Cerebras Systems Inc. (CBRS)

Cerebras builds AI infrastructure designed for speed. Its core innovation is the Wafer-Scale Engine (WSE), a wafer-scale processor the company describes as the first and only commercialized one. It says the WSE is 58 times larger than NVIDIA's B200 chip and has 2,625 times more memory bandwidth than the B200 package. The company says Cerebras Inference delivers answers up to 15 times faster than leading GPU-based solutions on leading open-source models. It also says many customers have seen more than 10 times faster training time-to-solution than on leading GPU systems of the same generation. Revenue was $510.0 million in 2025, up 76% from $290.3 million in 2024. The company reported net income of $237.8 million in 2025 and a net loss of $481.6 million in 2024, with a 2025 gross margin of 39%.

The offering has four parts. The hardware platform is the WSE-3, with 900,000 compute cores, 44 gigabytes of on-chip memory and 21 petabytes of memory bandwidth. Each WSE sits in a Cerebras CS-3 system with integrated cooling, power delivery and interconnect, and multiple CS-3 systems connect into AI supercomputers. The co-designed software platform includes the Cerebras Compiler, which compiles PyTorch models directly to the WSE, the Inference Serving Stack, and the Cluster Manager. AI Model Services provide co-development help with model selection, training data preparation, training, fine-tuning and optimized deployments. Customers can buy on-premises supercomputers, use Cerebras Cloud by the token or through dedicated capacity, access inference through partner clouds, or use hybrid deployments. Partner channels include AWS Marketplace, Microsoft Marketplace, IBM watsonx Model Gateway, Vercel AI Gateway, OpenRouter and Hugging Face. The company also operates and manages large clusters for some customers.

Customers and partners include hyperscalers, foundation model labs, AI-native and digital-native businesses, enterprises, and Sovereign AI initiatives. OpenAI selected Cerebras as its fast inference solution. AWS signed a binding term sheet to become the first hyperscaler to deploy Cerebras in its own data centers. Training customers include G42, MBZUAI, GlaxoSmithKline, Sandia National Laboratory and the U.S. Department of Defense. Workloads include real-time coding agents, deep research systems, and voice applications and digital twins. The company says its top ten customers by 2025 revenue increased their aggregate spend by about 80% within 12 months of their first purchase. Citing Bloomberg Intelligence, it estimates its combined addressable market in AI training infrastructure and inference at $251 billion in 2025, growing to $672 billion by 2029, a 28% CAGR. Inference is the faster-growing part of that market.

The text names NVIDIA and its GPU-based systems as the main performance comparison. It does not describe suppliers, manufacturing arrangements or headquarters.