Nvidia’s $108.5B Guarantees Shift AI Financing Risk
Nvidia isn’t just selling AI chips anymore. It has disclosed $108.5 billion in guarantees backing data-centre buildouts — which means the boom is now being financed by its biggest winner.
Nvidia has put $108.5 billion of guarantees behind AI infrastructure. That is not a chip company selling shovels in a gold rush. That is the bloke selling shovels quietly guaranteeing the mine’s rent if the miners can’t pay.
And if you own Nvidia, a broad US index fund, or anything remotely exposed to the AI trade, you need to understand what changed this week.
Nvidia’s $108.5 billion shift from supplier to financier
On August 26, Nvidia reported second-quarter fiscal 2027 revenue of $96.2 billion, up 106% from a year earlier. Data-centre revenue was $89 billion, up 117%. It generated $59.7 billion in quarterly net income. These are absurd numbers in the best possible sense: a real business, with real customers, producing real cash at a scale most companies will never see. ([nvidianews.nvidia.com](https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2027))
But buried in Nvidia’s quarterly filing was the more interesting number: a maximum gross exposure of $108.5 billion in guarantees tied to AI infrastructure.
That includes $105 billion in guarantees for SB Energy’s PORTS Technology Campus in Pike County, Ohio, backing leases for about 4.25 gigawatts of IT load for an OpenAI affiliate. It also includes $3.5 billion in land, power and shell guarantees for selected AI-cloud partners. The Ohio project is expected to begin coming into service in Nvidia’s fiscal 2029, and its guarantees are phased as nine data centres are completed. ([investor.nvidia.com](https://investor.nvidia.com/files/doc_financials/2027/NVDA-2027-Q2-10Q-Final-including-exhibits.pdf))
Let’s be precise, because precision matters when the zeros get silly. Nvidia has not written a $108.5 billion cheque. These are contingent obligations: Nvidia pays only if particular conditions are met and a tenant defaults, and the obligations cover defined parts of lease and power payments rather than every cost at the sites.
Still, contingent does not mean imaginary. A guarantee is a promise. It sits there quietly until it matters very loudly.
Nvidia says the guarantees can end if OpenAI achieves a satisfactory credit rating or when the applicable lease ends. It also has an option to provide support for another 3.8 gigawatts as the Ohio campus scales. ([investor.nvidia.com](https://investor.nvidia.com/files/doc_financials/2027/NVDA-2027-Q2-10Q-Final-including-exhibits.pdf))
That is the headline: the company at the centre of the AI infrastructure boom is no longer merely supplying the expensive hardware. It is helping make the projects financeable.
Why Nvidia is doing it — and why the market liked it
The simple answer is demand is outrunning the boring stuff needed to satisfy demand.
Everyone sees the chips. Fewer people see the constraints behind them: suitable land, enough power, completed buildings, financing capacity, memory supply and long-term customer contracts. You cannot deploy a rack of GPUs into a PowerPoint presentation. Someone needs to build the facility, connect the electricity and fund the whole exercise before a single AI token earns a cent.
Nvidia’s filing says exactly that: land, power, shell and capital are crucial to customer deployments, while smaller and less-capitalised AI companies may struggle to secure the long-dated infrastructure contracts and investment-grade funding they need. ([investor.nvidia.com](https://investor.nvidia.com/files/doc_financials/2027/NVDA-2027-Q2-10Q-Final-including-exhibits.pdf))
So Nvidia is stepping in where traditional project finance is too slow, too conservative or too fussy for the pace of the AI arms race.
Commercially, I understand it. If you have a product customers desperately want, but their bottleneck is the building around your product, you help remove the bottleneck. That is not madness. It is good operating.
Nvidia also forecast $108 billion, plus or minus 2%, in third-quarter revenue. More unusually, it forecast roughly 70% revenue growth for the next fiscal year ending January 2028, compared with the roughly 44% growth analysts had expected before the announcement. ([investing.com](https://www.investing.com/news/stock-market-news/nvidia-forecasts-quarterly-revenue-above-estimates-4877887))
It expanded its AWS partnership too, with Amazon set to deploy another 2 million Nvidia GPUs across its global infrastructure in 2027 and 2028. ([investing.com](https://www.investing.com/news/stock-market-news/nvidia-forecasts-quarterly-revenue-above-estimates-4877887))
That helps explain why investors initially treated the disclosure as part of a very bullish story. Nvidia is seeing so much demand that it is using its balance sheet and credibility to turn potential GPU demand into actual, installed capacity.
Fair enough. But “bullish” and “risk-free” are not synonyms. They have never been synonyms, despite what a hot market tells itself after three drinks.
The overlooked angle: Nvidia is creating demand and underwriting it
This is the bit investors should not skip because the earnings numbers are exciting.
When a supplier invests in, lends to, or guarantees the infrastructure of customers that then buy its products, the relationship becomes more complicated. Nvidia is not just observing AI demand from the sidelines. It is actively helping to create the conditions that let some of that demand happen.
