10 gigawatts. The equivalent of Belgium’s entire national electricity consumption. That is the scale of the data center OpenAI is negotiating to lease in southern Ohio - built on the site of a decommissioned Cold War uranium enrichment plant. And Nvidia, the world’s dominant GPU supplier, is reportedly in talks to guarantee $250 billion to make it happen (Wall Street Journal, July 2026).
A Three-Layer Deal Worth $600 Billion
The financial architecture involves three parties:
SB Energy - SoftBank’s power subsidiary - is building the 10-gigawatt campus in Piketon, Ohio, 68 miles south of Columbus. Total campus cost is expected to exceed $500 billion.
OpenAI wants to lease the entire facility. This would mark the first time OpenAI has directly leased a data center, rather than purchasing computing capacity from Microsoft Azure, AWS, or Oracle.
The problem: OpenAI is not yet profitable, which means it lacks an investment-grade credit rating. Without that rating, borrowing at favorable rates for a $500 billion lease is structurally impossible.
Nvidia steps in as a $250 billion guarantor - telling creditors that if OpenAI can’t service its lease obligations, Nvidia will cover it. Separately, Nvidia is in talks to provide up to $350 billion to finance OpenAI’s chip purchases for the campus.
Total potential commitment from one chip company to one AI company: $600 billion.
Phase one targets 800MW by 2028 - enough to power 640,000 homes. Full 10GW scale will take considerably longer. And no deal has been signed yet.
Why Would Nvidia Do This?
This is the most important question the headline numbers obscure.
Nvidia is not doing this from goodwill. By guaranteeing $250B and providing $350B in chip financing, Nvidia is locking OpenAI in as a structurally irreplaceable customer for decades.
When you are someone’s creditor and debt guarantor, the relationship fundamentally shifts. OpenAI cannot migrate to AMD, Google TPUs, or Amazon Trainium at scale when Nvidia holds $600B in financial leverage over the relationship.
This explains why OpenAI has raised its projected infrastructure spending to $750 billion through 2030 - up from $600 billion earlier this year. A significant share of that will be Nvidia chips.
For Nvidia, this is demand lock-in that doesn’t depend on normal customer budget cycles. It’s a hardware company using financial engineering to protect market share.
Breaking From Microsoft - Into Nvidia’s Arms
In 2025, Microsoft relaxed its exclusivity clauses. By April 2026, they restructured the partnership: OpenAI could distribute models through AWS, Google Cloud, and Oracle without being locked to Azure (Technerdo, 2026).
With the Ohio campus, OpenAI wants to go further: operating its own compute rather than renting from hyperscalers. Controlling 10GW of owned infrastructure means OpenAI could theoretically price APIs independently - without paying Azure or AWS their margin.
But the Ohio deal trades one dependency for another. This time, Nvidia is embedded in OpenAI’s balance sheet - not just as a chip supplier, but as a financial partner.
This is a structurally elegant move: Nvidia protects its chip market share not by outcompeting rivals on product, but by becoming financially inseparable from its largest customer.
What This Means for Markets Beyond the US
Most businesses in Vietnam and Southeast Asia access AI through two channels: directly via ChatGPT or OpenAI’s API, or indirectly through Azure OpenAI or AWS Bedrock.
When OpenAI operates its own compute at 10GW scale, it gains full pricing autonomy - in both directions.
The upside case: without sharing margin with Azure and AWS, OpenAI could pass savings downstream to API users. Businesses currently paying $0.15-$3.00 per million tokens could see lower rates as infrastructure economics improve.
The risk case: without hyperscalers acting as competitive checks on pricing, OpenAI sets its own margins. With $750 billion in infrastructure to recover over the next decade, the pressure to monetize aggressively is significant.
The Japan government has already invested $33 billion in natural gas infrastructure under a US trade deal to power this facility. The geopolitics of AI compute are accelerating.
What is certain: the AI infrastructure race is determining who controls compute capacity, who prices AI access, and who holds leverage in those pricing decisions. Vietnam and most of Southeast Asia have no voice in any of these negotiations.
The AI workflows being built today run on infrastructure whose economics are being renegotiated right now - in Ohio, between Nvidia and OpenAI, with no regional stakeholders at the table.
NateCue's Take
What most coverage misses: this isn't an OpenAI story. It's Nvidia converting from chip vendor to strategic creditor - a role change that makes them nearly impossible to replace. When you guarantee someone's debt, you're embedded in their capital structure. AMD and custom silicon alternatives can't compete with $600B in financial leverage. For marketers watching AI pricing: when OpenAI controls 10GW of its own compute by the late 2020s, the Azure/AWS margin layer disappears. That could compress API costs - or simply shift pricing power directly to OpenAI, with no competitive check from hyperscalers. Vietnam and Southeast Asian businesses are building workflows on infrastructure whose economics are being renegotiated right now. You have no seat at that table.