Breaking

Cerebras CEO Andrew Feldman to Discuss AI Scaling Limits at TechCrunch Disrupt 2026

September 30, 2026 at 10:45 EDT

Infra & Chips
AI Agents

Cerebras Systems CEO and co-founder Andrew Feldman is scheduled to take the Disrupt Stage on October 13 for "Can AI Keep Scaling?," a session on whether compute, energy and infrastructure constraints could slow the next phase of AI growth. The session puts one of the AI chip market's highest-profile hardware challengers at the center of a debate that has moved from model demos to power, latency and data-center economics.

Oct. 13, 2026 - Cerebras Systems

Can AI Keep Scaling?

Andrew Feldman will address the compute, energy and infrastructure constraints behind frontier AI, with Cerebras arguing that wafer-scale inference can change the economics of speed.

30x

CS-4 claimed inference speedup vs. GPU systems

750MW

OpenAI low-latency compute partnership

4,400

tokens per second per user on GPT-OSS-120B

2us

lowest wafer-to-wafer latency claimed for CS-4

CS-4 moves the compute scale

The rack-scale CS-4 is rated at six times the AI compute of a single-wafer CS-3 reference.

125 PFLOPSCS-3

750 PFLOPSCS-4

Why speed is the story

Prompt processing

High-throughput systems handle large inputs and context.

Fast decode

Wafer-scale memory bandwidth accelerates token generation.

Agent loop

More time remains for reasoning, checks and tool calls.

What supporters like

Lower latency can make coding agents, voice interfaces and live copilots feel more useful in production.

What still has to prove out

Benchmarks vary by workload and configuration, and broad deployment depends on customer concentration, power and data-center execution.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.