At TechCrunch Disrupt 2026, Andrew Feldman, CEO and co-founder of Cerebras Systems, addressed the escalating demands of AI computing and whether AI can continue to scale given constraints in compute, energy, and infrastructure. Cerebras has pioneered wafer-scale computing, a departure from conventional chip architectures, by building processors on entire silicon wafers rather than slicing them into individual chips. This innovation aims to meet the intensive workload of AI models. Cerebras recently introduced its next-generation wafer-scale AI system, CS-4, and has committed to a substantial deployment agreement with OpenAI, securing 750 megawatts of its systems for 2026 through 2028 following its $5.5 billion IPO in May.

Feldman emphasized that growing AI capabilities require more than just faster chips; they necessitate expanded data center capacity along with increased electricity and cooling infrastructure. Cerebras has responded by securing over 600 megawatts of data center capacity, either live or under contract, and plans to increase its manufacturing capacity tenfold in 2026. The company is also expanding geographically, with new data center capacity launching in Europe this year and plans to reach 200 megawatts there by the end of 2027. This illustrates that scaling AI entails a holistic approach, addressing the physical environment needed to support larger and more powerful AI compute loads.

The discussion at Disrupt highlighted the broader industry challenge of balancing technological innovation with sustainable infrastructure growth. Feldman’s experience offers insights into what it will take for AI hardware to keep pace with model demand and where limits may be met if scaling continues along current trajectories. His approach challenges long-standing assumptions about the architecture of AI compute hardware and underscores the need for inventive solutions in building AI systems that can handle future workloads efficiently.

TechCrunch Disrupt 2026 is gathering more than 10,000 industry professionals, 250+ speakers, and hundreds of startups to explore these pressing questions and the future of AI technology. Feldman’s session is part of over 200 activities planned from October 13–15 at Moscone West in San Francisco. Attendees have a chance to hear directly from a founder tackling AI’s biggest bottlenecks and participate in a larger conversation on what investments and innovations are essential to sustain AI’s rapid growth beyond today’s hardware capabilities.