Cerebras Systems has launched a new tool for AI developers that utilizes its large-scale chips to run applications, presenting a cost-effective alternative to industry leader Nvidia’s processors. The California-based startup’s latest offering aims to address challenges in accessing and running AI models, such as those used in applications like OpenAI’s ChatGPT, which typically rely on Nvidia graphics processing units (GPUs) that are often expensive and difficult to obtain through cloud providers.
According to CEO Andrew Feldman, Cerebras delivers unparalleled performance, surpassing what can be achieved by GPUs, while maintaining high accuracy and offering significantly lower prices. This positions Cerebras to capitalize on the rapidly growing AI inference market, expected to reach tens of billions of dollars as consumer and business adoption of AI tools increases.
Cerebras plans to make its inference tool available via a developer key and its own cloud platform. Additionally, the company will sell its AI systems to customers who wish to operate their own data centers. Feldman highlighted that Cerebras’ Wafer Scale Engines—chips the size of a dinner plate—overcome a common issue in AI data processing, where data from large models often requires multiple smaller chips. By enabling all the data to fit on a single large chip, Cerebras aims to deliver faster and more efficient AI processing.
The company also plans to offer a competitive pricing model, charging as low as 10 cents per million tokens, a unit of measurement for the amount of output data produced by large AI models. In a strategic move for future growth, Cerebras has filed a confidential prospectus with the Securities and Exchange Commission, indicating plans to go public.
