factories, and boring old infrastructure. Wafer-scale computing is a spicy bet, but the industry’s bigger problem is that brute force now comes with a utility bill.
Read more about this at: TechCrunch