Chips · Company
Cerebras (NASDAQ: CBRS) builds wafer-scale AI computing systems and the software used to train and serve large models faster than conventional GPU clusters. Its hardware platform is organized around a Wafer Scale Engine (WSE) and packaged into rack-scale or system-scale deployments (for example, CS-2, CS-3, and CS-4) that customers can run either on-premises or through Cerebras Cloud and partner clouds.
Cerebras’ differentiation is primarily architectural: the Wafer-Scale Engine is physically much larger than standard accelerator chips and is coupled with a software stack designed to exploit wafer-scale compute and memory bandwidth. Cerebras’ inference offering is presented as a low-latency, high-throughput alternative to GPU inference, and its Inference Cloud is positioned to be compatible with OpenAI API-style integration for developers. Commercially, Cerebras sells (1) wafer-scale training systems (Cerebras CS systems) and related deployments, and (2) fast inference capability via the Cerebras Inference Cloud API. Cerebras also emphasizes hybrid “disaggregated inference” approaches that combine its wafer-scale engine with external racks/accelerators (notably AMD Helios) and hyperscaler infrastructure (notably AWS, including a strategy referenced as integrating AWS Trainium with Cerebras CS-3 for different stages of inference). As of 2026-09-07, Cerebras is publicly traded after pricing and closing its initial public offering in May 2026, and it continues to expand its product roadmap with CS-4 (announced August 2026) and ongoing cloud/data-center capacity expansion discussed in investor materials.
| Headline | Source | Date |
|---|---|---|
| Ultrafast claims speeds of up to 750 tokens per second powered by CerebrasUltrafast says it can generate up to 750 tokens per second, positioning its model for business products and workflows where faster AI responses provide an advantage. | Updated | |
| Hugging Face and Cerebras bring Gemma 4 to real-time voice AIHugging Face and Cerebras are bringing Gemma 4 to real-time voice AI. | Updated | |
| Qwen3-Coder now supported in Vercel AI GatewayVercel AI Gateway now provides access to Qwen3-Coder from QwenLM through a unified API without requiring separate provider accounts. The integration includes usage and cost tracking, performance optimizations, retries, f… | Updated | |
| Cerebras and Compute Nordic Finland Announce New 165 MW AI Data Centre in Mikkeli, FinlandCerebras announced a new data center development with Compute Nordic Finland, describing 165 MW of AI compute capacity and expansion efforts in Europe. |
| Cerebras and Callosum Partner to Deliver Ultra-Low-Latency Heterogeneous Agentic InferenceCerebras announced a partnership with Callosum positioned around ultra-low-latency heterogeneous inference for agentic AI workflows, referencing Cerebras wafer-scale inference characteristics. |
| Cerebras Unveils CS-4: Up to 30 Times Faster than GPU-based SolutionsCerebras introduced CS-4, describing it as a rack-scale solution built from three Wafer Scale Engine 3 Turbo processors and positioned as delivering up to 30× faster inference versus GPU solutions; the press release also… |
| Cerebras Powers Ultrafast Mode for OpenAI’s GPT-5.6 SolCerebras announced powering an OpenAI model feature (UltraFast Mode) with Cerebras infrastructure, including claims about speed and throughput. |
| Cerebras Systems Fast Inference Cloud Business Nearly Quadruples in Second Quarter 2026Cerebras reported Q2 2026 results, including that core revenue more than doubled to $210 million and cloud business nearly quadrupled year-over-year, alongside operational updates such as 600 MW of data center capacity u… |