---

As artificial intelligence races from experimentation to everyday use, access to raw computing power has become the industry’s most critical bottleneck. OpenAI’s latest agreement with chipmaker Cerebras highlights how the battle for faster, more efficient AI is now being fought at the infrastructure level.

The deal also signals a shift in how AI leaders are diversifying beyond Nvidia’s dominant GPUs to gain an edge in speed, cost, and scale.

OpenAI locks in massive compute capacity

OpenAI has agreed to purchase up to 750 megawatts of computing power over the next three years from U.S.-based AI chipmaker Cerebras Systems, according to a joint announcement by the two companies on Wednesday.

The agreement, which a source familiar with the matter valued at more than $10 billion, is one of the largest infrastructure commitments OpenAI has made as demand for ChatGPT and other AI tools continues to surge.

Under the deal, OpenAI will rely on Cerebras-powered cloud services to run its AI models, particularly for inference, the stage where trained models generate responses for users in real time.

Why Cerebras caught OpenAI’s attention

Cerebras CEO Andrew Feldman said discussions began in August last year, after the company demonstrated that OpenAI’s open-source models could run more efficiently on its proprietary chips than on conventional graphics processing units (GPUs).

Unlike traditional chips, Cerebras designs wafer-scale engines, massive single-piece processors built specifically to handle the enormous workloads required by large language models.

After months of technical evaluations and negotiations, the two companies settled on a structure where Cerebras would sell cloud-based compute services rather than physical hardware directly to OpenAI.

Focus on inference and reasoning models

The partnership is heavily centered on inference and reasoning workloads, a fast-growing segment of AI computing.

Reasoning models, which pause briefly to evaluate multiple steps before generating answers, are significantly more demanding than standard text-generation systems. These models are increasingly seen as the next leap in AI capability, but they require far more computing power to operate at scale.

Cerebras will either build or lease data centers filled with its chips, while OpenAI will pay to use that capacity. The computing power will be rolled out in multiple phases through 2028, ensuring long-term supply as OpenAI’s products expand.

OpenAI: Faster responses are the goal

OpenAI framed the deal as a strategic move to improve performance rather than just scale.

“Integrating Cerebras into our mix of compute solutions is all about making our AI respond much faster,” the company said in a statement published on its website.

Speed has become a competitive differentiator in the AI market, particularly as users increasingly rely on chatbots for real-time assistance in work, education, and software development.

A broader shift in the AI infrastructure race

The agreement reflects a wider industry trend: inference is now as important as training.

While early AI investments focused on training massive models, companies are now racing to ensure those models can respond instantly and reliably to millions of daily queries. That shift has driven unprecedented demand for specialized chips and custom-built data centers.

As a result, AI companies are increasingly looking beyond Nvidia, the current market leader, to alternative chipmakers that can offer efficiency gains or cost advantages.

Boost for Cerebras ahead of IPO plans

For Cerebras, the OpenAI deal carries significance beyond revenue.

The company has been working to diversify its customer base, which has historically included UAE-based technology firm G42, both a major investor and one of its largest clients.

Landing OpenAI as a long-term customer strengthens Cerebras’ position as it prepares for a return to the public markets.

According to a Reuters report, Cerebras is planning to file for an initial public offering, targeting a listing in the second quarter of this year. This would mark its second IPO attempt after withdrawing a previous filing in October 2024.

Deep ties between OpenAI and Cerebras

The relationship between the two companies is not entirely new.

OpenAI CEO Sam Altman is an early investor in Cerebras, underscoring long-standing ties between the leadership teams. Those connections may have helped accelerate trust as OpenAI evaluates alternatives to traditional GPU-heavy infrastructure.

Cerebras, founded in 2015, competes directly with Nvidia and other AI chipmakers by focusing on purpose-built silicon for large-scale AI workloads.

OpenAI’s trillion-dollar ambitions

The Cerebras agreement fits into OpenAI’s far larger infrastructure vision.

According to Reuters, OpenAI is laying the groundwork for a future initial public offering that could value the company at up to $1 trillion, reflecting investor optimism around generative AI.

Altman has previously said OpenAI is committed to spending $1.4 trillion to develop 30 gigawatts of computing capacity, enough electricity to power roughly 25 million U.S. homes.

Such numbers illustrate just how energy- and capital-intensive the AI boom has become.

Growing concerns of an AI bubble

Despite soaring valuations and massive spending commitments, some investors and industry experts are urging caution.

Comparisons are increasingly being drawn to the dotcom boom, when aggressive infrastructure investments raced ahead of sustainable business models. Critics warn that not all AI demand projections may materialize as expected.

For now, however, companies like OpenAI appear determined to secure as much computing power as possible, betting that faster, more capable AI systems will translate into long-term dominance.

What comes next

The OpenAI–Cerebras partnership highlights a new phase of the AI race, where who controls the fastest inference infrastructure may matter as much as who builds the smartest models.

As reasoning-based AI becomes mainstream and competition intensifies, deals like this are likely to become more common, and more expensive.

Whether these investments prove visionary or excessive will depend on how quickly AI adoption turns massive compute spending into durable revenue.

(Attribution: With inputs from Reuters)

ALSO READ:  India’s New Smartphone Security Plan Sparks Privacy Backlash