Cerebras secures $25B in orders amid AI chip supply crunch

Cerebras CEO: $25B backlog, wartime-scale AI buildout, inference speed is key

AI buildout hits wartime scale.
Cerebras holds a $25B backlog.

Single data centers beat cities.
Power use is that high.

Supply is the constraint now.
OpenAI, Anthropic, Google, Microsoft, AWS face shortages.
Demand is pre-booked.
Delivery speed decides outcomes.

Not all AI spend pays.
Early AWS showed this mix.
Firms now triage workloads.
Frontier for hard problems.
Cheap models for routine work.

Inference speed is the bottleneck.
Inference equals thinking time.
Faster systems compress months.
Days replace months of thought.
Speed buys research and decisions.

OpenAI and Amazon build chips.
They avoid vendor dependence.
Control supply and pricing power.

Models split into two tracks.
Frontier and open source.
Premium for hard reasoning only.
Most tasks use open models.
Regulated buyers want on‑prem control.
Intelligence sovereignty drives purchases.

Video realism implies world models.
They understand motion and physics.
One stack makes films and controls robots.
Media and physical AI converge.