NVIDIA appears to be lending its own creditworthiness to AI cloud providers, helping them raise funds and encouraging investment in AI infrastructure. An analysis of the quarterly report (Form 10-Q) the company filed with the U.S. Securities and Exchange Commission (SEC) on August 26, 2026 shows that the maximum total of contracts and guarantees that serve to support customers' businesses and financing reaches $164.5 billion, equal to about 72% of shareholder equity as of July 26. CoreWeave, a "neocloud" that rents out GPUs, raises funds at an effective interest rate of about 10%, while NVIDIA borrows at 4.6%. Against this gap in creditworthiness, a picture emerges in which NVIDIA absorbs part of its customers' risk. However, the $164.5 billion is neither an actual loss nor a forecast of one; it is our own tally of contracts of differing natures. Taking into account NVIDIA's financial strength, including first-half net income of more than $118 billion, we examine how much risk the company is actually taking on.
Of roughly $530 billion in NVIDIA contracts and guarantees, customer-support items total $164.5 billion
NVIDIA's financial support for AI infrastructure operators had already come to light piecemeal. In a Form 8-K dated August 17, 2026, the company announced that it had signed a residual value guarantee agreement of up to $105 billion with SB Energy for a data center in Ohio used by OpenAI. We covered the details of that agreement in our earlier report. On September 9, 2025, NVIDIA also signed a purchase order with CoreWeave with an initial contract value of $6.3 billion, under which it is obligated to buy the company's unsold cloud computing capacity through April 13, 2032. The semiconductor research firm SemiAnalysis focused on the overall picture of such contracts in its September 11 analysis. It added up six line items in NVIDIA's 10-Q and pointed out that off-balance-sheet contracts and guarantees reach about $530 billion.
Adding up the three tables in NVIDIA's 10-Q reproduces that figure. Summing the $366 billion in the ordinary commitments table, $56 billion in the additional commitments table, and $108.5 billion in the guarantees table gives $530.5 billion, which matches SemiAnalysis's figure of about $530 billion apart from rounding. These, however, are not financial burdens of the same nature. The largest item is $279 billion to secure future semiconductor supply and production capacity, and the totals also include $29 billion in cloud service contracts for research and development and $25 billion in not-yet-commenced data center leases intended mainly for NVIDIA's own use. Treating all of these as credit guarantees for customers would misjudge the nature of the risk NVIDIA is actually bearing.
The items whose role in supporting customers' financing and business continuity is relatively clear are two entries in the additional commitments table and the amount in the guarantees table. The breakdown is $36 billion in AI cloud contracts, $20 billion in not-yet-commenced data center leases for third parties, and a maximum total of $108.5 billion in various guarantees, for a total of $164.5 billion. In this article, we classify these three items as "customer-support contracts and guarantees." This is our own classification based on the 10-Q's disclosures; NVIDIA does not book $164.5 billion as a single liability item in its financial reporting. Nor do we assert that every contract in the ordinary commitments table is spending for NVIDIA's own benefit.
The $20 billion in third-party data center leases are ones that NVIDIA signs itself, with the expectation of handing the contracts over to third parties later. According to the 10-Q, the lease terms are about 15 years and are expected to begin in fiscal 2028–2029. If the contracts can be transferred to third parties as planned, the subsequent burden falls on the transferees, but if the handover does not happen, NVIDIA itself could remain liable for the lease payments. In other words, NVIDIA is taking on the contractual burden first for facilities that are meant to be used by third parties.
The $36 billion in AI cloud contracts also functions as a mechanism to support customers' businesses. AI cloud providers procure NVIDIA GPUs and related products, while NVIDIA signs contracts to buy cloud services from those providers. The term is normally six years, but if a provider can sell services to a third party on more favorable terms, it is permitted to stop providing capacity to NVIDIA unilaterally and sell it to the third party. The more capacity is consumed by third-party use or NVIDIA's own R&D purposes, the smaller NVIDIA's contractual burden becomes. In addition, if certain conditions are met, NVIDIA can also receive a share of the cloud revenue earned from third parties.
