Reports indicate that OpenAI, currently preparing for an IPO, is in advanced negotiations for an unprecedented data center lease. The project, located on federal land in Pike County, southern Ohio, is planned with a total capacity of 10 gigawatts and is estimated to cost at least $500 billion to build. This move represents not only OpenAI's largest infrastructure commitment to date but also signals a new phase in the global AI arms race where power capacity equates directly to competitive advantage.
A unique aspect of this deal is the role being played by NVIDIA. According to informed sources, NVIDIA will not only supply all the GPU hardware for the facility but is also set, for the first time, to provide a credit guarantee using its own balance sheet. This guarantee would back OpenAI's lease payments and the project financing for SB Energy, establishing a novel model for deep chipmaker involvement in infrastructure funding. Concurrently, Google is reportedly providing a backstop arrangement for rival Anthropic's approximately $35 billion in TPU leasing obligations, indicating that the financial ties between AI giants and chip manufacturers have escalated from simple investments to balance-sheet-level guarantees.
Project Overview: From Cold War Uranium Plant to AI Powerhouse
The proposed data center site is strategically significant, located on the former Portsmouth uranium enrichment plant in Pike County, Ohio, which supplied weapons-grade uranium for the U.S. nuclear program during the Cold War before closing in 2001. In March 2026, the U.S. Department of Energy and Department of Commerce announced a public-private partnership with SoftBank Group and AEP Ohio to redevelop the site for advanced computing infrastructure.
According to the DOE, SB Energy will build 10 GW of new power generation at the location, with at least 9.2 GW coming from natural gas. The company is also investing $4.2 billion to upgrade and build new transmission lines in southern Ohio in collaboration with AEP Ohio. The project fully complies with the current administration's "Taxpayer Protection Pledge," meaning U.S. taxpayers will not bear the costs of the power and transmission infrastructure. SB Energy's investment stems from a U.S.-Japan trade and investment framework, under which Japan committed $550 billion in U.S. investments, with $33 billion allocated to this power project.
The developer, SB Energy, is an energy company established by SoftBank in 2019. SoftBank holds a controlling stake, with Ares Management and OpenAI also as shareholders. In January of this year, both OpenAI and SoftBank Group injected $500 million each into SB Energy. According to internal plans, construction on the power generation infrastructure began in Q2 of this year, with operations expected to commence gradually from late 2027 to early 2028. The first phase of the data center, with 800 MW of capacity, is slated for operation in 2028, with the full campus expected to be completed by the mid-to-late 2030s. If fully realized at 10 GW, it would be one of the largest AI data center campuses ever built globally, with a power output equivalent to a large nuclear power plant, capable of meeting the electricity needs of approximately 8 million U.S. households.
Industry Financing Shift: The Rise of Chipmaker Balance Sheet Guarantees
Global AI infrastructure investment is in an unprecedented expansion cycle. Research suggests the combined data center capital expenditure of the four largest U.S. hyperscale cloud providers will exceed $700 billion by 2026. However, a gap is widening between funding supply and investment demand—annual debt issuance required for AI infrastructure is estimated between $500 billion and $1 trillion, while the global high-yield bond market can only absorb about $250 billion per year.
In this context, the "chipmaker balance sheet guarantee" model being used by OpenAI and NVIDIA is emerging as a new paradigm for AI infrastructure financing. The core logic involves a highly credit-rated industry leader (like NVIDIA or Google) providing a financial guarantee to unlock large-scale debt financing, thereby solving the contradiction for high-growth AI startups that lack sufficient credit. Google's backstop for Anthropic and NVIDIA's previous guarantees for CoreWeave leases follow the same logic, allowing capital to flow more efficiently from global debt markets into AI infrastructure.
Deal Structure: NVIDIA's Evolving Role
The true innovation lies in the financing structure. Reports indicate OpenAI will sign a 20-year lease with SB Energy, gaining control over all equipment usage within the campus and beginning rental payments once the project is operational, with cumulative rent expected to reach tens of billions of dollars.
