Cerebras: Wafer-Scale Power for AI
DIP Market Fund Platform is Managed by Neovision Wealth Management Limited, which is authorised and regulated by the Financial Services Regulatory Authority ('FSRA') of the ADGM and only available to potential investors that meet the criteria for Professional Clients as defined by the FSRA. Any other person should not act or rely on any information on this website. This is not an offer to the public. Investing in private markets is speculative and involves significant risks, including illiquidity and the potential for a complete loss of capital. For informational purposes only. Interested parties should seek independent financial advice. Offering documents which outline the full terms, conditions, and risk factors, are available to Professional Clients upon request from Neovision Wealth Management Limited.
One-Pager
Short information
Presenation
Full information
🔓 Register* to get full access
About Company
Cerebras builds AI compute systems around a wafer-scale processor (WSE-3) and turnkey CS-3 hardware, and also offers hosted capacity via Cerebras Cloud / Condor Galaxy. The company was founded in 2016.
Key highlights
Hardware: CS-3 system powered by the WSE-3 wafer-scale chip (≈46,000 mm², ~4T transistors); designed to host very large models on a single system.
Cloud: Condor Galaxy network provides hosted training/inference capacity; available as a fully managed service.
Inference API: documented throughput on open models (e.g., ~450 tokens/s on Llama-70B; ~1,800 tokens/s on Llama-8B).
Software: CSoft SDK and open-model suite for fine-tuning and deployment on CS-3 or cloud.
Notable users/partners: deployments with G42 and U.S. national labs (e.g., Sandia, NETL); inference available via Hugging Face.
Market
Segment: AI infrastructure for model training and inference, delivered as on-prem appliances or hosted capacity.
Sunnyvale
USA
2015
founded
A. Feldman, G. Lauterbach
Founders
* Necessary checks are conducted to determine whether the client qualifies as a professional client, as defined by the FSRA Rulebook of ADGM.