Cerebras files for IPO, with AWS and OpenAI deals worth over $10 billion

US startup Cerebras Systems has filed for an IPO, and the listing could be one of the most significant in the AI semiconductor sector in recent years. The company is timing its debut after two strategic deals that reinforced its standing in the industry.

Cerebras was founded in 2016 in Silicon Valley and specializes in chips that accelerate neural networks. Its flagship product is the Wafer Scale Engine (WSE), a chip that occupies an entire silicon wafer. In size and transistor count, WSE far exceeds standard GPUs: the WSE-3 contains 4 trillion transistors, versus 80 billion in Nvidia's flagship H100.

WSE's architecture lets it process data with minimal latency. All computation happens on a single substrate, so data does not have to travel between separate chips over slow interconnects. The company positions its products as an alternative to Nvidia GPU clusters for training and inference of large language models. Instead of thousands of individual GPUs linked by high-speed networks, Cerebras offers a few WSE chips that deliver comparable performance with less communication overhead. The company claims its systems can cut training time for some models by tens of times.

Before filing, Cerebras signed two strategic agreements. The first is with Amazon Web Services: the company's chips will appear in AWS infrastructure, giving the startup access to enterprise customers worldwide through the largest cloud provider's platform. This noticeably shifts the balance of power: until now, AWS has relied mainly on its own Trainium and Inferentia chips, plus Nvidia GPUs. Adding Cerebras to the AWS ecosystem means the company passed a rigorous technical and commercial vetting; such partnerships are rare.

The second agreement is with OpenAI. According to sources, the contract is worth more than 10 billion dollars. Details of the deal are not disclosed, but it may involve supplying compute infrastructure for training and inference of next-generation models. For Cerebras, this is not just a big order but strategic validation: one of the most compute-hungry companies in the world chose its chips. OpenAI has traditionally worked with Nvidia-based Microsoft Azure infrastructure, so diversifying suppliers looks like a significant shift.

The IPO comes amid a wave of excitement around AI infrastructure. Investors are actively looking for ways to share in the sector's growth without betting on one specific model or service. AI semiconductors are one of the few segments with near-guaranteed demand: whichever models win the race, they need chips. Cerebras is positioning itself as the second major player after Nvidia, although questions about manufacturing scalability and the maturity of its software stack remain open.

The main signal for the industry is that the AWS and OpenAI deals show the biggest players are ready to diversify chip suppliers rather than stick with Nvidia alone. With GPU shortages and political risks around Taiwanese manufacturers, this creates a real niche. The key question is whether Cerebras can scale production and support a software stack at the level engineers used to the CUDA ecosystem expect.