One example of this business model is the contract with NVIDIA disclosed by SharonAI in an 8-K on June 12, 2026. The contract runs six years with a value of up to $4.88 billion, and SharonAI describes it as a model combining revenue sharing and credit enhancement. By linking NVIDIA's product sales to customers' cloud businesses through long-term contracts, the structure supports customers' expansion while securing demand for NVIDIA's own products.
Up to $164.5 billion is about 72% of equity, roughly ¥26 trillion
The maximum total of customer-support contracts and guarantees under NVIDIA's 10-Q, $164.5 billion, equals about 71.8% of shareholder equity of $228.984 billion as of July 26. The calculation adds $36 billion in AI cloud contracts, $20 billion in third-party data center leases, and a maximum total of $108.5 billion in guarantees, then divides by shareholder equity. Converted at the Bank of Japan's exchange rate of ¥158.34 to the dollar as of 5 p.m. on October 14, 2026, this comes to about ¥26 trillion.
| Customer-support contracts and guarantees | Maximum amount |
|---|---|
| AI cloud contracts | $36 billion |
| Not-yet-commenced data center leases for third parties | $20 billion |
| Land, power, and building guarantees for AI cloud providers | $3.5 billion |
| Guarantee for SB Energy | $105 billion |
| Total | $164.5 billion |
| NVIDIA shareholder equity (as of July 26, 2026) | $228.984 billion |
| Ratio to shareholder equity | About 71.8% |
There are, however, important caveats to this comparison. First, the $108.5 billion in total guarantees includes the $105 billion SB Energy guarantee signed on August 17, after the July 26 quarter-end. The reference dates for the denominator (shareholder equity) and the numerator (contracts and guarantees) therefore do not fully match. Also, the amounts in the guarantees table are "maximum gross exposure," not the amount of losses expected to occur in the future. The SB Energy guarantee takes effect as each lease commences, and the first facility is expected to go into operation in NVIDIA's fiscal 2029.
For the AI cloud contracts as well, NVIDIA's burden falls as third-party usage grows. Moreover, the SB Energy guarantee comes with an indemnification agreement from OpenAI to NVIDIA, and for the separate $3.5 billion land, power, and building guarantee, a partner company has deposited $712 million in an escrow account. The $164.5 billion calculation does not deduct such mitigating factors. In addition, AI cloud contracts and data center leases are agreements that involve consideration in the form of use of services or facilities, and differ in nature from guarantees under which payment arises from a counterparty's default. The figure of about 72% of shareholder equity is a comparative value showing the scale of contracts and guarantees NVIDIA has taken on, and does not mean the company will suffer losses of the same amount.
The composition of the amount is also heavily skewed. Of the $164.5 billion, $105 billion is the SB Energy guarantee tied to the group of data centers in Ohio used by OpenAI. About 64% of the total we classified as customer-support items is thus linked to a single large project. SB Energy, the counterparty to the guarantee, is a SoftBank Group company, so this financing structure is closely relevant to Japanese investors as well.
NVIDIA at a 4.6% effective rate, CoreWeave at about 10%: a support structure built on the credit gap
Behind NVIDIA's ability to support customers' financing is the difference in creditworthiness between NVIDIA and AI cloud providers. Comparing U.S. dollar-denominated senior unsecured notes maturing in 2031, the bonds NVIDIA issued in June 2026 with a 4.50% coupon had an effective interest rate of 4.60%. By contrast, the bonds CoreWeave issued in April of the same year with a 9.75% coupon are listed in the company's quarterly report as having an effective rate of about 10%. The roughly 5-point gap between the two companies shows a large difference in funding costs.
That said, this comparison alone cannot strictly measure the effect of the difference in creditworthiness. The two bonds were issued at different times, and their terms and market conditions were not entirely identical. Also, what is being compared here are effective rates at issuance, not current yields reflecting subsequent market prices. CoreWeave's figure of about 10% likewise relates to a specific bond and does not indicate the average funding rate for neoclouds as a whole. Even so, the fact that NVIDIA can raise funds at a lower rate is an important condition for providing credit enhancement to customers.
