OpenAI has entered a major computing infrastructure agreement worth over $10 billion with AI hardware startup Cerebras Systems, a move that marks one of the company’s most significant efforts yet to diversify its chip supply chain and secure capacity for AI model training and inference. Under the multiyear deal, OpenAI will purchase up to 750 megawatts of computing power from Cerebras through 2028, underscoring the surging demand for specialised AI processing as generative models scale rapidly.
Unlike traditional graphics processing units (GPUs) typically supplied by industry giants such as Nvidia and AMD, Cerebras’s architecture is built around “wafer-scale” engines — exceptionally large chips designed specifically for deep learning tasks. These chips, which can dramatically accelerate AI model inference and training, are central to OpenAI’s strategy of matching the right computing systems to the workloads they run, according to industry analysts.
The deal represents both a technical and strategic shift for OpenAI. For years, the company has relied heavily on Nvidia’s GPU ecosystem, including deploying tens of gigawatts of GPU power under partnerships that have sometimes involved joint investment and procurement agreements. However, the rapidly increasing need for real-time AI computing — particularly for inference workloads that power responses to user queries — has encouraged OpenAI to pursue alternatives that can deliver low-latency performance and greater efficiency.
Cerebras, valued at around $8.1 billion and backed by significant investor interest, had previously been seen as a niche player in the AI chip market. This agreement elevates its profile substantially, as the startup will now supply critical computing resources to one of the world’s most active developers of large language models. Its wafer-scale engines offer a fundamentally different approach to AI acceleration by integrating extremely large matrices on a single silicon die, reducing communication overhead and potentially enabling faster model execution.
For OpenAI, diversifying its infrastructure suppliers is also a hedge against supply constraints and concentration risk. In recent years, the company has expanded its chip partnerships; in 2025 it agreed to purchase AI processors from AMD — including a multi-gigawatt deployment — and has been working on custom silicon development with Broadcom as part of longer-term goals to reduce reliance on any single vendor.
Analysts warn that the AI hardware arms race may intensify as more companies seek compute capacity that can keep pace with increasingly complex generative models. While Nvidia continues to dominate the market with its GPU products, competition from specialised architectures like those offered by Cerebras could reshape the competitive landscape — particularly if these designs prove more cost-effective or better suited to specific types of AI workloads.
However, some observers caution that integrating diverse hardware types into a unified infrastructure introduces operational complexities, including software compatibility and workload orchestration. OpenAI will need to ensure that its systems can dynamically route tasks to the most efficient processors while maintaining reliability at scale.
In the broader AI ecosystem, this deal highlights how compute capacity — once a behind-the-scenes concern — has become a core strategic asset. As companies race to develop more advanced AI capabilities, securing high-performance chips and building resilient infrastructure will be essential to maintaining a competitive edge.
