← Back to feed

Nvidia is in advanced talks to provide $250 billion in financing guarantees for a 10-gigawatt AI data center in southern Ohio, according to a Wall Street Journal report. The full project — being developed by SoftBank through its energy subsidiary — is estimated to cost as much as $500 billion and would be the largest compute infrastructure project ever announced.

OpenAI is in advanced talks to lease the site. Anthropic, Microsoft, and Google have also contacted Commerce Secretary Howard Lutnick about access to the project’s capacity.

Scale

10 gigawatts is roughly ten times the size of the largest individual hyperscaler data center currently in operation. The figure is large enough that it reframes the question: this is not a data center. It is a compute utility. SoftBank developed the site through its energy subsidiary, which provides the power infrastructure. Nvidia’s role is to backstop the financing that makes the lease viable for a tenant like OpenAI.

For context:

• The entire US data center market currently consumes approximately 50 to 70 gigawatts
• The largest previously announced single AI project was OpenAI’s Project Camellia in Georgia, reported at $750 billion across multiple campuses
• Nvidia’s $250 billion guarantee would represent roughly 2.5 times its current annualized revenue

Nvidia’s Strategic Position

Nvidia’s involvement is not philanthropic. By guaranteeing project financing, Nvidia positions itself as the default GPU supplier for the compute that will run inside the facility. A 10GW facility running Nvidia’s latest Vera Rubin racks would represent tens of billions of dollars in ongoing GPU sales. The financing guarantee effectively pre-sells the hardware.

The move also pushes back against a risk that has been building in 2026: hyperscalers and AI labs moving toward custom silicon (Broadcom, Marvell, Amazon Trainium, Google TPUs) as an alternative to Nvidia hardware. Anchoring a $250-billion-financed compute facility to Nvidia’s supply chain makes switching costs structural rather than contractual.

The Competition for Capacity

Anthropic, Microsoft, and Google contacting Howard Lutnick about the Ohio project is not a coincidence. Each lab faces the same constraint: training runs for frontier models have been outgrowing available compute for two consecutive years. A 10GW site with guaranteed financing changes the calculus on what is trainable. The lobbying activity suggests each company sees the Ohio site as potentially the most important compute resource of the decade and wants access before it is locked by OpenAI.

Lutnick’s involvement reflects the project’s geopolitical dimension. The current administration has made AI infrastructure investment a domestic priority, and a SoftBank-NVIDIA-OpenAI project of this scale is exactly the kind of commitment the Commerce Department has been courting. The approval process for power interconnections, permitting, and potentially direct government contracts on the compute would all run through Commerce.

SoftBank’s Bet

Masayoshi Son has been positioning SoftBank as the infrastructure layer for the AI era since 2023. SoftBank’s Vision Fund made early bets on OpenAI and Arm; this project extends the pattern into physical compute. Developing a 10GW site through an energy subsidiary means SoftBank captures the land, power, and infrastructure margin — not just the equity in the tenant.

The timing follows SoftBank’s announced $100-billion-plus US AI investment commitment made at the White House in January 2026.

Key Numbers

• Site capacity: 10 gigawatts (largest single AI compute project ever proposed)
• Estimated project cost: up to $500 billion
• Nvidia financing guarantee: $250 billion
• Primary tenant in talks: OpenAI
• Other parties in contact: Anthropic, Microsoft, Google
• Developer: SoftBank via energy subsidiary, southern Ohio
• Source: Wall Street Journal report, July 27, 2026