The interest rate gap itself does not prove a transfer of credit risk. What matters is the structure of the contracts NVIDIA has taken on. For example, under the purchase order with CoreWeave, if computing capacity at the company's data centers goes unsold, NVIDIA is to purchase that unsold capacity. Thanks to the September 9, 2025 contract with NVIDIA, CoreWeave can secure a certain level of demand, which backs up its business plan and financing. In return, NVIDIA takes on the burden if computing capacity does not sell sufficiently.
It is not possible to tell from NVIDIA's 10-Q how much of the $6.3 billion CoreWeave purchase order is included in the $164.5 billion breakdown we tallied. It should therefore not be added to the $164.5 billion separately. Also, even with such a purchase agreement in place, CoreWeave bore an effective rate of about 10% when it issued its 2031 notes. Even if NVIDIA's support helps customers raise funds, it does not by itself eliminate customers' own credit risk.
The SB Energy guarantee makes the credit-enhancement mechanism clearer still. According to the 8-K NVIDIA filed on August 17, the guarantee covers lease contracts for a group of data centers with a total IT load capacity of about 4.25GW to be built at the PORTS Technology Campus in Ohio. IT load refers mainly to the power consumed by information-processing equipment such as servers and network gear. The cumulative cap on the guarantee is $105 billion, and NVIDIA's payment obligation arises if OpenAI fails to pay rent, or if a default under the lease occurs due to insolvency, among other cases.
Under the guarantee mechanism, if OpenAI defaults, NVIDIA bears the difference between a guaranteed minimum value and the amount recovered through re-leasing, sale, or similar means. By contract, NVIDIA also has options such as taking over the lease, having the lessor look for a new tenant, or proceeding with a sale process. The guarantee period is up to 20 years from the start of each lease, but the guarantee also ends if OpenAI terminates a lease in accordance with the contract or obtains a sufficient credit rating. This termination condition tied to a credit rating plainly shows that NVIDIA's guarantee serves to supplement OpenAI's own creditworthiness.
In short, through contracts to purchase AI cloud computing capacity and through data center guarantees, NVIDIA is taking on part of the risks of insufficient demand and payment default that had been concentrated on the customer side. This may make it easier for customers and business partners to use long-term contracts as backing for financing. If financial institutions factor NVIDIA's ability to pay into their assessments, there is room to raise funds on better terms than if they relied solely on the customers' own creditworthiness. However, the guarantees are limited to the scope set by contract, and NVIDIA is not taking on all of its customers' debts or business risks.
First-half net income of $118 billion shows financial strength, but linked risks remain
The biggest counterargument when assessing NVIDIA's burden is the company's earning power, along with the fact that many of the contractual burdens are conditional. Net income for the first half of fiscal 2027 (the six months through July 26, 2026) reached $118.01 billion, up sharply from $45.197 billion in the same period a year earlier. That profit exceeds the maximum total of guarantees, $108.5 billion. Moreover, the burden under AI cloud contracts declines as third-party use of capacity progresses, and the SB Energy guarantee creates no payment obligation unless events such as an OpenAI payment default occur. Even if a guarantee were actually triggered, NVIDIA's burden would be limited to the scope set by contract, with amounts recovered through re-leasing or sale of the facilities taken into account.
SemiAnalysis also assesses that, as long as NVIDIA's financial base remains strong, the balance of risk and reward in these support contracts favors the company. By supporting customers' capital spending, it can increase GPU sales opportunities and maintain its competitiveness in the AI infrastructure market. CoreWeave also completed an issuance of $4.2 billion in convertible notes due 2033 with a 2.875% coupon on September 22, 2026, which shows that it can still raise funds from capital markets. However, because convertible notes carry a right to convert into shares, their coupons tend to be lower than those of ordinary senior unsecured notes. It is therefore not possible to simply compare the 2.875% rate with the earlier 9.75% bonds and conclude that CoreWeave's creditworthiness has improved significantly.
