ANALYSIS The AI infrastructure boom is scaling faster than the financial and environmental frameworks meant to contain its risks — a gap now visible in uninsured megaprojects, renegotiated backstops, and carbon footprints rivaling national industrial sectors.
Why it matters
Nvidia revised its planned financial guarantee for the Ohio OpenAI campus from $250 billion to less than $120 billion5,12. Meta and BlackRock's $14 billion El Paso data center project is not insured against total loss, exposing lenders to credit risks and big potential liabilities7. An FT analysis of 60 of the largest planned US data centers found they could together produce 101.5 million tons of CO2 per year, or approximately 7% of US power sector emissions in 20256. ◆ Taken together, these developments reveal that the largest data center projects ever attempted are being financed, built, and powered under conditions that leave lenders, investors, and the climate exposed to tail risks that no single party has agreed to absorb.
The big picture
OpenAI signed a 20-year, 10-gigawatt data center deal in Ohio with SoftBank's SB Energy, capping weeks of negotiations over what would be the largest data center project announced to date2. Nvidia agreed to backstop a portion of the value of the completed data center. But the backstop itself was renegotiated downward: Nvidia revised its plans so that it would initially guarantee less than $120 billion, down from the $250 billion previously discussed. The Wall Street Journal reported the change was made after investors raised concerns about Nvidia's risk exposure tied to large financing commitments.
Separately, Nvidia is in talks to invest as much as $3 billion in SB Energy10,11. Nvidia has discussed investing half of the $3 billion when the Ohio project deal is signed and the other half as part of SB Energy's planned initial public offering. SB Energy is aiming to go public as soon as next month and could raise at least $5 billion in the IPO. SB Energy, founded in 2019, develops large-scale power and data center infrastructure projects and is also backed by OpenAI.
Meanwhile, Meta and BlackRock's $14 billion El Paso data center project is not insured against total loss, exposing lenders to credit risks and big potential liabilities. Insurers are balking at the cost of full coverage for gigawatt-scale campuses. And an FT analysis of 60 of the largest planned US data centers found they could together produce 101.5 million tons of CO2 per year, or approximately 7% of US power sector emissions in 2025 — the equivalent of 27 coal plants or 24 million cars per year.
ANALYSIS Nvidia's renegotiation of the Ohio backstop from $250 billion to less than $120 billion is the clearest signal yet that even the company most enriched by the AI buildout is unwilling to underwrite its full scale. The revised guarantee covers only the first phase of the project, meaning the remaining capacity depends on financing structures that do not yet exist. Nvidia's parallel move to partner with six major financial institutions to launch compute financing platforms aimed at raising over $500 billion in third-party capital for AI infrastructure suggests a strategy of distributing risk outward rather than absorbing it on-balance-sheet.
The El Paso insurance gap and the Ohio backstop renegotiation point to the same structural problem: the projects are too large and too novel for existing risk-transfer markets. Insurers balk at total-loss coverage for gigawatt-scale campuses; investors balk at open-ended guarantees. The result is that lenders and equity holders are carrying residual risk that traditional infrastructure finance would have laid off.
The environmental dimension compounds the financial exposure. If the 60 largest planned facilities produce 101.5 million tons of CO2 annually, the regulatory and reputational surface area for these projects widens considerably — particularly for publicly traded entities like Meta and for SB Energy as it pursues an IPO.
Nvidia's dual role — as chip supplier, equity investor in SB Energy, and financial guarantor of the Ohio campus — creates an unusual concentration of exposure to a single asset class. The $3 billion equity stake under discussion would deepen that entanglement just as the backstop terms are being scaled back.
What's next
SB Energy's IPO, targeted for as soon as next month, will be the first market test of investor appetite for pure-play AI data center development at this scale. OpenAI is still discussing a binding lease for the full 10-gigawatt project in Ohio, meaning the final scope — and the final risk allocation — remain unresolved.