OpenAI and Nvidia are exploring a novel chip leasing arrangement as part of their recently announced $100 billion partnership to build massive AI data centers, allowing OpenAI to access Nvidia’s advanced GPUs without the upfront capital outlay of purchasing them. According to a September 23, 2025, report from The Information, the two companies are discussing this leasing model to structure the deal, which involves deploying at least 10 gigawatts of Nvidia systems starting in late 2026. For AI researchers, investors, and tech executives searching OpenAI Nvidia chip leasing, AI data center partnership 2025, or Nvidia leasing model, this potential shift from outright sales to leases could redefine how hyperscalers procure compute power, offering OpenAI flexible scaling while ensuring Nvidia a steady revenue stream amid soaring demand for its Blackwell and Rubin GPUs.
The leasing discussions are part of last-minute negotiations that sealed the megadeal during President Trump’s UK visit, blending equity investments with hardware commitments to fuel OpenAI’s superintelligence ambitions.
The Leasing Model: A New Twist on the $100 Billion Deal
Under the proposed structure, OpenAI would lease Nvidia’s AI chips for deployment in its data centers, rather than buying them outright, as confirmed by executives familiar with the talks. This approach addresses OpenAI’s massive compute needs—estimated at over 1 million GPUs by year-end—while mitigating the financial burden of a $50 billion per gigawatt buildout, of which $35 billion typically goes to Nvidia hardware.
Deal highlights:
- Initial Commitment: Nvidia’s $10 billion investment kicks off upon a definitive purchase agreement for the first gigawatt, but leasing could replace traditional sales.
- Scale: 10 gigawatts total, starting with Rubin platform chips in H2 2026.
- Equity Angle: Nvidia takes a non-controlling stake in OpenAI, with no voting rights, ensuring alignment without control.
OpenAI CEO Sam Altman emphasized compute’s centrality: “Everything starts with compute—it’s the basis for the economy of the future.” Nvidia CEO Jensen Huang, who personally delivered the company’s first DGX supercomputer to OpenAI in 2016, views the partnership as symbiotic.
| Aspect | Traditional Model | Leasing Proposal |
|---|---|---|
| Procurement | Outright Purchase | Rental/Lease Terms |
| Upfront Cost | $35B per GW | Lower Initial Outlay |
| Revenue for Nvidia | One-Time Sales | Recurring Fees |
| Flexibility for OpenAI | Fixed Assets | Scalable Access |
Strategic Drivers: Scaling AI Without Massive Capex
The leasing idea emerges from OpenAI’s escalating infrastructure demands: Training models like GPT-5 requires exaflop-scale resources, but upfront costs strain its $500 billion valuation and Microsoft-backed finances. Leasing allows rapid deployment while preserving cash for R&D.
For Nvidia:
- Recurring Revenue: Shifts from lumpy sales to predictable income, akin to cloud models.
- Customer Lock-In: Ensures long-term GPU utilization amid supply constraints.
- Diversification: Complements investments like $5 billion in Intel and $700 million in Nscale.
This isn’t exclusive—OpenAI continues collaborations with AWS, Google, and others to avoid single-vendor risks.
Market Reactions: Positive for Nvidia, Questions on Structure
Nvidia shares rose 4.4% post-announcement, adding $170 billion to its market cap, reflecting investor enthusiasm for the deal’s scale. However, analysts quipped about the “infinite money loop”: Nvidia invests in OpenAI, which buys (or leases) Nvidia chips.
Broader implications:
- AI Ecosystem: Accelerates OpenAI’s edge over rivals like Google DeepMind.
- Regulatory Scrutiny: DOJ/FTC may eye the equity stake for antitrust.
- Industry Precedent: Could inspire similar models for hyperscalers like Meta or xAI.
Challenges: Execution and Vendor Risks
Power shortages and construction timelines pose hurdles—each gigawatt requires $50 billion total, with leasing easing only Nvidia’s portion. OpenAI’s for-profit restructuring adds complexity, needing Microsoft/SoftBank nods.
Conclusion: Leasing the Path to Superintelligence
OpenAI’s potential shift to leasing Nvidia chips in the $100 billion deal is a pragmatic evolution, blending capital efficiency with compute scale to chase superintelligence. As Huang and Altman negotiate, it could redefine AI infrastructure economics. For those tracking Nvidia OpenAI partnership, Rubin deliveries in 2026 will test the model. Will leasing unlock trillions in AI value, or tangle in red tape? The servers are spinning. CNBC
