OpenAI and Nvidia have signed a letter of intent for a sweeping strategic partnership that pairs capital with compute at unprecedented scale. Under the plan, Nvidia will invest up to $100 billion in OpenAI as the startup deploys at least 10 gigawatts of Nvidia systems to train and run its next generation of models—part of OpenAI’s march toward “superintelligence.” The first phase is slated to come online in the second half of 2026 on Nvidia’s upcoming Vera Rubin platform.
While the two companies have collaborated for years—from early DGX systems to the breakout of ChatGPT—the new arrangement formalizes a deeper alignment of incentives. Nvidia gains a financial stake—via non-voting shares, once the definitive agreement is signed—in the world’s most prominent AI lab, already one of its largest customers. OpenAI, most recently valued around $500 billion, secures both capital and a guaranteed runway of advanced chips amid intense global demand and growing supply-chain constraints.
How the deal is structured
According to people familiar with the terms, the pact involves two intertwined tracks. After finalizing a definitive purchase agreement for the hardware, Nvidia will make an initial $10 billion investment, with further tranches tied to deployment milestones as new systems are installed. OpenAI can then use the cash to procure additional Nvidia data center platforms and networking gear. The companies also plan to co-optimize their roadmaps, aligning OpenAI’s model and infrastructure software with Nvidia’s silicon, systems, and software stack.
The compute footprint is eye-watering: 10GW of systems—roughly the electricity needs of 8+ million U.S. homes—to be delivered on Vera Rubin beginning late 2026. That capacity is intended to power training and inference for OpenAI’s next-gen models, while expanding the “AI factory” concept across multiple partners, including Microsoft, Oracle, SoftBank and the broader Stargate initiative to build massive AI data centers worldwide.
Strategic rationale—and competitive ripples
For OpenAI, leadership hinges on access to cutting-edge compute. “Everything starts with compute,” CEO Sam Altman said, framing infrastructure as the base layer of the future economy. The tie-up gives OpenAI an inside track on Nvidia’s newest platforms while reinforcing continuity with existing partners. Importantly, sources say the deal does not derail OpenAI’s parallel plans to develop custom AI chips—projects reported to involve Broadcom and TSMC—nor does it alter the company’s long-standing relationship with Microsoft.
For Nvidia, the agreement shores up demand visibility and further entrenches its status as the default AI hardware provider at a time when rivals—from AMD to in-house silicon efforts at hyperscalers—are seeking to chip away at its lead. Analysts suggested such megadeals could counterbalance risks tied to export controls and lost China sales, while sending a message that alternate platforms remain multiple cycles behind. Nvidia’s stock climbed to a record intraday high, rising as much as 4.4% after the announcement; Oracle, a data-center partner to OpenAI in Stargate, gained about 6%.
What gets built—and when
The near-term milestone is the first gigawatt of compute targeted for the second half of 2026, delivered on Vera Rubin. The staging gives Nvidia time to ramp next-gen architectures and manufacturing capacity, and gives OpenAI time to synchronize software, data pipelines, and model training schedules. In parallel, the firms will work on performance co-design—tightening how model parallelism, compiler toolchains, and networking fabric evolve alongside GPU and NIC roadmaps.
OpenAI says the investment supports a user base now topping 700 million weekly active users across consumers, enterprises, SMBs, and developers. The company frames the expanded infrastructure as necessary to push the frontier of capability while productizing models safely and at scale.
Regulatory backdrop
The scale and entanglement of the partnership is likely to draw scrutiny. The U.S. Department of Justice and Federal Trade Commission agreed in mid-2024 to delineate oversight of major AI players, potentially paving the way for deeper probes into the roles of Microsoft, OpenAI, and Nvidia. Antitrust experts note that a tight coupling between the dominant AI chip supplier and a leading model developer could alter market incentives and make it harder for competitors to gain a foothold—whether AMD in accelerators or rival foundation-model labs in software. For now, the current U.S. administration’s more business-friendly posture suggests a higher bar for intervention, though regulatory winds can shift quickly.
Bigger picture: the compute era’s flywheel
For all the headlines about dollar figures, the core bet is on a compute-driven flywheel: more advanced chips and systems enable larger, longer training runs; that yields more capable models; those models attract more users and enterprise workloads, which in turn justify even greater infrastructure outlays. Nvidia founder and CEO Jensen Huang cast the moment as the “next leap forward” in a decade-long partnership, while OpenAI co-founder Greg Brockman emphasized scaling benefits “to everyone” as the company pushes the frontier.
The letter of intent leaves room for adjustments as both sides finalize details in the coming weeks. But the direction of travel is unmistakable: the AI stack’s leaders are locking in multi-year, multi-billion-dollar commitments that fuse capital, chips, and software into a single strategic arc. If executed as planned, OpenAI’s next models will be trained on some of the densest AI infrastructure ever built—and Nvidia will remain at the center of the industry’s most powerful growth engine.
