OpenAI inks $10B Cerebras compute pact - AI News Today Recency

# OpenAI Inks $10B Cerebras Compute Pact for AI Inference

OpenAI has announced a multi-year agreement worth over $10 billion with AI chipmaker Cerebras to deliver 750 megawatts of computing power through 2028[2]. The landmark deal represents a significant strategic move for OpenAI to enhance its AI infrastructure and deliver faster response times for complex tasks, marking what could be a breakthrough moment for Cerebras in the competitive AI chip market[3].

The collaboration underscores OpenAI's commitment to building a diversified compute portfolio that optimizes performance across different workloads. According to Sachin Katti, OpenAI's infrastructure executive, the partnership will provide "faster responses, more natural interactions, and a stronger foundation to scale real-time AI to many more people," with particular interest in using Cerebras systems for AI applications in coding[2][3].

The Strategic Importance of Low-Latency Inference

The core value of this partnership lies in Cerebras' specialized capability for low-latency inference—the ability to process AI queries and return responses with minimal delay[2]. Unlike traditional GPU-based systems from companies like Nvidia, Cerebras' wafer-scale chip architecture is engineered specifically for fast AI computations[2].

This speed advantage addresses a critical market need. Andrew Feldman, Cerebras' co-founder and CEO, emphasized that "this extraordinary demand for fast compute" is driving the market right now[3]. The company has already demonstrated impressive performance metrics, running OpenAI's gpt-oss-120B model at inference speeds of 3,000 tokens per second on its infrastructure[3].

OpenAI's Diversified Compute Strategy

OpenAI's decision to partner with Cerebras reflects a broader strategy to avoid over-reliance on any single chip supplier. By building what the company calls a "resilient portfolio that matches the right systems to the right workloads," OpenAI is positioning itself to handle diverse AI applications more efficiently[2].

The 750 megawatts of computing capacity will be delivered starting this year and continuing through 2028, providing OpenAI with substantial computational resources for inference tasks[2]. This approach complements OpenAI's existing infrastructure investments and demonstrates the company's commitment to maintaining technological flexibility in a rapidly evolving AI landscape.

Cerebras' Path to Mainstream Recognition

For Cerebras, this deal represents a transformational moment for the company. Founded over a decade ago and introducing its distinctive wafer-scale GPU in 2019, Cerebras has operated somewhat on the periphery of the AI chip market dominated by Nvidia[2][3].

The OpenAI partnership validates Cerebras' technology and provides the company with unprecedented visibility and credibility. This momentum is already reflected in the company's valuation; Cerebras is currently in talks to raise $1 billion at a $22 billion valuation, having raised a total of $1.8 billion to date[3]. Notably, OpenAI CEO Sam Altman is already an investor in Cerebras, and the companies had previously explored an acquisition before settling on this partnership model[2].

Frequently Asked Questions

What exactly is the Cerebras-OpenAI deal about? OpenAI has committed to purchasing 750 megawatts of computing power from Cerebras over a multi-year period through 2028 in a deal valued at over $10 billion[2]. The compute will be used to power AI inference applications, enabling faster response times for complex tasks[2].

Why is low-latency inference important for AI? Low-latency inference reduces the time it takes for AI models to process queries and return responses. This is critical for real-time applications and user experience, particularly for tasks that currently require significant processing time[2]. Cerebras' technology is specifically designed to excel at this capability[3].

How does Cerebras' technology differ from Nvidia GPUs? Cerebras uses a wafer-scale chip architecture designed specifically for AI workloads, which the company claims delivers faster performance than traditional GPU-based systems like Nvidia's offerings[2]. The technology has demonstrated inference speeds of 3,000 tokens per second for OpenAI models[3].

When will Cerebras start delivering compute to OpenAI? Cerebras will begin delivering the 750 megawatts of computing power starting this year (2026) and continue through 2028[2]. The deal was reportedly signed nearly two months before the January 2026 announcement[3].

What applications will benefit most from this partnership? OpenAI infrastructure executives have indicated particular interest in using Cerebras for AI applications in coding[3]. More broadly, the compute will be used for inference tasks that require fast response times and real-time processing capabilities[2].

Why didn't OpenAI just acquire Cerebras instead? While OpenAI had previously considered acquiring Cerebras, the companies ultimately chose a partnership model instead[2]. This approach allows OpenAI to maintain its diversified compute strategy while giving Cerebras independence to serve other customers and pursue its own growth trajectory, including potential future IPO plans[3].