Two people familiar with the negotiations revealed that OpenAI is entering an in-depth negotiation stage and plans to lease a data center park with a planned total scale of 10 gigawatts on federal land in Ohio. This cooperation is also expected to receive financial support from Nvidia. People familiar with the matter said that if the park is fully built based on the current prices of chips, manpower, electricity and building materials, the total investment in the park will be at least US$500 billion. OpenAI will sign a long-term lease agreement and control all equipment in the computer room; the first phase of the project is expected to be put into operation in 2028. After it is put into operation, OpenAI will pay the rent on schedule.

Negotiations have not yet been finalized and plans may change. If the cooperation is implemented, this Ohio project will become OpenAI's largest infrastructure investment to date; even if it has signed large server rental orders with cloud vendors such as Microsoft, Amazon, and Oracle in the next ten years, building a large-scale computing power center by itself is enough to reflect its strong willingness to control its own computing power. This is reminiscent of the "Stargate Project" announced by OpenAI at the White House in January 2025: It was originally intended to form a joint venture with Oracle and SoftBank to invest in data center production capacity worth US$500 billion, but the plan ultimately came to nothing.

The new project site is located in southern Ohio and belongs to the U.S. Department of Energy. SB Energy, a company controlled by SoftBank Group, is currently responsible for the early development. In the cooperation plan negotiated by the two parties, all computer rooms are equipped with NVIDIA hardware; NVIDIA will rely on its strong balance sheet to provide a back-up guarantee to provide financial guarantee for OpenAI's rent repayment and SB Energy's follow-up project financing.

For Nvidia, this cooperation model is also a new attempt. NVIDIA has long expressed its willingness to support customers' data center financing, but has never undertaken a project of this size before.

Nvidia's chip competitor Google has already done something similar: Google provided debt and rent guarantees for Anthropic's data center project leasing Google tensor processors.

Preparations for project start

SB Energy held a groundbreaking ceremony for the project as early as March this year, attended by SoftBank founder Masayoshi Son, U.S. Secretary of Energy Chris Wright and other high-level government and enterprise officials; however, OpenAI and Nvidia’s entry plan has never been disclosed to the public before.

Many places across the United States are promoting gigawatt-level large-scale data centers, such as the Abilene, Texas, computing center operated by Oracle Cloud for OpenAI. The power supply scale of 10 GW is equivalent to 4.5 times the power generation of the Hoover Dam, and its size far exceeds the existing major computer rooms. The completion time of the entire park has not yet been determined, and the first phase of the 800MW computing power computer room is expected to be delivered in 2028.

Many of OpenAI’s previous giant infrastructure ideas have failed to come to fruition.

Last year, OpenAI and NVIDIA reached a preliminary agreement: NVIDIA would invest up to US$100 billion to help OpenAI build its own data center with a total scale of 10 GW, and also support GPU leasing cooperation. The two parties ultimately did not implement the agreement; earlier this year, Nvidia completed a US$300 billion equity investment in OpenAI.

When the Stargate project failed at the beginning of this year, OpenAI shelved its short-term plan to build its own computing center. Nowadays, the potential cooperation with SB Energy can greatly improve the control of its own computing power and is also expected to reduce the cost of computing power. Sam Altman’s long-term goal has been to build a multi-gigawatt computing cluster in the same park. As early as early 2024, OpenAI had negotiated with Microsoft about a large computer room project costing US$100 billion.

In the long run, the cost of self-owned computing power may be lower, but the initial investment is still very high.

People familiar with the matter said that during the 20-year lease period of the Ohio site, OpenAI’s rental expenses alone will be at least hundreds of billions of dollars. Hardware such as chips and servers usually account for 70% of the total cost of data centers. Based on this calculation, the purchase of Nvidia AI chips will cost approximately US$350 billion.

Two people familiar with the matter revealed that OpenAI is negotiating for supporting financing for chip procurement. It is temporarily unclear whether Nvidia or other institutions will provide chip leasing or purchase loans.

OpenAI is urgently preparing for the supply of power and chips required for the next generation of AI systems, while striving to achieve sustainable funding. In the next five years, the cloud computing power leasing contracts it has signed with Oracle, Microsoft, and Amazon Cloud Technology total at least US$665 billion. OpenAI has secretly submitted its IPO application on Monday, and this huge amount of computing power procurement expenditure is bound to become the core focus of investors' review.

OpenAI is Nvidia's largest customer. It has also finalized purchase agreements for AMD chips and Amazon Trainium chips, and has also joined hands with Broadcom to develop customized AI chips.

Site Transformation: The Rebirth of a Former Nuclear Weapons Raw Materials Factory

SB Energy was established in 2019, with SoftBank as the major shareholder, and investors also include Ares Asset Management and OpenAI. OpenAI announced in January this year that it would invest US$10 billion in SB Energy and entrust it to build and operate a 1.2 GW data center in Milam County, Texas. SB Energy disclosed in May that it planned to IPO in the United States, with news that its target valuation exceeds US$500 billion.

The Ohio project originated from an agreement between SB Energy and the former Trump administration: the company transformed an abandoned uranium enrichment plant in Pike County, Ohio, more than 50 miles south of Columbus. U.S. Department of Energy employees Tim Walsh and Ankur Bansal took the lead in promoting the project. Following the instructions of Energy Secretary Wright, they began to revitalize land owned by the department last year for the construction of AI supporting power plants and data centers.

The pace of negotiations accelerated significantly in February this year. The U.S. Department of Commerce officially announced that as part of Trump's new trade agreement with Japan, SB Energy will spend US$33 billion to build a new 9.2 GW natural gas power plant in Ohio.

The property rights of the power plant are owned by the U.S. government and operated by SB Energy; the company promises to fully bear the cost of power grid upgrades and not pass the costs on to the public.

At the groundbreaking ceremony in March, Commerce Secretary Howard Lutnick was also present. Officials vigorously promoted the giant AI data center equipped with this power plant, saying it was among the world's top AI infrastructure projects; they also emphasized that this land, which used to produce nuclear weapons raw materials, has been commercially revitalized and has great revitalization value.

The ownership of the land by the U.S. Department of Energy is a major advantage and can reduce the public resistance that is common for large data centers across the United States. Most commercial computer room sites often face community protests and local planning board approval resistance.