OpenAI has reportedly reached a multi-billion dollar deal to purchase computing power from startup Cerebras Systems, backed by CEO Sam Altman. This is the latest in a series of chip and cloud deals signed by ChatGPT maker OpenAI.
Chips designed by Cerebras
OpenAI plans to power its popular chatbot with chips designed by Cerebras, the two companies announced Wednesday. OpenAI has committed to purchasing up to 750 megawatts of computing power from Cerebras over the next three years. The deal is valued at over $10 billion, according to sources familiar with the matter.
Cerebras' AI chips claim to run AI models and generate responses faster than industry leader Nvidia. OpenAI CEO Dennis Altman is a personal investor in Cerebras, and the two companies explored a collaboration in 2017.
According to OpenAI's announcement, Cerebras builds dedicated AI systems designed to accelerate the long-term output of AI models. Its unique speed advantage stems from integrating massive computing power, memory, and bandwidth onto a single giant chip, eliminating bottlenecks in traditional hardware that cause inference speeds to drop.
Integrating Cerebras into our computing solutions portfolio aims to significantly improve the responsiveness of AI. When you ask a complex question, generate code, create an image, or run an AI agent, a loop runs in the background: you send a request, the model thinks, and then returns a result. When AI can respond in real time, users can leverage it to accomplish more, stay engaged longer, and run higher-value workloads.
Low latency capabilities integrated into the inference stack
We will integrate this low-latency capability into our inference stack in phases and scale it across a wide range of workloads.
"OpenAI's computing strategy is to build a resilient portfolio of systems that match the right systems to the right workloads," said Sachin Katti of OpenAI. "Cerebras adds a dedicated low-latency inference solution to our platform. This means faster response times, more natural interactions, and a more robust foundation to extend real-time AI to a wider audience."
"We are excited to partner with OpenAI to bring the world's leading AI models to the world’s fastest AI processors. Just as broadband revolutionized the internet, real-time inference will revolutionize AI, enabling entirely new ways to build and interact with AI models," said Andrew Feldman, co-founder and CEO of Cerebras.
This capacity will be rolled out in phases until 2028.
Bargaining chips in negotiations with Nvidia
OpenAI is ramping up its data center capacity acquisition efforts to prepare for the next phase of growth. With over 900 million weekly active users, executives have repeatedly stated they are facing severe computing resource shortages.
OpenAI is also seeking cheaper and more efficient alternatives to Nvidia chips. Last year, OpenAI announced a partnership with Broadcom to develop custom chips and a separate agreement to use AMD's new MI450 chip.
OpenAI and Cerebras began discussions last fall and signed a letter of intent before Thanksgiving, Cerebras CEO Andrew Feldman stated in an interview.
In a video interview with the Wall Street Journal, Feldman demonstrated a series of demonstrations showing chatbots powered by Cerebras chips responding to users much faster than those using competitor processors. He stated that it was precisely because his chips could handle AI computations more quickly that OpenAI reached a cooperation agreement with him.
"The driving factor in the market right now is the extraordinary demand for fast computing," Feldman said.
OpenAI's infrastructure director, Sachin Katti, said the company began considering a partnership with Cerebras after engineers expressed a desire for chips to run AI applications (especially for coding) faster.
"OpenAI's biggest revenue metric is computing power," Katti said in an interview. "Our computing power has tripled every year for the past two years, and our revenue has also tripled every year."
According to sources familiar with the matter, Cerebras is in talks to raise $1 billion at a valuation of $22 billion, which would nearly triple its valuation. Previously, *The Information* reported on the company's funding negotiations.
According to data from market research firm PitchBook, the company has raised a total of $1.8 billion, not including newly raised funds, with investors including Benchmark, UAE company G42, Fidelity Management & Research Co., and Atreides Management.
Chip startups focused on inference (the process of running trained AI models to generate responses) are in high demand as AI companies vie for cutting-edge technologies that provide fast, cost-effective computing power.
Nvidia and Groq sign licensing agreement
In December, Nvidia signed a $20 billion licensing agreement with Groq, enabling it to use chips developed by the startup, also designed for such tasks. In September, the chip giant also signed a preliminary agreement with OpenAI to sell up to 10 gigawatts of chips, but that agreement has not yet been finalized.
Cerebras, founded about a decade ago, has struggled to establish itself in the semiconductor market in recent years. When it filed for an IPO in 2024, the company disclosed that most of its revenue came from a single customer—Abu Dhabi-based G42. Cerebras withdrew its IPO plans the following year, instead raising $1.1 billion through a private placement. The deal valued the company at $8.1 billion. Feldman stated that Cerebras has since signed new collaboration agreements with IBM and Meta.
penAI is facing growing investor concerns about its ability to pay for computing service contracts. Last year, the company generated approximately $13 billion in revenue, a small fraction of the nearly $600 billion in new cloud contracts it has signed with Oracle, Microsoft, and Amazon. OpenAI CEO Richard Altman stated that the company will use future revenue growth to pay for these phased contract payments.
According to court documents released in the pending lawsuit between Altman and Elon Musk (OpenAI co-founder), OpenAI discussed a partnership with Cerebras in 2017. In an interview, Altman stated that around the same time, he rejected Musk's offer to acquire Cerebras, the developer of ChatGPT.
According to the Wall Street Journal, OpenAI is in the early stages of a new large-scale funding round to continue funding its ambitious growth plans. This new investment is expected to be completed before its initial public offering (IPO), at which point OpenAI's valuation (before the new investment) could reach $830 billion.
Source: Compiled from WSJ
