OpenAI has entered into a multibillion-dollar agreement to purchase computing capacity from AI chip startup Cerebras Systems, aiming to bolster infrastructure as demand for ChatGPT continues to accelerate and strain its data center and network resources.
Under the agreement, OpenAI will deploy Cerebras-designed chips to handle portions of its ChatGPT inference workloads, committing to as much as 750 megawatts of compute capacity over a three-year period, according to a Wall Street Journal report.
The deal underscores the growing pressure that large-scale AI deployments are placing on power supply, networking, and inter–data center connectivity, as OpenAI evaluates faster and more cost-effective alternatives to Nvidia’s market-leading GPUs.
OpenAI leadership has previously cautioned that the company is running up against compute constraints, noting that its products are now used by more than 800 million people each week. This scale has made it necessary to bring on additional infrastructure partners.
The Cerebras agreement is part of a broader push by OpenAI to diversify its infrastructure stack. Other initiatives include developing a custom AI chip in collaboration with Broadcom and planning deployments of AMD’s latest accelerators, all aimed at managing costs and reducing dependence on Nvidia.
At the industry level, hyperscale AI platforms are driving a shift away from monolithic, general-purpose clusters toward more layered and heterogeneous infrastructure architectures.
“OpenAI’s move toward Cerebras inference capacity reflects a broader shift in how AI data centers are being designed. This move is less about replacing Nvidia and more about diversification as inference scales.”
Prabhu Ram, VP of the industry research group at Cybermedia Research.\
Prabhu Ram
