Cerebras Systems, a maker of AI inference accelerators, issued second-quarter 2026 results below Wall Street expectations, sending shares down 15% in extended trading. Despite raising full-year guidance, the company's revenue missed forecasts. CEO Andrew Feldman told CNBC that AI demand remains 'through the roof' and that enterprises are willing to pay premium prices for Cerebras' specialty inference chips.
Gross margins are expanding as fast inference commands premium pricing, Feldman explained. The company's performance reflects broader strength in AI accelerator demand, particularly for inference workloads as enterprises move AI models from research labs into production deployments. However, the revenue miss signals execution challenges or slowing booking patterns even amid strong demand signals.
For architects: specialty inference chips (Cerebras, Graphcore, others) are a wedge play competing against Nvidia's dominance in training. Inference margins are fat but order books are tightening. Watch whether premium-priced inference players can sustain their multiple if customers begin to diversify sourcing or if software frameworks commoditize inference across GPUs. This earnings miss amid raised guidance is a yellow flag on demand elasticity at premium price points.