Nvidia Commits Up to $105 Billion in AI Data Center Guarantees as Capital Bottlenecks Loom

Nvidia filed an 8-K on August 17, 2026, confirming residual-value guarantees covering approximately 4.25 IT-gigawatts at SB Energy's PORTS-Pike campus in Pike County, Ohio, with aggregate payment capped at $105 billion. Nvidia retains sole discretion over an additional ~3.8 IT-GW option. OpenAI is the 20-year tenant. SB Energy builds, owns, and operates. Nvidia supplies compute hardware exclusively for initial deployment and is investing $1.5 billion of equity into SB Energy.

Guarantees activate as qualifying capacity enters service, expected from 2028. Nvidia covers the gap between a guaranteed minimum lease value and whatever SB Energy recovers through replacement tenants or asset sales if OpenAI defaults or becomes insolvent.

What Nvidia is actually guaranteeing

The obligation attaches to land, power, and shell economics—physical infrastructure residual value, not individual GPUs. If triggered, Nvidia can assume the lease, require reletting, initiate a sale, or defer for a year while covering specified costs.

OpenAI has agreed to reimburse Nvidia for any payments made. That protection carries obvious wrong-way credit exposure: the circumstances most likely to trigger the guarantee—OpenAI distress—are those in which OpenAI's reimbursement capacity would be weakest. Notably, the guarantees can terminate once OpenAI obtains a satisfactory credit rating (threshold undisclosed), making Nvidia a de facto bridge-credit provider until traditional capital markets underwrite OpenAI directly.

The circularity problem

Nvidia's credit support lowers the weighted average cost of capital for AI infrastructure. Cheaper financing means more projects built, more Nvidia hardware ordered, and stronger cash flows to write the next round of guarantees.

The vulnerability is correlation. A downturn severe enough to trigger these guarantees—weak AI demand, falling GPU utilization, tenant credit deterioration—would simultaneously compress Nvidia's revenue and valuation multiples. Nvidia absorbs guarantee losses exactly when its core business is under pressure.

Scale against the balance sheet

The $105 billion is a maximum contingent amount phased in over years. As of April 26, Nvidia held ~$50.3 billion of cash and marketable debt securities, plus $30.2 billion of marketable equity securities, against $259.5 billion in total assets. The cap runs roughly 2.1× liquid assets and 40.5% of total assets—but only ~1.9% of Nvidia's ~$5.49 trillion market capitalization.

NVDA closed August 17 at approximately $225.01, flat on the day, underperforming the SOXX semiconductor ETF by roughly 1.6 percentage points. The market did not price this as costless secured demand.

Why the revenue arithmetic justifies the risk

Nvidia says one technology generation at PORTS-Pike could involve approximately 1.5 million GPUs and $150–200 billion of Nvidia revenue. OpenAI's broader commitments represent roughly 12 GW of Nvidia compute, potentially reaching 16 GW, with Jensen Huang describing the opportunity through 2030 as approximately $600 billion. A $105 billion contingent backstop behind infrastructure yielding multiples of that across successive refresh cycles has clear commercial logic.

The constraint has migrated

Nvidia's 8-K, paired with the separate $500-billion-plus capital mobilization announced with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, crystallizes one conclusion: the binding constraint on AI deployment has migrated from chip supply to the cost and availability of financing.

Nvidia is converting its balance sheet into a sales-enablement instrument. Each dollar of contingent credit exposure can generate several dollars of GPU gross profit across hardware generations. Every 100 basis points of borrowing-cost reduction on $100 billion of financed infrastructure represents roughly $1 billion per year in financing economics—a subsidy Nvidia can offer without cutting hardware prices.

The second-order consequence matters more than the guarantee itself. Nvidia-backed and hyperscaler-backed facilities will attract cheaper debt. Unbacked merchant developers and subscale neoclouds face tighter capital, producing a financing squeeze without any uniform spread blow-out. Credit bifurcation is the operative dynamic.

The scarce asset in AI infrastructure is no longer the GPU. It is the financeable megawatt—power capacity paired with a creditworthy tenant and credible residual-value support. Nvidia has positioned itself as the provider of that credibility. The August 26 Q2 FY2027 earnings and 10-Q, containing the full guarantee forms, amortization schedules, and accounting treatment, will clarify whether the correct analogy is prudent credit enhancement or vendor financing on a scale the technology industry has not attempted since the telecom build-out of the late 1990s.

not investment advice

Sources: https://d18rn0p25nwr6d.cloudfront.net/CIK-0001045810/988230c7-dd29-466b-92dc-5cf34f65b60f.pdf