Cerebras CEO Feldman to discuss AI scaling limits at Disrupt 2026

As artificial intelligence models grow increasingly capable, they demand ever-larger amounts of computing power, energy, and supporting infrastructure. The question facing the industry: how much further can this expansion realistically go?

Cerebras Systems, led by CEO and co-founder Andrew Feldman, has spent the last decade pursuing a different path to AI hardware. Rather than relying on traditional chip architectures, the company has bet on wafer-scale computing—building processors across entire silicon wafers instead of cutting them into separate chips. Today, Cerebras delivers AI compute both through on-premise systems and its own cloud infrastructure.

A decade of unconventional thinking#

Feldman's journey in computing infrastructure spans companies like Force10 Networks and Riverstone Networks before co-founding SeaMicro, an energy-efficient microserver company that AMD acquired in 2012. He founded Cerebras in 2015 with a mission to make wafer-scale computing practical—a goal the industry long dismissed as impractical. The company's specialized architecture targets the extreme demands of modern AI workloads.

Cerebras' approach is gaining momentum. The company raised $5.5 billion in its May IPO and signed a multiyear agreement to deploy 750 megawatts of its systems for OpenAI between 2026 and 2028. This August, the company unveiled its latest generation hardware, the CS-4.

Infrastructure is the real bottleneck#

Scaling AI isn't just about faster processors—it requires entire ecosystems of supporting systems. As of August, Cerebras had secured over 600 megawatts of data center capacity that is either operational or contracted for delivery by the end of 2027. The company is simultaneously expanding manufacturing capacity more than tenfold through 2026, opening its first European data center this year, and planning to grow European capacity to 200 megawatts by end of 2027.

The real constraint on AI advancement increasingly lies not in chip design alone, but in the physical infrastructure needed to power, cool, and house these systems at scale.

What's next at Disrupt#

Feldman will speak at TechCrunch Disrupt 2026, October 13–15 at Moscone West in San Francisco, addressing whether today's hardware approaches can sustain AI's growth trajectory. His insights will span the computational demands ahead, infrastructure requirements, and potential breaking points if conventional systems fail to evolve fast enough.