What it means that Nvidia is negotiating a $3 billion investment in SoftBank's SB Energy for OpenAI's data center — August 16, 2026
· CompaniesAutomation
Flash edition: Nvidia negotiates investing up to $3 billion in SB Energy (SoftBank), the subsidiary building OpenAI's Ohio campus, while cutting its guarantee from $250 to under $120 billion. Circular financing, energy as the bottleneck, and what to do with your AI contract.
Flash edition. Nvidia is negotiating to invest up to $3 billion in SB Energy, the energy subsidiary of SoftBank that is building OpenAI's macro data center in Ohio. This is the smallest figure in the deal, yet it best explains how the AI you use is being financed. Complement to today's Radar.
What happened
According to a report by The Information and picked up by Reuters on August 15, Nvidia has discussed putting up to $3 billion into SB Energy: half upon signing the Ohio project and the other half within the IPO the subsidiary is preparing for next month, through which it aims to raise at least $5 billion (Source). This equity entry is the small piece of a much larger negotiation: about $100 billion in credit support from Nvidia for the campus (Source). And it comes with the fine print trimmed: on Friday, the Wall Street Journal reported that Nvidia will initially guarantee less than $120 billion, down from the $250 billion discussed in July, and only for the first phase of a 10-gigawatt construction on Department of Energy land in Pike County, Ohio (Source). Separately, GPU financing is being negotiated that could reach $350 billion. Neither Nvidia nor SB Energy have confirmed anything.
Why it matters
This is circular financing in plain sight: the chipmaker provides money and guarantees so that its customer can buy its chips, and in the process takes a stake in the entity that supplies the electricity. The setup works as long as demand grows and credit remains cheap. The relevant part for your company is not the stock market drama, but where the price you pay per token comes from: today's inference cost is sustained by financial engineering and infrastructure that does not yet exist. The cut in the guarantee from $250 billion to less than $120 billion is the first public sign that even Nvidia is measuring that risk. And there is a second message: the bottleneck is no longer chips, it's energy. That's why the world's largest GPU manufacturer is buying a piece of a utility company.
For your company
Three concrete decisions, none urgent but all cheap to make now. One: stop building the business case on today's price. Calculate the cost per completed task—not per million tokens—and check what margin you have left if the API becomes 50% more expensive. If the project only works with current prices, it's not a project, it's a bet. Two: if you are signing or renewing a contract in the coming months, ask for a price cap for 12-24 months, prior notice of rate changes, and data and prompt portability. It's a good time: providers want committed volume to justify precisely these investments. Three: real portability, not declared. Isolate model calls behind your own layer and maintain a set of evaluations that you can run against two different providers in an afternoon. If you prefer that architecture built and audited without spending a quarter on it, an AI consultancy can get it resolved in weeks.
Frequently Asked Questions
Does this mean AI prices will go up?
Not in the short term, and probably the opposite: while the race for capacity lasts, providers will continue to compete downwards. The risk lies in the medium term, when that infrastructure needs to be amortized. That's why it's wise to lock in contract prices now instead of assuming the downward trend is permanent.
My company is not an OpenAI customer. Does this still affect me?
Yes, indirectly. Nvidia, energy, and data center capacity are the common ground for all providers: Anthropic, Google, or any open model you run in the cloud depend on the same chain. If it gets stressed, it gets stressed for everyone.
Is this a bubble?
The question has no useful answer for an SME. The one that does is: does your automation still make sense if the cost of AI doubles? If the answer is yes, the bubble is the investors' problem, not yours.