Nvidia’s $518 Billion of Commitments and $118 Billion of Guarantees: Reading the Fine Print of the Q3 CFO Commentary

Nvidia’s Q3 CFO commentary lists about $518B of commitments and $118B of guarantees. What the fine print says, how it changed, and why AI-capex watchers care.

Most coverage of Nvidia’s November 17 report focused on revenue and guidance. The more unusual disclosures sit a few pages further into the CFO commentary, under the headings “Commitments,” “Additional Commitments” and “Guarantees.” Taken together, they show a company that has signed up for $518 billion of future supply, cloud, lease, equity and capital spending obligations, has agreed to a further $32 billion of contingent purchases from neocloud partners, and carries a maximum gross exposure of $118.0 billion under guarantees that support the construction of data centers it does not own.

On the same day, Nvidia announced expanded long-term agreements with packaging and substrate suppliers and created a board committee dedicated to national security and export controls. This article sets out what the documents say about each item, how large the figures are relative to the company’s own revenue and balance sheet, and which questions the disclosures leave open. It does not assess whether the commitments are prudent; it describes them.

Key takeaways

  • Supply commitments rose to $390 billion at October 25, 2026, from $279 billion a quarter earlier, which the company says is “primarily related to the multi-year procurement of memory supply.”
  • Cloud service agreements and data center lease commitments together reached $102 billion, up from $74 billion. The leases have terms of up to twenty years and are expected to commence between the fourth quarter of fiscal 2027 and fiscal 2033.
  • Including $20 billion of equity investment commitments and $6 billion of capital expenditures, the commitments table totals $518 billion, of which $222 billion falls in fiscal 2028.
  • Separate neocloud agreements of $32 billion take effect only if the partners secure financing and the capacity is deployed and accepted.
  • Maximum gross exposure under guarantees is $118.0 billion, of which $105.0 billion relates to credit support for SB Energy’s PORTS-Pike Technology Campus in Ohio, which will host Nvidia infrastructure under 20-year leases to OpenAI. Nvidia says it has made no payments under any guarantee to date.
  • New or amended supplier agreements cover Kinsus, Samsung Electro-Mechanics, SPIL and Amkor, with capacity additions running into 2028 and mid-2029.
  • A new board-level National Security and Export Control Oversight Committee will meet at least quarterly.

The commitments table

The CFO commentary presents future commitments by fiscal year. Nvidia’s fiscal year ends in late January, so “2028” in the table refers to the year ending in January 2028, not calendar 2028.

Commitment ($ billions)Rest of FY27FY28FY29FY30FY31FY32+Total
Supply and capacity69203921088390
Cloud service agreements27764228
Data center leases not commenced—12446374
Equity investments51122——20
Capital expenditures6—————6
Total82222103221673518

A few features of the table are worth spelling out. Supply and capacity obligations are heavily front-loaded: $272 billion, or roughly 70% of the $390 billion, falls in the remainder of fiscal 2027 and fiscal 2028, by the blog’s arithmetic. That is consistent with the company’s description of locking in components “to meet demand for the next several years.” The data center lease line is the opposite: $63 billion of the $74 billion falls in fiscal 2032 and later, reflecting long leases that have not yet started.

The company also says it “may assign some of these leases to third parties in the future.” That sentence matters, because an assignment could move some of the long-dated obligations off Nvidia’s books, but the commentary gives no sense of how likely or how large such assignments might be.

How big is $518 billion?

Raw figures of this size are hard to read without a yardstick. The comparisons below use only numbers Nvidia published in the same filing. They are the blog’s arithmetic and are meant for scale, not as financial ratios with any accepted meaning.

ComparisonFigureRatio
Total commitments vs revenue for the first nine months of fiscal 2027 ($293.7B)$518BAbout 1.8x
Fiscal 2028 commitments vs fourth-quarter revenue guidance annualized (4 × $130.0B)$222BAbout 0.43x
Supply commitments vs inventory on the balance sheet ($39.7B)$390BAbout 9.8x
Guarantee exposure vs shareholders’ equity ($266.7B)$118.0BAbout 0.44x
Guarantee exposure vs cash, equivalents and marketable debt securities ($53.5B)$118.0BAbout 2.2x

The annualized guidance comparison is crude because the fourth quarter includes an extra week and because annualizing a single quarter assumes nothing changes. It is included only to show that fiscal 2028 commitments are large but not out of proportion to the revenue run rate the company is guiding to. The guarantee comparisons describe maximum exposure, which the company says builds up over time and would only be called on if partners defaulted.

