OpenAI has agreed to pay chip startup Cerebras more than $20 billion over the next three years to use servers powered by the company’s chips. As part of the deal, OpenAI could also receive an equity stake in the firm.
This didn’t come out of nowhere.
In January 2026, OpenAI announced a deal with Cerebras to deliver 750 megawatts of computing power through 2028, in an arrangement worth over $10 billion. The latest deal more than doubles that. What started as a supplier relationship is turning into something much deeper.
Who Is Cerebras, and Why Does OpenAI Want Their Chips?
Before we move further, you need to understand what Cerebras makes, because it’s genuinely different from what the rest of the chip industry is building.
Cerebras follows a technical path distinct from traditional players like Nvidia. Its flagship processor is roughly the size of a dinner plate and is designed specifically for large-scale parallel data processing. The chip uses Static Random-Access Memory (SRAM) to store data directly on the processor, reducing the need to move data between the chip and external storage, a potential bottleneck in Nvidia’s systems.
The WSE-3 wafer-scale chip contains 4 trillion transistors and 900,000 cores, delivering claimed 21x performance over Nvidia’s DGX B200 at one-third the cost and power. It basically does the same job faster, cheaper, and with less energy.
OpenAI’s Head of Compute Infrastructure Sachin Katti put it simply: “Cerebras adds a dedicated low-latency inference solution to our platform. Inference is the process by which an AI model like ChatGPT actually generates a response when you ask it a question. It’s the most computer-hungry part of running an AI product at scale, and the faster and cheaper you can do it, the better.
Currently, Cerebras chips are already providing compute power for a new OpenAI code model called Codex-Spark, though it is only available to a limited number of users
Most people read “$20 billion chip deal” and move on. But the financial structure behind this agreement is worth understanding, because it’s unconventional in a very deliberate way.
Under the deal, OpenAI will receive warrants for a minority stake in Cerebras, with its ownership potentially increasing as its spending rises. OpenAI has also agreed to provide Cerebras with about $1 billion to help fund the development of data centres that would run its AI products. If total spending reaches $30 billion over three years, that could translate into warrants representing up to 10% of Cerebras.
Then there’s the accounting angle. The information characterises the arrangement as a “working capital deposit” structure rather than a traditional procurement contract. Through this arrangement, OpenAI can book the $1 billion data centre financing provided to Cerebras as an asset and record a portion of the chip-related payments as interest income. This is important because it helps OpenAI’s balance sheet look healthier, which matters a lot when you’re preparing for a potential IPO.
OpenAI plans to use this agreement as a template for similar arrangements with other cloud service providers and chip manufacturers going forward. In other words, this isn’t a one-off deal. It’s a blueprint for how OpenAI plans to fund and structure its entire infrastructure strategy.
Breaking Free from Nvidia
Here’s the uncomfortable part of this. OpenAI, like almost every major AI company, is dangerously dependent on Nvidia. NVIDIA makes the GPUs that power the AI industry. It became the first company to reach a $5 trillion market capitalisation in October 2025, as investors sought to capitalise on further AI growth. When one company controls the hardware that an entire industry runs on, that’s a problem.
Despite the Cerebras agreement, Nvidia still holds an overwhelming position in OpenAI’s computer landscape. OpenAI plans to spend $45 billion on computers this year, rising to $90 billion next year, with cumulative spending over the next five years exceeding $650 billion, the vast majority still tied to leasing Nvidia chips.
But the direction is clear. OpenAI is actively building a world where it isn’t held hostage by a single chip supplier. The Cerebras deal is one piece of that. The “working capital deposit” template suggests more deals are coming.
This deal is transformative for Cerebras in a different way. Cerebras had been dealing with a serious problem: 87% of its revenue came from a single UAE-based client called G42, which raised regulatory red flags when the company first filed for an IPO in 2024.
The OpenAI relationship changed the story. CFIUS granted regulatory clearance after Cerebras restructured G42’s equity stake to non-voting shares. The OpenAI contract then gave Cerebras a credible path away from G42 revenue dependency. A $1 billion Series H round in February 2026, led by Tiger Global with participation from AMD, Fidelity, Benchmark Capital, and others, tripled the valuation to $23 billion.
The deal strongly propels Cerebras’s plan to restart its IPO at a valuation of approximately $35 billion. The company is targeting a Nasdaq listing under the ticker CBRS, raising approximately $2 billion with Morgan Stanley as lead underwriter, which would make it one of the 10 largest semiconductor IPOs in history and the first pure-play Nvidia alternative to reach public markets during the current AI infrastructure cycle.
This deal is three things at once. It’s a massive infrastructure bet by OpenAI on faster, cheaper inference chips. It’s a strategic move to reduce dependence on Nvidia. And it’s a financial engineering play that makes OpenAI look better on paper ahead of its own IPO.
For Cerebras, it’s a lifeline that became a launchpad.
The AI industry has always been about who has the most computers. What’s changing now is who controls it, and how deep the relationships between buyers and builders are starting to run.
Read also: GMX launches tokenised gold and silver trading on-chain