Cerebras (CBRS) is a publicly traded AI Accelerators company founded in 2016 and based in Sunnyvale, USA.
| Date | Stage | Amount | Valuation | Lead investors |
|---|---|---|---|---|
| May 14, 2026 | IPO | $5.5B | $56.4B | — |
| Feb 1, 2026 | Series H | $1B | $23B |
| Tiger Global Management |
| Sep 1, 2025 | Series G | $1.1B | $8.1B | — |
| Nov 1, 2021 | Series F | $250M | $4B | Alpha Wave Ventures, Abu Dhabi Growth Fund |
| Nov 1, 2019 | Series E | $270M | $2.4B | — |
| Nov 1, 2018 | Series D | $88M | $1B | Altimeter Capital |
| Jan 1, 2017 | Series C | — | — | Vy Capital |
| Dec 1, 2016 | Series B | — | — | Coatue Management |
| May 1, 2016 | Series A | $27M | — | Benchmark, Foundation Capital, Eclipse Ventures |

Motley Fool publishes an opinion comparison of Cerebras and SpaceX as potential 2026 IPO investments.

Cerebras shares fall after its first post-IPO earnings report, though analysts cite strong demand for its AI chips.

Cerebras overclocks its WSE-3 waferscale engine and launches the Nexus CS-4 system for faster AI inference.
The largest chip ever built, fabricated from a single TSMC wafer with roughly 900,000 AI-optimized cores and 4 trillion transistors on one ~46,000mm² die. By keeping an entire model on-chip with enormous on-die memory bandwidth, the WSE-3 sidesteps the multi-GPU communication bottlenecks of clusters. It is the compute heart of Cerebras's systems and the company's core architectural bet for frontier AI.
A data-center appliance that packages a single WSE-3 with power, cooling, and I/O so customers can deploy wafer-scale compute without assembling GPU clusters. CS-3 systems can be linked into clusters (Cerebras's 'Condor Galaxy' supercomputers) to train very large models, and are sold or rented to enterprises, national labs, and partners like G42. It is Cerebras's flagship hardware product.
A high-speed inference cloud service that runs popular open large language models at token-generation rates far exceeding GPU-based offerings, exploiting the WSE's on-chip memory bandwidth. Marketed as the fastest commercial LLM inference available, it lets developers access Cerebras hardware via API. The service is central to Cerebras's strategy of monetizing wafer-scale silicon through cloud inference, not just hardware sales.
8 patents on file, but none with both an extractable figure and an abstract on Google Patents yet.