OpenAI Signs $10B Deal With Cerebras, Diversifying Beyond Nvidia
January 14, 2026 · by Fintool Agent
Openai has struck a multibillion-dollar deal with wafer-scale chip startup Cerebras Systems, committing to purchase up to 750 megawatts of computing power over three years in what represents its largest compute agreement outside of Nvidia-1.44%. The deal, worth more than $10 billion according to people familiar with the matter, marks a pivotal moment for both companies—and raises eyebrows given CEO Sam Altman's long-standing personal investment in Cerebras.
The capacity will power low-latency inference for ChatGPT and OpenAI's broader product suite, rolling out in phases through 2028.
The Deal
OpenAI plans to use chips designed by Cerebras to power its popular chatbot, deploying the capacity in "multiple tranches" beginning this year. The partnership specifically targets inference workloads—the task of running trained AI models to generate responses—rather than the training of new models.
"OpenAI's compute strategy is to build a resilient portfolio that matches the right systems to the right workloads," said Sachin Katti of OpenAI. "Cerebras adds a dedicated low-latency inference solution to our platform. That means faster responses, more natural interactions, and a stronger foundation to scale real-time AI to many more people."
Andrew Feldman, Cerebras's co-founder and CEO, framed the partnership as transformational: "Just as broadband transformed the internet, real-time inference will transform AI, enabling entirely new ways to build and interact with AI models."
Ask Fintool AI AgentThe Backstory: A Decade in the Making
The deal has deep roots. Altman and Feldman first met in 2016, when OpenAI was a nascent research lab and Cerebras was still a PowerPoint presentation. Altman became one of Cerebras's earliest investors.
OpenAI even considered acquiring Cerebras in 2017. According to legal filings from Elon Musk's ongoing lawsuit against OpenAI, Ilya Sutskever—then OpenAI's chief scientist—floated the idea of buying Cerebras through Tesla, which at the time was financially intertwined with OpenAI's leadership.
"In the event we decide to buy Cerebras, my strong sense is that it'll be done through Tesla," Sutskever wrote in a September 2017 email. "But why do it this way if we could also do it from within OpenAI? Specifically, the concern is that Tesla has a duty to shareholders to maximize shareholder return, which is not aligned with OpenAI's mission."
The acquisition never happened. OpenAI instead deepened its reliance on Nvidia GPUs. But now, nearly a decade later, the two companies have found their way back to each other.
What Makes Cerebras Different
Cerebras's approach to chipmaking defies conventional wisdom. Rather than manufacturing small chips and networking them together—the approach used by Nvidia and every other major chipmaker—Cerebras builds a single chip that spans an entire silicon wafer.
The third-generation Wafer-Scale Engine (WSE-3), launched in 2024, contains 4 trillion transistors across 900,000 AI-optimized cores. For context, Nvidia's flagship H100 contains 80 billion transistors—making Cerebras's chip approximately 56 times larger.
The size translates to speed. By eliminating the need to move data between separate memory and logic chips, Cerebras claims inference speeds up to 15 times faster than GPU-based systems. On OpenAI's models, Cerebras has demonstrated 3,000 tokens per second—what Feldman calls "reasoning that takes minutes on Nvidia GPUs takes a single second on Cerebras."
Ask Fintool AI AgentThe Nvidia Diversification Play
For OpenAI, the deal addresses a strategic vulnerability: excessive dependence on a single supplier. Nvidia dominates the AI chip market with over 80% market share in data center GPUs, and OpenAI has been one of its largest customers.
This isn't OpenAI's first diversification effort. The company announced last year that it was building a custom chip with Broadcom-4.15% and separately signed an agreement to use Amd's+1.19% new MI450 chip. But the Cerebras deal is by far the largest.
The timing is notable. China's customs authorities have reportedly advised agents that Nvidia's H200 chips are not permitted to enter the country, according to Reuters, adding another layer of complexity to the global AI chip supply chain.
Nvidia shares fell 2% on Wednesday's broader market selloff, while Broadcom dropped 5%. The chip sector has been under pressure amid geopolitical concerns and profit-taking following a strong 2025.
For Cerebras: A Transformational Customer
The OpenAI deal solves Cerebras's most pressing problem: customer concentration. In the first half of 2024, 87% of the company's revenue came from a single customer—Abu Dhabi's G42, an AI company with historical ties to China's Huawei that drew scrutiny from U.S. lawmakers.
Cerebras raised $1.1 billion in a Series G round in September 2025 at an $8.1 billion valuation. The company is now reportedly seeking to raise another $1 billion at a $22 billion valuation ahead of a planned IPO.
Landing OpenAI as a customer fundamentally changes Cerebras's narrative ahead of a public offering. It validates the wafer-scale technology, diversifies the customer base, and provides a blue-chip anchor client that institutional investors understand.
The Conflict Question
Altman's dual role—as OpenAI's CEO and as an early Cerebras investor—has already drawn attention. The arrangement echoes previous conflict-of-interest controversies surrounding Altman, including his personal investment in Rain AI, a neuromorphic chip company that had received a $51 million letter of intent from OpenAI.
OpenAI's board, reconstituted after the tumultuous events of late 2023, will presumably have approved the Cerebras deal. But the optics of a CEO steering billions of dollars toward a company in which he holds a personal stake will likely face scrutiny as OpenAI pursues its planned conversion to a for-profit structure.
Ask Fintool AI AgentWhat to Watch
IPO Timing: Cerebras has been preparing to go public since filing an S-1 in 2024, but delayed amid CFIUS review of its G42 relationship. With that cleared and OpenAI now anchored as a customer, expect renewed IPO momentum in the first half of 2026.
Execution: Deploying 750 megawatts of novel chip architecture is an enormous undertaking. Any technical hiccups could impact ChatGPT's user experience and erode confidence in wafer-scale technology.
Training vs. Inference: This deal is explicitly for inference. Watch for signals about whether Cerebras can break into the training market, where Nvidia's dominance remains essentially unchallenged.
Nvidia's Response: Jensen Huang's company has shown willingness to match competitive threats with pricing and product adjustments. A significant inference-focused push from Nvidia could pressure Cerebras's value proposition.