That does not mean the demand is fake. Nvidia’s revenue, margins and cash generation are plainly real. Its second-quarter gross margin was 75%, and it returned about $26 billion to shareholders through buybacks and dividends during the quarter. ([nvidianews.nvidia.com](https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2027))
But it does mean investors should ask a better question than, “How many GPUs can Nvidia sell?”
Ask this instead: How much of the next wave of GPU spending depends on Nvidia making the financing, property and power maths work first?
Nvidia disclosed $99 billion of equity investments and $25 billion of equity-investment commitments as of July 26. It also said it had signed preliminary memorandums with large capital providers intended to mobilise more than $500 billion of third-party capital over time for AI infrastructure. Nvidia was clear those preliminary arrangements may not become definitive agreements. ([investor.nvidia.com](https://investor.nvidia.com/files/doc_financials/2027/NVDA-2027-Q2-10Q-Final-including-exhibits.pdf))
That is a powerful ecosystem strategy. It is also a concentration of risk.
If AI customers keep growing into their contracts, Nvidia looks clever: it secured capacity early, locked in an ecosystem and protected a massive hardware opportunity.
If weaker AI-cloud operators cannot monetise their compute, if capital dries up, if power projects run late, or if AI spending gets rationalised, Nvidia’s role has shifted from vendor to stakeholder in the mess.
That is the difference. A pure supplier can say, “Sorry mate, we sold the product.” A supplier that guarantees the lease has more skin in the outcome.
Reuters reported on August 27 that Nvidia had paused some agreements in a newer program that offered credit support to AI-cloud companies in exchange for a share of revenue, following concerns that the structure could invite antitrust scrutiny. Nvidia said its broader business model to expand compute access remains in place and is continuing to evolve because demand is high. ([investing.com](https://www.investing.com/news/stock-market-news/nvidia-pauses-revenuesharing-deals-with-ai-cloud-companies-wsj-reports-4880320))
That pause matters. It tells you management is pushing hard, but it is not blindly charging into every financing structure put in front of it. Good. They should be cautious. When your market value and your customers’ funding ecosystem become tightly intertwined, regulatory and reputational risk arrive before the invoice does.
This is not a reason to dump Nvidia — it is a reason to stop worshipping it
The lazy bear case is that Nvidia is propping up its own buyers and therefore the whole AI boom is rubbish. That is too cute by half.
The company’s financial performance is extraordinary, and the demand base is broader than a handful of hyperscalers. Nvidia says AI labs could represent about one-quarter of its business next year, while AI clouds, enterprises, sovereign buyers and industrial customers are expanding. ([investing.com](https://www.investing.com/news/stock-market-news/nvidia-forecasts-quarterly-revenue-above-estimates-4877887))
The lazy bull case is equally daft: Nvidia has beaten expectations, so every risk is irrelevant.
No. The correct view is that Nvidia has earned the right to be admired, not the right to be owned without thinking.
For individual investors, this is a position-sizing issue more than a prediction issue. If Nvidia is 2% or 3% of a diversified portfolio, you can appreciate the upside while accepting that one company has real operational, financing, supply-chain and regulatory risks.
If it has become 20%, 30% or more because it has had a belter of a run, you do not own a diversified portfolio. You own a very successful bet that is now trying to convince you it is a permanent personality trait.
The same applies to index investors. Nvidia’s results matter beyond Nvidia because it is a heavyweight in major indices and a barometer for the wider AI trade. Axios put it neatly: Wall Street reads Nvidia’s quarterly numbers as a test of whether the AI trade is slowing. This quarter, the numbers reassured investors — but Axios also noted the stock had fallen the day after five of its previous six earnings reports. ([axios.com](https://www.axios.com/2026/08/27/nvidia-ai-earnings-stock))
That is the market reminding you that a brilliant company and a predictable share price are entirely different things.
What this means for you
First, do a proper concentration check today. Not a vibe check. Open your brokerage account and calculate what percentage of your investable portfolio sits in Nvidia, AI-chip stocks, AI-cloud names and broad US technology funds. Add them together. Most people are less diversified than they think.
Second, separate the business from the stock. Nvidia can keep growing like mad while its shares still deliver a rough year if expectations, interest rates, margins or financing risks move the wrong way. Great companies are not always great purchases at every price.
Third, pay attention to the infrastructure layer. Nvidia’s filing makes clear that the next bottlenecks are not just silicon. They are power, land, completed data centres, memory and financing. The people making money from AI will not all have “AI” in their ticker symbol.
Finally, don’t confuse a guarantee with a disaster — but don’t ignore it because it is hidden in footnotes. Nvidia’s $108.5 billion exposure may turn out to be a masterstroke that accelerates years of profitable growth. It may also be the moment the market realises the AI boom needs more than amazing chips to keep moving.
Either way, the lesson is useful: when the supplier starts financing the ecosystem, stop looking only at the sales chart. Look at who is carrying the risk when the music slows down.