Neocloud agreements: contingent, but growing

Beyond the main table, Nvidia lists “Additional Commitments” under agreements with neocloud partners, a line it says was formerly referred to as “AI cloud agreements.” The total is $32 billion, spread as $6 billion in fiscal 2028, $7 billion in fiscal 2029, $6 billion in fiscal 2030, $5 billion in fiscal 2031 and $8 billion in fiscal 2032 and thereafter.

The commentary sets out the conditions clearly. The commitments “take effect only if the neoclouds secure financing to purchase the data center infrastructure products and the related compute capacity is deployed, made operational, and accepted by us.” If those conditions are not met in time, “certain commitments could be reduced or eliminated.” Under the arrangements, Nvidia earns revenue on the upfront sale of infrastructure and, if certain criteria are met, participates in a revenue share from the neoclouds’ third-party customers.

The structure is circular in a way that readers should understand without overstating. A neocloud buys Nvidia hardware, and Nvidia commits to buy back compute capacity from that neocloud once it is running. The company frames this as enabling “broader access to our AI infrastructure.” It also means a portion of future revenue is supported by Nvidia’s own purchase commitments. The documents do not say how much of third-quarter revenue came from customers that are party to these agreements.

The commentary also explains the reasoning for this support in unusually direct language: “Neoclouds and frontier labs are seeing extraordinary demand for AI infrastructure, yet many are growing faster than their balance sheets and long-term credit profiles can support.” That is the company’s own description of its customers’ financing constraints.

The guarantees

The guarantee disclosure is the most novel part of the filing. Nvidia says it has “land, power, shell, and hardware financing arrangements, under which we may be required to make payments to third parties if certain of our partners default on their obligations.”

Guarantee categoryMaximum gross exposure at October 25, 2026
Land, power and shell guarantees for neoclouds$3.5 billion
SB Energy Corp. credit support guarantees$105.0 billion
Power and systems credit support guarantees$9.5 billion
Total$118.0 billion

The SB Energy item dominates. According to the commentary, Nvidia entered into guarantees in August 2026 “to provide credit support on the land, power, and shell buildout to secure approximately 4.25 gigawatts (GW) across nine buildings at SB Energy’s PORTS-Pike Technology Campus in Ohio, which will exclusively host NVIDIA infrastructure under 20-year leases to OpenAI, subject to limited exceptions.”

The timing of the exposure is spelled out. Each guarantee becomes effective only when its related building is completed and ready for service, the first is expected in fiscal 2029, and exposure “builds over time as each building comes online, up to a maximum of $105 billion, and then declines over each 20-year lease term as OpenAI makes lease payments.” Nvidia also has an option to provide credit support in phases for approximately 3.8 additional gigawatts as the site scales.

Three points follow directly from that language. The $105 billion is a ceiling, not a current liability. None of it is effective until at least fiscal 2029. And the exposure ultimately depends on a single lessee’s ability to make lease payments over two decades. The commentary does not describe collateral, recourse or how the guarantees would rank against other creditors; those details may appear in the 10-Q.

Why the balance sheet barely shows it

Unconditional purchase obligations, leases that have not commenced and guarantees that are not yet effective generally do not appear as liabilities on a balance sheet. That is why Nvidia’s total liabilities of $105.5 billion at October 25 look modest next to $518 billion of commitments and $118 billion of guarantee exposure. The commitments are disclosed precisely because they are not otherwise visible.

That does not make them hidden in any improper sense; disclosure is the mechanism by which investors are meant to see them. It does mean that headline leverage measures based on reported debt, which stood at $32.4 billion of long-term borrowings, understate the scale of the company’s forward obligations. Readers who track the broader discussion of off-balance-sheet exposures linked to AI infrastructure will recognize the pattern.

