Cerebras Systems

The CS-3 is built on a single 46,225 mm² silicon wafer containing 900,000 compute cores and 4 trillion transistors, compared to roughly 80 billion transistors on a single Nvidia H100 GPU. By building the entire accelerator on one wafer rather than dicing it into hundreds of small chips, Cerebras eliminates the inter-chip networking bottlenecks that slow GPU clusters during AI training. The result is a system that can train a 24-trillion-parameter model on a single installation with linear performance scaling.

The UAE-based AI conglomerate G42 accounted for 87% of Cerebras's $249 million in first-half 2024 revenue (per the firm's SEC filing, September 2024). The two firms have a strategic partnership that includes building a nine-system CS-3 cluster in Dallas and joint go-to-market efforts in cloud computing. This concentration makes Cerebras highly sensitive to any slowdown in G42's AI infrastructure spending, a risk the firm itself flagged in its IPO prospectus.

Cerebras has partnered with Mayo Clinic since at least 2022 to train large language models on de-identified patient records and genomic data using the CS-2 and CS-3 systems (per the firm's public communications). The partnership aims to build foundation models that can assist in clinical diagnosis and treatment planning without sending sensitive data outside the hospital's firewalls, a use case Cerebras pitches as a competitive strength of its on-premise appliance model.

Cerebras Cloud, launched in 2023, provides on-demand access to CS-2 and CS-3 compute via third-party hosting partners Cirrascale Cloud Services and G42. Users can select pre-configured model-training environments or reserve bare-metal CS-3 instances for workloads requiring maximum performance. The cloud model gives Cerebras a recurring-revenue stream and allows smaller AI labs to access wafer-scale compute without purchasing a $2 million-plus system.

Cerebras has raised venture funding through at least Series F, with a total reported raise of over $720 million from investors including Benchmark, Foundation Capital, Alpha Wave Global, Coatue, and Altimeter Capital (per The Wall Street Journal, 2024). The firm filed for an IPO in September 2024, aiming to list on the Nasdaq under ticker CBRS. As of the filing date, no public listing had been completed.

Cerebras uses a three-tier go-to-market model: direct sales of CS-3 appliances to large enterprises and government labs, cloud instances sold through Cirrascale and G42, and strategic technology partnerships such as the Mayo Clinic collaboration. The firm's own sales force targets national labs, energy companies, and large hospital systems, while the cloud partners handle smaller-scale AI developers who prefer on-demand access over capital purchases.

There is no publicly disclosed philanthropic foundation associated with Cerebras Systems. The firm describes its healthcare partnerships — notably with Mayo Clinic — as commercial arrangements rather than charitable initiatives, though the underlying genomic AI research has public-health implications that Cerebras highlights in its scientific publications.