Nvidia's $3B Power Play: The $500B Ohio AI Campus

Nvidia is in talks to invest as much as $3 billion in SB Energy, the SoftBank-backed developer behind OpenAI's planned 10-gigawatt data center campus in Ohio — a project that could cost $500 billion and would be one of the largest single AI infrastructure builds ever attempted.

The news, reported by The Information and confirmed by Reuters and CNBC on August 15, signals something bigger than one investment: power — not chips — is becoming AI's binding constraint.

The deal in one paragraph

SB Energy, a SoftBank Group subsidiary, is developing a massive AI data center campus on federal land at the Department of Energy's Portsmouth Site in Pike County, Ohio. OpenAI is in talks to lease up to 10 gigawatts of capacity there — roughly the output of ten large nuclear reactors — and the buildout would require about $100 billion in financing, with the total project cost potentially reaching $500 billion including power generation.

Nvidia's role would be to invest up to $3 billion in SB Energy as part of that financing structure. For context, OpenAI and SoftBank each already invested $500 million in SB Energy in January 2026, and the company has signed a 1.2-gigawatt lease with OpenAI for next-generation data centers.

Why this matters: chips were the bottleneck. Now it's power.

For two years, the AI buildout was constrained by GPU supply — who could get Nvidia's chips first. That era is ending.

Today the constraint is electricity. A single 10-gigawatt campus needs as much power as a major metropolitan area. SB Energy's plan for the Ohio site includes building 10 gigawatts of new power generation, at least 9.2 GW of it natural gas, to feed the data center load — a mini power utility built from scratch alongside the compute.

This is why Nvidia is getting involved at the equity level, not just the chip level. Its products are the compute, but the compute is worthless without power. Investing directly in the power developer locks in capacity for its customers — and gives Nvidia a stake in the infrastructure boom its own chips are driving.

The $500B number, explained

The Ohio campus is the centerpiece of OpenAI's Stargate expansion. OpenAI, Oracle, and SoftBank announced five new Stargate sites as part of a $500 billion, 10-gigawatt U.S. infrastructure buildout, and the Ohio project is the largest single component.

The scale is hard to overstate:

• 10 GW of IT load — enough to power roughly 7-8 million homes
• ~9.2 GW of new natural gas generation to serve it
• Federal land — the DOE-owned Portsmouth Site, a former uranium enrichment facility, being repurposed for the AI era
• $100B+ in direct financing, with total project cost estimates up to $500B

No private project of this kind has ever been attempted. The closest comparison is the Manhattan Project — not in secrecy, but in the scale of a single-site energy and industrial buildout.

What's in it for each side

OpenAI gets guaranteed power and compute capacity without building a power company. Its 1.2 GW lease is already signed; the 10 GW campus would make it the largest single consumer of AI compute capacity in the world.

SoftBank and SB Energy get a captive customer with effectively unlimited demand, plus a blue-chip equity partner in Nvidia that brings credibility — and GPUs — to the project.

Nvidia gets three things: revenue (the campus will buy its chips by the tens of thousands), strategic leverage (it controls the power that makes its products useful), and a hedge — if the AI buildout slows, it owns a piece of the energy infrastructure rather than only the depreciating silicon.

The risks

Nothing this large is low-risk.

• Financing risk. A $100 billion financing backstop doesn't materialize overnight. If capital markets tighten, the campus gets scaled back — and the 10 GW headline becomes a 4-5 GW reality.
• Power politics. Natural gas generation at this scale faces permitting, environmental, and community opposition. The DOE land helps, but 9.2 GW of new gas is a political target.
• Demand risk. The entire thesis assumes AI compute demand keeps growing at current rates. If model efficiency improves faster than expected — or if the AI bubble narrative returns — the campus could be overbuilt.
• Execution risk. No one has built a single-site energy-plus-compute project this large. Timelines will slip; the question is by how much.

Bottom line

Nvidia's $3 billion SB Energy talks are small relative to its balance sheet — roughly a rounding error against its cash position — but they are strategically enormous. They confirm that the AI buildout has entered its energy phase, where the winners are the companies that control power, land, and compute together.

The Ohio campus, if built at anything close to 10 GW, would reshape the U.S. power market, the AI infrastructure market, and the competitive balance between hyperscalers in one stroke. Watch for the final Nvidia-SB Energy terms — and for how much of the $500 billion actually gets financed.