By Chinonyerem Emmanuel
In November, fresh off another blockbuster earnings report, Nvidia CEO Jensen Huang couldn’t resist reminding investors just how central his company has become to the artificial intelligence revolution. Speaking about OpenAI, the most closely watched startup in the sector, Huang put it bluntly: “Everything that OpenAI does runs on Nvidia today.”
At the time, the statement rang true. Nvidia dominates the market for AI chips, sits atop the global rankings as the world’s most valuable company, and supplies the hardware backbone for nearly every major AI model in existence. But the AI boom is accelerating faster than any single company can comfortably support — and OpenAI knows it.
Behind the scenes, OpenAI is executing one of the most aggressive infrastructure expansion strategies in modern tech history, quietly reshaping the AI hardware landscape in the process. While Nvidia remains a cornerstone partner, competition is heating up, and OpenAI is moving decisively to diversify its chip supply, hedge against bottlenecks, and secure unprecedented amounts of computing power for the years ahead.
That strategy took another dramatic step forward this week.
On Wednesday, OpenAI announced a massive $10 billion agreement with Cerebras Systems, a smaller but ambitious chipmaker best known for its unconventional, wafer-scale processors.
Though Cerebras is still a relatively young player compared with giants like Nvidia and AMD, the deal signals OpenAI’s willingness to bet big on alternative architectures, and to help propel emerging hardware companies toward the public markets.
The Cerebras announcement is just the latest chapter in a dizzying series of partnerships between OpenAI and the companies that design the processors powering large language models and increasingly complex AI workloads.
Over the past year alone, OpenAI has committed more than $1.4 trillion to long-term infrastructure deals with chipmakers and systems partners, including Nvidia, Advanced Micro Devices, and Broadcom — an investment spree that has helped fuel its rise to an estimated $500 billion private-market valuation.
The message to the market is unmistakable: OpenAI wants as much compute as it can possibly get.
As demand for AI models explodes across enterprise, consumer, and government sectors, OpenAI is racing to lock down the raw processing power needed to train, deploy, and scale its technology globally. That urgency has translated into multi-gigawatt agreements, equity-linked warrants, custom chip development, and deep partnerships that stretch years into the future.
Nvidia remains OpenAI’s most entrenched partner, dating back to the company’s earliest efforts to build large language models long before ChatGPT ignited the generative AI boom. In 2025, the relationship reached an entirely new magnitude.
After investing in OpenAI in late 2024, Nvidia pledged $100 billion to support the deployment of at least 10 gigawatts of Nvidia-based systems, a power footprint roughly equivalent to the annual electricity consumption of 8 million U.S. households.
According to Huang, that scale translates into between 4 million and 5 million GPUs, underscoring the sheer industrial size of OpenAI’s ambitions. “This is a giant project,” Huang said at the time. The first phase is expected to come online later this year using Nvidia’s Vera Rubin platform, though Nvidia has cautioned investors that the arrangement has yet to progress into a binding long-term contract.
At the same time, OpenAI has made clear it does not intend to rely on Nvidia alone.
In October, the company revealed plans to deploy six gigawatts of AMD GPUs over multiple hardware generations. The deal includes warrants that could give OpenAI up to a 10% stake in AMD, tying the partnership not just to technology but to long-term financial alignment. AMD CEO Lisa Su framed the agreement as essential to building a resilient AI ecosystem capable of keeping pace with demand.
Broadcom, meanwhile, is helping OpenAI design and deploy custom AI accelerators, known internally as XPUs, in a 10-gigawatt project slated to run through 2029. While revenue from the partnership is expected to ramp slowly, Wall Street has already rewarded Broadcom’s AI ambitions with a valuation north of $1.6 trillion.
Then there’s Cerebras. OpenAI’s new deal will bring 750 megawatts of Cerebras’ wafer-scale processors online through 2028, delivering what the company claims are inference speeds up to 15 times faster than traditional GPU systems. For Cerebras, the agreement represents a crucial marquee customer at a time when it’s weighing a future IPO after previously pulling back from the public markets.
And OpenAI’s shopping list may not be finished.
Cloud giants like Amazon and Google are already deeply embedded in OpenAI’s infrastructure strategy. In November, OpenAI signed a $38 billion cloud agreement with Amazon Web Services, and discussions continue around a potential $10 billion-plus investment from Amazon talks that could eventually include AWS’s in-house AI chips, Trainium and Inferentia.
Google Cloud also supplies computing capacity, though OpenAI has so far avoided Google’s proprietary TPUs.
Intel, notably, remains on the sidelines, a missed opportunity that now looms large. The chipmaker reportedly passed on an early chance to invest in OpenAI and help supply its hardware, a decision that left Nvidia free to entrench its dominance. Intel is now scrambling to catch up, with new AI-focused GPUs expected to reach customers in late 2026.
Taken together, OpenAI’s strategy reflects a simple but profound reality: AI supremacy will be determined not just by algorithms, but by access to power, silicon, and scale. Nvidia may still sit at the center of that universe for now, but OpenAI is building an ecosystem designed to ensure it never depends on a single gravitational force again.
In the AI arms race, compute is king. And OpenAI is buying the crown jewels in bulk.
Source: CNBC
Original article
Comments