Nvidia to Invest $3B in SB Energy as Ohio AI Data Centre Deal Takes Shape

Nvidia is in talks to invest approximately $3 billion in SB Energy, a SoftBank subsidiary developing a massive data centre campus in Ohio for OpenAI, according to a report by The Information published on 16 August 2026. The investment is one part of a broader deal that also has Nvidia providing up to $120 billion in credit support for the project, down from an earlier figure of $250 billion. A deal could be signed as soon as this weekend, per The Wall Street Journal.

What happened

Detail	Figure
Nvidia planned equity investment in SB Energy	~$3 billion
Nvidia credit guarantee (revised)	Less than $120 billion
Previous guarantee figure	$250 billion
Ohio data centre target capacity	10 gigawatts (GW)
Phase one capacity backed by Nvidia	~5 GW
SB Energy IPO target raise	$5 billion
Potential chip financing deal value	Up to $350 billion
Original MoU signed (Nvidia + OpenAI)	September 2025

Nvidia plans to split its $3 billion equity commitment into two tranches: the first half paid when the Ohio project is formally signed, and the second half timed to SB Energy’s initial public offering (IPO), which could happen as soon as next month.

SB Energy, founded in 2019 and backed by OpenAI, builds large-scale power and data centre infrastructure. The Ohio site is targeting 10 GW of capacity and, according to The Information, could become the largest data centre project ever announced.

How the financing is structured

The deal has several moving parts. Nvidia’s revised credit guarantee of less than $120 billion would cover the first phase of roughly 5 GW. That guarantee backs the data centre lease and the debt needed to build the campus, giving lenders confidence in the project’s financial stability and thereby lowering borrowing costs.

OpenAI is still finalising a binding lease for the full 10 GW, which could be signed within days. Separately, Nvidia is also in talks to finance OpenAI’s purchase of chips, a deal that could total up to $350 billion across the entire project. The US government controls the power supply for the Ohio site, and Japan is providing additional funding under a recent trade agreement.

Why was the guarantee cut from $250B to $120B?

Nvidia officials had raised concerns about the company’s exposure when using its own balance sheet to underwrite demand for its AI chips. Trimming the initial guarantee to less than $120 billion was explicitly aimed at reducing that risk perception among investors, according to The Wall Street Journal’s reporting.

This restructuring follows a longer history of delays. In September 2025, Nvidia and OpenAI signed a memorandum of understanding (MoU) committing Nvidia to help develop at least 10 GW of computing capacity and invest up to $100 billion. That agreement stalled over the same internal concerns about financial exposure.

Why it matters

The scale here is hard to overstate. A 10 GW data centre campus dwarfs most existing facilities by an order of magnitude. If it gets built, it sets a new floor for what “large-scale AI infrastructure” means, and signals that the major players are betting that AI compute demand will stay enormous for the foreseeable future.

For businesses watching the AI market, this deal has a practical implication: chip supply and compute costs are unlikely to fall sharply in the near term. Projects at this scale take years to build, and the financial engineering required (sovereign funding, chipmaker credit guarantees, IPO proceeds) tells you how capital-intensive the AI hardware layer has become. Anyone budgeting for AI workloads in 2026 and 2027 should not count on prices dropping fast.

The SB Energy IPO is also worth watching. A $5 billion raise tied directly to AI infrastructure gives investors a relatively direct way to bet on compute growth outside of Nvidia’s own stock.

Our take

Nvidia guaranteeing demand for its own chips is a clever structure until it isn’t. By providing credit support for a project that will buy Nvidia GPUs, the company is effectively financing its own revenue. That works as long as AI spending stays strong and Nvidia’s balance sheet holds. But it concentrates risk in a way that makes the internal pushback entirely understandable.

For smaller businesses, none of this changes what you should be doing with AI today. The infrastructure story is about hyperscale compute, not the tools you use to automate a workflow or improve a customer-facing product. If you are exploring how to put AI to practical use, the relevant question is whether the models and APIs you rely on will remain affordable and available. Given that multiple cloud providers are racing to build capacity, they likely will. The Ohio project is insurance for that future, not a signal to wait.

Our AI integration work for clients rarely involves GPU clusters. It involves picking the right models, connecting them to real business data, and measuring results. That work is happening now, regardless of how the Nvidia-OpenAI financing lands. For context on how AI infrastructure spending compares to open-weight alternatives, see our earlier coverage of Morgan Stanley’s ROIC analysis on open-weight AI models.

What to do about it

1. Watch the SB Energy IPO date: if it prices next month, the terms will signal how public markets value AI infrastructure at scale.
2. Track Nvidia’s balance-sheet disclosures in its next earnings report for any mention of contingent liabilities from credit guarantees.
3. Review your own AI compute costs now. If you are on a cloud provider with data centre exposure in the US Midwest, capacity expansion should benefit you over the next 12 to 24 months.
4. Do not delay AI adoption waiting for cheaper compute. API pricing from major providers has been falling independently of these infrastructure timelines.

The practical takeaway: the Nvidia-OpenAI deal is infrastructure politics at a scale most businesses will never touch directly, but it does confirm that AI compute will remain well-funded and available through major cloud providers for the foreseeable future.

Source: Bing News · OpenAI