On the other hand, NVIDIA's strong finances are not the same as the risks it has taken on being small. A particular concern is the possibility of losses arising at the same time across multiple contracts. SemiAnalysis points out that the time when neoclouds become unable to meet their obligations on lease payments or computing capacity purchases is likely to correlate with the time when NVIDIA's own business and cash generation slow. If AI computing demand falls enough that neoclouds lose the ability to pay, NVIDIA's GPU sales could slow at the same time.
This is a risk common to NVIDIA's support contracts. If AI demand is robust, cloud providers' facility utilization rises, and as sales to third parties grow, NVIDIA's purchase burden also falls. By contrast, if demand drops sharply and rents and computing capacity sale prices fall, cloud providers' revenue declines, making it more likely that NVIDIA's purchase obligations and guarantees translate into actual burdens. If NVIDIA's own product demand and earning power deteriorate in the same phase, even contracts of a size that poses no problem in normal times could become a heavy financial burden.
What follows is our editorial analysis. First-half net income of $118.01 billion shows that NVIDIA currently has extremely high earning power. But net income does not directly mean cash on hand, and there is no guarantee that the same profit can be secured in a future downturn. What matters is how far a deterioration in NVIDIA's own earnings would overlap with payments under customer-support contracts when AI demand slows. Losses cannot be forecast directly from the $164.5 billion figure, but the possibility that part of the risk in AI infrastructure investment is being concentrated on NVIDIA's balance sheet needs to be examined continuously.
The "50% of equity" benchmark to test in the next 10-Q
This analysis does not predict that NVIDIA will suffer losses of up to $164.5 billion. What we are focusing on is the structure in which NVIDIA, through contracts, is taking on part of the demand shortfall and credit risk that customers would otherwise have borne. As an indicator to measure its scale continuously, we use the ratio of the total of three items (AI cloud contracts, third-party data center leases, and the maximum total of guarantees) to shareholder equity. Against the current figure of about 71.8%, if this ratio falls below 50%, it would be grounds to revisit our view that risk is concentrated on NVIDIA.
The first chance to check is the 10-Q for the third quarter of fiscal 2027. The previous report was filed on August 26, about a month after the July 26 quarter-end, and the next may be disclosed at a similar time. We need to check whether new contracts and guarantees are added, whether existing commitment amounts decline through third-party use or contract transfers, and how much shareholder equity grows. Even if the ratio falls below 50%, though, we must evaluate separately whether the reason is an actual reduction in risk or simply an increase in shareholder equity.
Another important item to check is the treatment of the SB Energy guarantee for OpenAI. If OpenAI obtains the sufficient credit rating specified in the 8-K and the guarantee's termination condition is met, the guarantee burden of up to $105 billion that NVIDIA has taken on would change substantially. Also, if neoclouds such as CoreWeave become able to issue unsecured notes maturing around 2031 at effective rates in the 5% range without relying on NVIDIA's purchase contracts or credit enhancement, that would be evidence that customers' financing has improved on their own creditworthiness. This too would support the view that the need for NVIDIA to supplement credit is declining.
Going forward, therefore, three changes will be the focus of the assessment. First, whether the ratio of customer-support contracts and guarantees to shareholder equity falls below 50%. Second, whether OpenAI obtains a sufficient credit rating and meets the termination condition of the SB Energy guarantee. Third, whether neoclouds become able to raise long-term unsecured funds at effective rates in the 5% range without depending on NVIDIA's underwriting contracts. If these come to pass, the view that credit risk is concentrated on NVIDIA will need to be revised.
NVIDIA's AI infrastructure strategy has moved beyond selling GPUs to a stage where it is also involved in the financing and demand securing that customers need to deploy those GPUs. Supporting customers' investment secures future product demand, but if the AI market slows, those support contracts could become a burden on the company itself. What will matter going forward is less the sheer size of the up-to-$164.5 billion in contracts and guarantees than how much of an actual burden arises as AI demand changes, and whether it stays within a range NVIDIA's financial strength can absorb.