← Home
Odd Lots · May 21, 2026 · 51m

Why Cerebras CEO Andrew Feldman Built The World's Largest Computer Chip

Andrew Feldman, CEO and founder of Cerebras, discusses why his company built chips 58 times larger than average—about the size of a dinner plate. The massive form factor enables blazing-fast AI inference. Feldman covers the history of GPU design, competition between open and closed-source AI models, Cerebras's manufacturing relationship with TSMC, and what the company's IPO means for the AI hardware landscape.

This summary was generated from show notes and public descriptions, not from a full transcript review. Details may contain inaccuracies.

Highlights

Larger Chips Enable Faster AI Inference
Cerebras built chips 58 times larger than industry standard—dinner-plate sized—because size directly enables faster inference speeds for AI queries.
GPU History and AI Hardware Competition
Feldman discusses how GPU architecture evolved and how Cerebras's approach differs from the GPU-dominated AI chip market.
Open vs. Closed-Source AI Model Economics
Feldman weighs in on competitive dynamics between proprietary AI models (like GPT) and open-source alternatives, and how this affects hardware demand.
Cerebras-TSMC Manufacturing Partnership
Feldman discusses how Cerebras manufactures its massive chips through Taiwan Semiconductor Manufacturing Company (TSMC).

Editorial

Cerebras IPO and AI Hardware Market Timing
Feldman appeared on Odd Lots during the week of Cerebras's IPO, reflecting investor appetite for AI chip makers.

Misc

Cerebras chips are dinner-plate sized—a radical departure from miniaturization orthodoxy in chip design
58x larger than average chips, yet this size is the entire business premise
Was this useful?