Cerebras raises $1.1B as Andrew Feldman bets faster AI inference changes what's possible — not just speed

Summary

Cerebras has closed a $1.1 billion funding round, announced by CEO Andrew Feldman. The company builds inference infrastructure it claims runs 20 to 50 times faster than Nvidia GPUs, and has stacked its own chip design, systems, and cloud on top of that foundation.

Feldman's core argument is that speed isn't a feature — it's a category shift. He draws the analogy directly: when the internet was slow, Netflix mailed DVDs; when it got fast, Netflix became a studio. Faster AI inference doesn't just compress existing workflows, it opens up things that weren't possible before. A deep research task that takes 20 minutes today, or a coding task that takes four minutes, could run in seconds — and that changes what engineers actually do with the remaining time.

The concrete example he points to: Anthropic rebuilt the Claude.ai front end using Claude Code in roughly 20 to 30 hours of AI compute time, a task that would have taken human engineers months. Feldman's contention is that if that same task ran in a minute and a half, the implications stop being about productivity and start being about what companies can become.

On the pace of improvement, Feldman goes beyond a simple cost curve. He frames it as half the price and five times faster roughly every 12 months — a steeper trajectory than Moore's Law suggests on cost alone.

Customers named in the segment include Cognition and Figma, cited as early examples of companies where faster inference is translating into new product possibilities rather than just faster versions of existing ones.

With business described as "ripping" and a billion-dollar round now closed, Cerebras is positioning itself as the infrastructure layer for a world where AI latency is the binding constraint — not model capability.