"AI shovel sellers" become "AI infrastructure bank"! Nvidia (NVDA.US) is reportedly in talks to provide a $250 billion financial guarantee for OpenAI to lease data centers

According to The Smart Finance, citing sources familiar with the matter, Nvidia (NVDA.US) is in talks to provide around $250 billion in guarantees to help OpenAI lease computing resources in a major data center project in southern Ohio, USA.

The report states that the deal would enable the AI lab to lease resources from a 10-gigawatt data center hub currently being developed by SoftBank Group in Ohio. SoftBank’s founder Masayoshi Son previously said the project could cost around $500 billion. Nvidia’s guarantee would help SoftBank secure additional financing to support the construction of the data center.

According to the report, OpenAI has been negotiating for weeks with relevant parties to lease capacity from the data center. As planned, this data center complex will become one of the world’s largest computing infrastructures, serving as a landmark project that reflects the massive compute demand of AI development. It is also a core project promoted by both SoftBank and the Trump administration, with the latter calling it a major victory for investment.

Panorama of the 10GW Project

The proposed data center is strategically sited at the former Portsmouth Uranium Enrichment Plant in Pike County, southern Ohio—a facility that once supplied weapons-grade uranium for the US nuclear arsenal during the Cold War before shutting down in 2001. On March 20, 2026, the US Department of Energy and the Department of Commerce jointly announced a public-private partnership with SoftBank Group and AEP Ohio to redevelop the site for advanced computing infrastructure.

According to information released by the Department of Energy, SB Energy will build a new 10GW power generation facility on the site, with at least 9.2GW coming from natural gas, and will invest $4.2 billion with AEP Ohio to upgrade and expand transmission lines in southern Ohio. The project fully complies with the Trump administration’s "taxpayer protection pledge," with the costs of electricity and transmission infrastructure not borne by US taxpayers. SB Energy’s generation investment stems from a US–Japan trade and investment framework agreement under the Trump administration, under which Japan pledged $550 billion of investment in the US, including $33 billion for this power generation project.

The development entity, SB Energy, is an energy company founded by SoftBank in 2019, with SoftBank as the controlling shareholder and Ares Management as well as OpenAI among its shareholders. In January this year, both OpenAI and SoftBank each injected $500 million into SB Energy. Per SB Energy’s internal plans, construction of the power infrastructure began in Q2 this year, with phased operations expected from late 2027 to early 2028. Phase one of the data center, with 800MW of capacity, is expected to go online in 2028, with full build-out of the campus anticipated to continue into the mid-to-late 2030s.

If the campus reaches the planned 10GW capacity, it will be one of the largest AI data center campuses in history. Ten gigawatts equals the full output of a large nuclear power plant and could supply power to about 8 million American households.

Concerns Over Circular Financing Remain

Notably, this potential transaction completely squashes previous market rumors that "Nvidia is scaling back its $100 billion-level support to OpenAI." Instead of shelving a $100 billion direct investment due to OpenAI’s IPO plans, Nvidia is pivoting to a $250 billion project finance guarantee—substantially increasing the scale and substance of its capital commitment.

However, the deal also highlights the circular structure behind the ever-larger transactions in the AI industry—big tech companies provide capital for the projects, and these projects, in turn, generate business for those companies. Nvidia has already invested billions of dollars into multiple AI industry players. CEO Jensen Huang aims to accelerate adoption of the company’s own products and remove bottlenecks that could hinder AI application growth.

Nvidia is also investing $1 billion independently in Korea’s Naver to support its AI data center construction. At the same time, Nvidia is working with Korea’s SK Group to build AI data centers exceeding 2GW capacity. For Nvidia, deepening ties with SK Group helps to strengthen its access to high-bandwidth memory (HBM) chips.

Jensen Huang has said that investments in firms like Anthropic PBC and OpenAI not only benefit Nvidia’s own business development but also generate investment returns. Nonetheless, critics worry these actions could artificially create and amplify demand in the AI sector. In recent weeks, investors have grown cautious about high valuations of tech stocks, fearing current compute buildouts may outpace future AI service demand needs.

Nvidia Morphs Into an “AI Infrastructure Bank”

Global AI infrastructure investment has entered an unprecedented expansion cycle. According to research agencies, the combined data center capital expenditure of America’s four hyperscale cloud providers will exceed $700 billion by 2026. However, the gap between capital supply and infrastructure demand is widening—annual debt issuance required for AI infrastructure is about $500 billion to $1 trillion, while the global high-yield bond market can only absorb about $250 billion per year.

Against this backdrop, the “chip vendor balance sheet guarantee” model, as adopted by OpenAI and Nvidia, is emerging as a new paradigm for AI infrastructure financing. Its core logic: industry-leading companies with the highest credit ratings (such as Nvidia or Google) act as financial guarantors to leverage large-scale debt financing, thereby solving the "high growth but insufficient credit" dilemma for AI startups. Google’s backstop financing for Anthropic, and Nvidia’s prior guarantee for CoreWeave’s lease, follow the same logic. These arrangements let capital flow more efficiently from global debt markets into AI infrastructure.

Some analysts point out that this setup essentially forms a “capital closed loop”—Nvidia provides a financial guarantee to OpenAI, OpenAI secures capital to lease Nvidia’s hardware, and Nvidia, through the guarantee, locks in hardware shipments and long-term rental income. SemiAnalysis summarizes this structure as “hyperscaler endorsement,” stating that it is effectively addressing the duration-mismatch issues in data centers, where lease terms exceed 15 years and payback periods are around eight years.

Notably, competitors are following suit. As previously reported, Google is providing backstop arrangements for Anthropic’s TPU leasing obligations of about $35 billion, spanning five data centers across New York, Texas, Louisiana, and Indiana. This transaction is supported by structured debt via an SPV led by Apollo and Blackstone, with Broadcom providing residual value guarantees on senior tranches.

Nvidia spent three years building the CUDA moat, making it the default accelerator for AI training. But the real engine is not the software stack, it’s the financing layer. When Nvidia guarantees leasing projects, it embeds itself into the capital structure of every large AI facility, a position far harder to replace than any codebase. Thus, Nvidia is no longer just the default supplier; it has effectively become the lender. In an industry where construction capital is scarce, Nvidia’s position is extremely solid.

However, financial guarantees have a dual nature—they lock in demand but also create off-balance-sheet risk exposures. If OpenAI fails to achieve profitability quickly enough to fulfil the $500 billion lease commitment, or if AI inference’s economics cannot justify the capital spent on these facilities, Nvidia will be liable. The CoreWeave transaction capped its risk at $6.3 billion, but the exposure here could be magnitudes higher.

From an industry perspective, this potential deal will further reinforce Nvidia's moat in the AI compute sector. Combining Nvidia’s recently disclosed supplier purchase commitments of up to $95.2 billion with around $90 billion in ecosystem investments over the past 16 months—covering over 145 AI industry companies—Nvidia is attempting to solidify ultimate dominance in next-generation AI infrastructure through financial means, hardware supply, and energy procurement barriers. This not only secures long-term demand from the largest clients, but also significantly raises the entry barrier for rivals such as AMD and Intel.