OpenAI is making another major move to secure the computing power behind ChatGPT. According to a Wall Street Journal report, the company has signed a multi-billion dollar agreement with Cerebras Systems, a chip startup known for building unusually large AI processors.
The deal is valued at more than $10 billion and gives OpenAI the option to purchase up to 750 megawatts of computing capacity over three years. For a company that says it now serves well over 900 million weekly users, the agreement points to a simple reality: the AI race is increasingly a race for compute.
A large deal aimed at a severe compute shortage
OpenAI leadership has repeatedly described its computing resources as being under a “severe shortage.” That shortage matters because large AI services depend on vast amounts of processing power, both to train models and to run them for users.
The Cerebras agreement is focused on expanding capacity for ChatGPT. The source article states that Cerebras chips will power ChatGPT, placing the startup’s hardware inside one of the most visible AI products in the market.
The relationship between the companies is not entirely new. Court documents from the legal battle between Sam Altman and Elon Musk show that OpenAI and Cerebras had considered a partnership as far back as 2017. Altman, OpenAI’s CEO, is also a personal investor in Cerebras.
The size of the new agreement makes it more than a routine supplier deal. It suggests OpenAI is trying to build a wider base of infrastructure partners at a time when compute has become one of the biggest constraints on AI growth.
What makes Cerebras different
Cerebras was founded roughly a decade ago and has taken a different path from traditional chipmakers. Instead of relying on many smaller processors, the company builds an entire silicon wafer into one unified chip called the Wafer Scale Engine, or WSE.
The idea behind the Wafer Scale Engine is to reduce the communication problems that appear when large numbers of separate GPUs need to coordinate with each other. Cerebras places hundreds of thousands of AI-optimized compute cores on a single chip, with direct on-chip communication through the silicon.
The company’s chip generations have advanced quickly. The first WSE-1 launched in 2020, followed by WSE-2 about a year later. The third-generation WSE-3 features 900,000 AI cores and 4 trillion transistors.
Cerebras presents its systems as a fast inference alternative to GPU clusters from vendors like Nvidia. According to Cerebras, a single CS-3 system can replace large GPU clusters made up of hundreds or thousands of graphics processors.
That inference angle is important. Inference is the work of running an AI model after it has been built, which is what happens when people use products like ChatGPT. As user demand grows, fast and efficient inference becomes a central infrastructure challenge.
Why this matters for Cerebras
The OpenAI agreement arrives after Cerebras faced difficulty gaining broader traction in the semiconductor market. When the company filed for an IPO in 2024, most of its revenue came from one customer: Abu Dhabi-based G42.
Cerebras later pulled its IPO and raised $1.1 billion privately instead. The company is now in talks for a funding round at a $22 billion valuation and is building multiple data centers across North America and Europe.
A large agreement with OpenAI could change how the company is perceived. It places Cerebras in direct connection with a major AI platform and gives its wafer-scale approach a high-profile use case.
The deal also comes as demand for alternatives to conventional GPU clusters is rising. Cerebras has long argued that its architecture can handle AI workloads in a different way. OpenAI’s decision to buy capacity from the company gives that argument a much larger stage.
OpenAI’s broader push beyond Nvidia
The Cerebras deal is part of a wider effort by OpenAI to find cheaper alternatives to Nvidia chips. The company is working with Broadcom to develop its own chip and has also signed a deal for AMD’s MI450 chip.
Nvidia still remains central to the AI chip market, but the source article frames the Cerebras agreement as another sign of growing competition in inference. OpenAI also signed a preliminary agreement with Nvidia in September for up to ten gigawatts of chips, although that deal has not been finalized yet.
Nvidia has also moved in the fast inference chip market. In December, it signed a $20 billion licensing agreement with Groq, another chip startup focused on fast inference systems similar to Cerebras. As part of that deal, roughly 90 percent of Groq’s workforce is moving to Nvidia.
Taken together, these moves show how quickly the AI hardware landscape is shifting. OpenAI is not relying on a single path. It is signing capacity deals, pursuing custom chip work, and bringing more suppliers into its infrastructure plans.
The bigger signal
The main takeaway is not just that OpenAI is spending heavily. It is that compute capacity has become a strategic resource for AI companies with large user bases.
For OpenAI, more capacity could help ease pressure around ChatGPT usage. For Cerebras, the agreement could validate its wafer-scale chip strategy after years of trying to break through in a difficult semiconductor market.
For the broader AI industry, the deal highlights a market that is no longer only about model quality. The companies that can secure enough power, chips, data centers, and inference capacity may shape what AI products can actually deliver at scale.