Cerebras Systems

Cerebras Systems is an American artificial intelligence company specializing in the design of wafer-scale deep learning accelerators. Founded in 2016 by a team of semiconductor industry veterans, the company is headquartered in Sunnyvale, California. Its primary innovation is the Cerebras WSE, the largest computer chip ever built, which is designed to dramatically accelerate machine learning and high-performance computing workloads. Cerebras has attracted significant investment from prominent venture capital firms and has established partnerships with major entities in the pharmaceutical industry, national laboratories, and the United States Department of Energy.

History

Cerebras Systems was co-founded in 2016 by Andrew Feldman, Gary Lauterbach, Sean Lie, Michael James, and Jean-Philippe Fricker, all of whom had extensive backgrounds at companies like SeaMicro, Advanced Micro Devices, and Broadcom. The company emerged from stealth mode in August 2019 at the Hot Chips symposium, unveiling its first wafer-scale engine. This announcement was a significant event in the semiconductor industry, challenging the conventional multi-chip module approach. In 2021, Cerebras launched its second-generation WSE-2 and the CS-2 system, followed by the even larger WSE-3 in 2024. The company's growth has been supported by substantial funding rounds led by investors like Benchmark Capital and Altimeter Capital.

Technology

The core of Cerebras technology is the wafer-scale engine, a single integrated circuit that uses an entire semiconductor wafer from a foundry like TSMC. This design incorporates hundreds of thousands of AI accelerator cores and offers orders of magnitude more on-chip memory and memory bandwidth than traditional GPU clusters from Nvidia or AMD. The architecture avoids the von Neumann bottleneck by keeping data movement extremely local, which is critical for training large language models and scientific computing. To manage the immense thermal output, Cerebras developed a proprietary liquid cooling system. The software stack includes support for popular machine learning frameworks like PyTorch and TensorFlow.

Products

Cerebras's flagship product is the CS-2 system, a purpose-built AI supercomputer built around the WSE-2 or WSE-3 chip. The system is delivered as a complete appliance, including the wafer-scale engine, its specialized cooling system, and the supporting Cerebras Software Platform. For larger deployments, the company offers the Cerebras CS-2 Condor Galaxy, a networked AI supercomputing fabric that links multiple CS-2 systems. These products are designed to train neural networks significantly faster than clusters of Nvidia A100 or Nvidia H100 GPUs, particularly for models with hundreds of billions of parameters.

Partnerships and applications

Cerebras has formed strategic partnerships to deploy its systems for complex computational problems. In the pharmaceutical industry, it collaborates with AstraZeneca and GlaxoSmithKline for drug discovery and generative AI in bioinformatics. Major national laboratories, including Lawrence Livermore National Laboratory and Argonne National Laboratory, use the technology for scientific simulations and climate research projects funded by the United States Department of Energy. The company is also part of the Condor Galaxy project with G42, an Abu Dhabi-based AI holding company, to build one of the world's largest AI supercomputer networks.

Financials and funding

Cerebras Systems has raised substantial capital through multiple private funding rounds. Early investors included Benchmark Capital and Foundation Capital. A significant Series F round in 2021, led by Altimeter Capital and Alpha Wave Global, valued the company at over $4 billion. Other notable investors in subsequent rounds have been Moore Strategic Ventures and the Abu Dhabi Growth Fund. While the company remains privately held, its high valuation reflects strong investor confidence in its challenge to established players like Nvidia in the burgeoning AI hardware market.