OpenAI's Sam Altman Reaffirms Close Partnership With Cerebras
Key Takeaways
- •Sam Altman publicly reaffirmed on X that Cerebras remains a close OpenAI partner after rumors that GPT-6.1 Sol's initial deployment used Nvidia GPUs rather than Cerebras hardware.
- •Market speculation in early October 2026 pressured Cerebras shares, exposing a customer-concentration vulnerability tied to its largest buyer.
- •On January 14, 2026, OpenAI to purchase up to 750 megawatts of Cerebras wafer-scale inference capacity over three years, in a deal initially valued above $10 billion and potentially exceeding $20 billion.
- •Cerebras capacity is slated to roll out in phases through 2028 to speed applications like ChatGPT, positioned as a complement to OpenAI's existing suppliers Nvidia and AMD.
- •Altman has personally held roughly 89,373 Cerebras shares since February 2017, a dual customer-CEO and shareholder role likely to keep governance questions in focus.

Sam Altman has publicly described Cerebras as a close partner of OpenAI, and the timing of the message was no coincidence. The reaffirmation, shared on X, followed a stretch of speculation about whether the relationship between OpenAI and the AI chipmaker was cooling, during which Cerebras shares came under pressure. A single reassuring message from the most prominent CEO in AI tends to carry weight in such moments.
What Sparked the Speculation
In early October 2026, market rumors began circulating about OpenAI's newest Ultrafast-tier model, GPT-6.1 Sol. According to those reports, the model is using Nvidia GPUs for its initial deployment rather than Cerebras hardware.
The detail landed awkwardly. Cerebras pitches its technology specifically for ultra-low-latency inference — in practical terms, hardware built to make AI respond faster. An “Ultrafast” model running on another vendor's silicon looked, to some investors, like a snub.
Altman's message appears designed close that gap. In his telling, partners remain partners, even when a particular launch runs on someone else's chips.
The Deal Behind the Drama
On January 14, 2026, OpenAI and Cerebras Systems announced a multi-year partnership under which OpenAI committed to purchase up to 750 megawatts of Cerebras' wafer-scale AI inference capacity over the following three years.
The megawatt-based framing has become a common yardstick for AI infrastructure deals, since power capacity is the practical limit on how much computing a facility can run. Cerebras' contribution is also distinctive: its wafer-scale design packs processing elements across an entire silicon wafer rather than onto discrete chips — an architecture the company pitches as purpose-built for fast inference.
The deal was initially valued at more than $10 billion, and subsequent reports indicated the total could potentially exceed $20 billion once extensions and related financing are counted.
The capacity is set to roll out in phases through 2028. The plan is to fold Cerebras' hardware into OpenAI's inference stack — the part of the operation that actually answers user requests — with the goal of making applications like ChatGPT faster and more responsive.
A Relationship With a Long History
Altman has been a personal investor in Cerebras since at least February 2017, holding approximately 89,373 shares of the company since then — a stake that appreciated significantly after Cerebras went public in May 2026.
The two companies explored ways to work together around that same period, and their collaboration has since extended to OpenAI's open-weight GPT-OSS models, giving the January deal a foundation rather than a cold start.
Both sides have framed Cerebras as a complement to OpenAI's existing suppliers, notably Nvidia and AMD, rather than a replacement for them. The pitch is dedicated, low-latency capacity sitting alongside the general-purpose GPU fleet.
What It Means for Cerebras and Its Investors
The October selloff exposed a structural vulnerability for Cerebras: customer concentration. When a single buyer accounts for a contract of this size, every rumor about that buyer moves the stock.
Analysts believe the market overstated the risk. The agreement is multi-year, the capacity rolls out in phases through 2028, and one model's initial deployment choice does not rewrite a three-year purchase commitment. How much of the phased capacity actually comes online, and where future Ultrafast-tier models run, will be the practical tests of that argument.
Altman's personal stake adds another layer. He is both the CEO of the customer and a longtime shareholder in the supplier — a dual role that makes his public reassurance notable and will likely keep governance questions in the conversation.
OpenAI is spreading its bets across Nvidia, AMD, and Cerebras, and each supplier now competes not just for signed deals but for the flagship launches that prove those deals matter.