Cerebras CEO Andrew Feldman to discuss wafer‑scale AI scaling at TechCrunch Disrupt 2026

Cerebras Systems CEO Andrew Feldman will take the main stage at TechCrunch Disrupt 2026 to argue that wafer‑scale AI processors can keep pace with rising compute, energy, and infrastructure demands, citing a $5.5 billion IPO and a multiyear OpenAI contract for 750 megawatts of hardware. The company also reported more than 600 megawatts of data‑center capacity live or under contract for delivery by the end of 2027.

Key points

• Feldman will present a session titled “Can AI Keep Scaling?” at Disrupt.
• Cerebras raised $5.5 billion in its May 2026 IPO.
• The firm signed a multiyear agreement with OpenAI for 750 MW of systems (2026‑2028).
• Over 600 MW of data‑center capacity is live or under contract for 2027 delivery.
• Cerebras plans to increase manufacturing capacity tenfold in 2026.

What happened

Cerebras Systems announced that its co‑founder and CEO, Andrew Feldman, will appear on the Disrupt stage in San Francisco on October 13‑15. The session, “Can AI Keep Scaling?” will examine the growing demand for compute, energy, and infrastructure, and how Cerebras’ wafer‑scale approach differs from conventional chip designs. Feldman’s appearance is part of a larger program that includes more than 200 sessions, 250 speakers, and 300 exhibiting startups.

The company’s recent financial milestones underscore the stakes. In May 2026 Cerebras completed a $5.5 billion initial public offering, the largest for an AI‑hardware firm to date. In August, it introduced the CS‑4 generation of its wafer‑scale AI infrastructure and disclosed a multiyear agreement with OpenAI to deploy 750 megawatts of Cerebras systems through 2028. The same announcement noted that Cerebras already has over 600 megawatts of data‑center capacity live or under contract for delivery by the end of 2027, and that it will boost manufacturing capacity more than tenfold during 2026.

Cerebras also revealed plans to bring its first European data‑center capacity online this year, targeting 200 megawatts of capacity in Europe by the end of 2027. These expansions aim to match the projected surge in AI workloads that demand ever‑larger compute envelopes.

The backstory

Andrew Feldman co‑founded Cerebras in 2015 after a career that included SeaMicro, an energy‑efficient micro‑server startup acquired by AMD in 2012, and senior roles at Force10 Networks and Riverstone Networks. Cerebras’ core innovation is wafer‑scale computing: rather than slicing a silicon wafer into individual dies, the company fabricates a single processor that spans the entire wafer, eliminating inter‑die communication bottlenecks.

Wafer‑scale designs were long considered impractical because of yield challenges and thermal management. Feldman’s team invested a decade in engineering solutions, culminating in the CS‑4 system that promises higher bandwidth and lower latency for demanding AI models. The broader industry has responded to AI’s exponential growth with larger GPUs and specialized ASICs, but many firms still rely on conventional chip architectures that face physical and power limits.

Scaling the compute wall

The fundamental obstacle to AI growth is the mismatch between model size and the physical resources required to train them. Conventional chips hit diminishing returns as power density rises and cooling becomes prohibitive. Cerebras’ wafer‑scale approach sidesteps some of these limits by integrating thousands of cores on a single substrate, reducing data movement and energy per operation. Feldman’s Disrupt talk is likely to stress that scaling compute without scaling infrastructure—power, cooling, and data‑center space—creates a bottleneck that only a holistic solution can address.

However, wafer‑scale processors also demand bespoke manufacturing lines and larger footprints, which can constrain supply chain flexibility. Cerebras’ tenfold manufacturing expansion in 2026 suggests the company is betting on economies of scale to offset these disadvantages. If the strategy succeeds, customers could achieve higher throughput with fewer physical servers, reducing overall data‑center real estate.

Competitive implications

Cerebras’ aggressive rollout positions it against established GPU vendors and emerging AI‑accelerator startups. The OpenAI contract signals confidence from a leading AI developer, potentially prompting rivals to explore alternative architectures or to double down on modular designs that fit existing data‑center ecosystems. Companies that already own extensive conventional GPU farms may face a decision point: retrofit with wafer‑scale units or continue scaling out traditional hardware.

Investors will watch Cerebras’ ability to meet its 2027 capacity commitments. Failure to deliver could erode trust in wafer‑scale viability, while successful deployment would validate a new path for AI infrastructure and could reshape hardware procurement strategies across the sector.

What builders and businesses should do

• Evaluate whether current GPU‑centric stacks can meet projected AI workloads beyond 2027.
• Conduct a cost‑benefit analysis of integrating wafer‑scale systems versus expanding conventional data‑center capacity.
• Monitor Cerebras’ manufacturing rollout for signs of supply‑chain stability.
• Consider pilot projects with Cerebras’ CS‑4 platform to benchmark performance against existing accelerators.
• Align energy‑budget planning with the higher power density requirements of wafer‑scale hardware.

The bottom line

Cerebras will use TechCrunch Disrupt to argue that wafer‑scale processors can break the current AI‑compute ceiling, backed by a $5.5 bn IPO and a 750 MW OpenAI deal. Watch the rollout of its expanded manufacturing capacity and European data‑center footprint for early signals of market traction.