Cerebras: Inference & Compute

Cerebras Systems is a pioneering AI hardware company founded in 2015 by Andrew Feldman, Gary Lauterbach, Michael James, Sean Lie, and Jean-Philippe Fricker, who previously worked together at SeaMicro (sold to AMD for USD 334 million in 2012). The company revolutionized AI computing with its Wafer-Scale Engine (WSE), the world's largest chip that uses an entire wafer instead of cutting it into individual chips. The CS-3 system contains 4 trillion transistors across 900,000 AI cores with 44GB of on-chip SRAM, delivering 21 petabytes per second of memory bandwidth—7,000× more than NVIDIA's H100.

Cerebras offers both hardware systems and cloud inference services. The CS-3 hardware system is priced at approximately USD 2-3 million per unit, targeting large enterprises, research institutions, and well-funded AI labs. For more accessible options, Cerebras provides cloud-based inference with competitive rates: a Developer Tier at USD 0.10-0.60 per million tokens depending on model choice, making cutting-edge AI accessible without massive capital investments. Cloud training on CS-2 systems is available at USD 60,000 per week or USD 1.65 million per year.

Cerebras' wafer-scale architecture delivers 10-70× faster inference speeds than GPU-based solutions and achieved 210× speedup over NVIDIA H100 in carbon capture simulations. The on-wafer interconnect bypasses latency bottlenecks of multi-GPU setups, enabling simpler programming models and handling huge models without typical GPU memory constraints. While manufacturing yields and high costs present challenges, Cerebras' breakthrough technology addresses fundamental bottlenecks in AI computing, positioning it as a serious challenger to NVIDIA's dominance in the AI accelerator market.