At a Glance
- OpenAI signed a multi-year compute deal with AI chipmaker Cerebras worth over $10 billion.
- Cerebras will deliver 750 megawatts of compute from 2026 through 2028.
- The partnership targets faster, real-time responses for OpenAI users.
- Why it matters: Low-latency inference could make AI interactions feel instant at global scale.
OpenAI has locked in a multi-year supply of specialized AI compute by partnering with Cerebras, the wafer-scale chipmaker that has quietly become a critical piece of the post-ChatGPT infrastructure boom.
The agreement, announced Wednesday, commits Cerebras to deliver 750 megawatts of compute capacity to OpenAI starting this year and running through 2028. A source familiar with the terms told News Of Philadelphia the contract is valued at more than $10 billion; Reuters published the same figure.
Speed Over Scale
Both companies framed the alliance as a speed play rather than a raw-capacity move. OpenAI said the new systems will cut response times for queries that today require extended processing. Cerebras CEO Andrew Feldman likened the shift to broadband’s impact on the early internet, arguing that “real-time inference will transform AI.”
OpenAI’s Sachin Katti echoed the latency focus, stating the deal adds “a dedicated low-latency inference solution” to the firm’s compute mix. The goal, Katti said, is “faster responses, more natural interactions, and a stronger foundation to scale real-time AI to many more people.”
A Rising Supplier
Cerebras has spent over a decade developing wafer-scale processors purpose-built for AI workloads. The company claims its systems outperform GPU-based clusters on both throughput and energy efficiency. Industry attention spiked after ChatGPT’s 2022 debut triggered a race for faster training and inference hardware.
Despite filing for an IPO in 2024, Cerebras has repeatedly delayed going public while continuing to raise private capital. On Tuesday, reports surfaced that the firm is in talks for a new $1 billion funding round at a $22 billion valuation.
Existing Ties
The new supply pact deepens an already intertwined relationship. OpenAI CEO Sam Altman is an existing investor in Cerebras, and the AI lab previously weighed acquiring the chipmaker outright. Those talks did not culminate in a purchase, but they set the stage for Wednesday’s supply agreement.
OpenAI described its broader compute strategy as building “a resilient portfolio that matches the right systems to the right workloads.” Adding Cerebras gives the company a specialized path for workloads where millisecond-level latency matters, complementing the massive GPU clusters it already operates.
Market Ripple
The deal underscores how AI giants are moving beyond off-the-shelf GPUs to lock in custom or semi-custom silicon and architectures. By guaranteeing Cerebras billions in revenue, OpenAI also secures priority access to future generations of wafer-scale hardware, potentially giving it an edge in real-time applications such as voice agents, coding assistants, and autonomous systems.
For Cerebras, the contract validates its pivot from research novelty to production supplier, providing cash flow that could further delay-or even render unnecessary-a public listing in the near term.
Key Takeaways
- $10 billion-plus supply deal runs through 2028, making Cerebras a core OpenAI partner.
- 750 megawatts of compute capacity targets low-latency inference, not bulk training.
- Existing investor ties and prior acquisition talks show relationship long in the making.
- Funding talks for $1 billion at a $22 billion valuation signal continued private-scale ambitions.