NVIDIA occupies a pivotal position. The company is reportedly in discussions to use its balance sheet to provide credit support for OpenAI's lease payment obligations and SB Energy's future project financing. On the hardware side, NVIDIA will supply all GPU hardware for the facility, ensuring the entire technology stack from foundational compute to applications runs within its ecosystem. Analysis suggests this creates a "capital loop"—NVIDIA provides a financial guarantee to OpenAI, enabling OpenAI to lease NVIDIA hardware, while NVIDIA locks in hardware shipments and long-term rental income through the guarantee. This structure, described as "hyperscaler endorsement," is seen as effectively solving the duration mismatch problem for data centers, where leases often exceed 15 years but payback periods are around 8 years.
Competitive Moves and Strategic Implications
In a parallel development, it was reported that Google is providing a backstop arrangement for Anthropic's roughly $35 billion TPU leasing obligations across five data center locations. This deal involves structured debt issued by an SPV led by Apollo and Blackstone, with Broadcom providing additional residual value guarantees on the senior tranche.
This transaction is a multi-dimensional strategic move within the AI compute competition. For NVIDIA, it represents its most aggressive use of its balance sheet to date, transitioning from hardware supplier to a financial gatekeeper. While NVIDIA previously agreed to purchase up to $6.3 billion in unsold compute capacity from CoreWeave in 2025 and guaranteed its data center lease payments in February of this year, the scale of the potential guarantee for OpenAI is far larger. With quarterly data center revenue of approximately $73.1 billion (annualized around $290 billion), backing a project of this magnitude would significantly expand its balance sheet exposure.
The core logic for NVIDIA is not merely hardware supply but becoming an implicit controller of OpenAI's capital structure. If OpenAI cannot meet its lease payments, NVIDIA would become the "payer of last resort" to SB Energy. This reverse exposure forces NVIDIA to ensure its largest client remains financially healthy, altering the long-term dynamic between them. The guarantee acts as a demand-lock mechanism, ensuring hardware it delivers is utilized and preventing clients from abandoning purchases if utilization drops.
For OpenAI, the lease structure reveals a significant shift in its compute strategy. The company had previously advanced plans for custom AI chips designed with Broadcom, targeting mass production in 2026. However, analysis suggests NVIDIA's financial guarantee model substantially weakens OpenAI's ability to move away from NVIDIA accelerators, as migrating to custom chips or alternatives would require a complete restructuring of its capital arrangement. Furthermore, as OpenAI's IPO application is under SEC review, its massive cloud leasing obligations and the associated financial guarantees will be a core focus for regulators assessing its long-term profitability and financial sustainability.
Energy and Infrastructure Challenges
The project faces significant energy challenges. According to the DOE announcement, 9.2 GW of SB Energy's plan will come from natural gas, with the remaining 800 MW potentially from other sources. While large-scale new natural gas generation faces policy debate under previous clean energy frameworks, it is currently a preferred baseload power source for AI infrastructure under the current regulatory focus on affordability.
A more critical constraint is grid expansion. Estimates suggest the U.S. will need to add about 148 GW of generation capacity by 2030 to meet AI data center demand. However, grid infrastructure expansion is severely lagging, with substation and transmission line construction taking 5 to 13 years. SB Energy has committed $4.2 billion to upgrade transmission lines, but it remains a major challenge whether this investment can be completed in sync with the data center's first phase in 2028.
While the project relies primarily on natural gas, nuclear power is emerging as a long-term energy option for AI infrastructure. Traditional U.S. nuclear plants could start supplying power to data centers by the second half of 2027, and small modular reactors are expected to enter commercial operation from around 2030. Whether the Ohio project's later phases (post-2030) can incorporate clean baseload energy like nuclear will impact its long-term carbon compliance.
Comments