The supplier announcements

Nvidia’s separate November 17 release, “NVIDIA Expands Commitments With Supply Chain Partners to Support AI Infrastructure Buildout,” explains part of the increase in supply obligations. The agreements described are:

PartnerWhat was announcedTiming stated
Kinsus Interconnect TechnologyNew multiyear agreement to expand advanced ABF package substrate capacity dedicated to Nvidia by more than 3.5x from its Youth campus in Taoyuan, TaiwanProduction expected to begin in the second half of 2027; additional capacity targeted for the third quarter of 2028
Samsung Electro-MechanicsAmended substrate supply agreement to improve capabilities at existing factories in Sejong and Busan, South KoreaNot stated
Amkor TechnologyMultiyear agreements announced in July and October 2026 totaling nearly $3 billion, including prepayments, supporting packaging expansion in Arizona and South KoreaNot stated
SPILMultiyear agreement supported by multibillion-dollar funding from Nvidia to expand advanced packaging in TaiwanContracted capacity increase by mid-2029
TSMCVera Rubin NVL72 and Vera CPUs use TSMC’s 5nm, 3nm and CoWoS packagingVolume shipments began in the third quarter
FoxconnDeepened collaboration on next-generation GPU and networking platformsNot stated

The release also recalls October commitments that included “an aggregate 2.5x increase in substrate capacity over three years,” a doubling of semiconductor testing capacity from 2026 to 2027 and a 15x increase in SPIL’s optical-engine capacity by mid-2027. Debora Shoquist, Nvidia’s executive vice president of operations, called the supply chain “a strategic advantage.”

Prepayments and supplier funding are a form of commitment that reaches beyond purchase orders. When a buyer finances a supplier’s capacity, it shares the risk that demand falls short of the capacity being built. The release does not quantify the SPIL funding beyond “multibillion-dollar.”

Memory: the single largest driver

The commentary attributes the jump in supply commitments “primarily” to multi-year memory procurement. The same documents note, in the Edge Computing discussion, that “higher memory and system prices continue to affect consumer PC demand.” Read together, the two statements suggest Nvidia is locking in memory supply in a tight market. The company does not disclose prices, suppliers or volumes for its memory contracts, so the economics cannot be assessed from the filing.

The export-control committee

The board’s new National Security and Export Control Oversight Committee is a standing committee “responsible for board-level oversight of the company’s national security policies, export control compliance and global trade risk management.” It is chaired by director Suzanne Nora Johnson, with directors Stephen C. Neal and John O. Dabiri as members, and it will meet at least quarterly, including in executive session. The company’s National Security Team and Global Trade Compliance Team report to the committee and, for day-to-day operations, to interim CEO Colette Kress. The committee will retain independent outside experts in law, national security and trade compliance.

Board chairman Harvey Jones said in the announcement that “advanced silicon technologies and the emergence of artificial and superintelligence present unprecedented challenges for national security,” and that the company is committed “to working constructively with government and industry partners.” The release does not cite a specific regulatory action or investigation as the reason for the committee, and this article does not speculate about one.

The commercial relevance is visible in the earnings documents. Hopper shipments to China were $2.4 billion in the third quarter, and the fourth-quarter outlook assumes no China Data Center compute revenue at all. Governance over export compliance therefore sits directly on a revenue line.

Equity stakes and the “infrastructure financiers”

The commitments table includes $20 billion of future equity investments, and the commentary says these “are focused on frontier labs, infrastructure financiers, and other private companies, subject to certain contingencies.” The balance sheet shows non-marketable securities of $65.5 billion at quarter end, up from $22.3 billion in January, and the cash flow statement shows $58.1 billion of equity security purchases in the first nine months of the fiscal year.

The cash flow statement also contains a financing line labelled “Groq, Inc.” with an outflow of $2,944 million in the first nine months. The commentary does not explain it, and this article does not attempt to characterize the transaction.

What changed in a single quarter

The commentary gives prior-quarter figures for two of the lines, which allows a simple comparison of how quickly the obligations are growing relative to the business.

LineEnd of Q2 FY27End of Q3 FY27Change
Supply commitments$279B$390B+$111B (about +40%)
Cloud service agreements plus data center leases$74B$102B+$28B (about +38%)
Quarterly revenue, for reference$96.2B$115.8B+20%

The percentages are the blog’s arithmetic. Both commitment lines grew roughly twice as fast as quarterly revenue. That is not by itself a warning sign, since commitments are signed ahead of the revenue they are meant to support, but it does mean the company’s forward obligations are expanding faster than its current sales. If demand slowed, the gap between what Nvidia has agreed to buy and what it can sell would be the first place pressure would show, most likely through inventory and purchase-obligation charges. Nvidia has not reported any such charge for the quarter, and the documents give no indication that one is expected.

A short primer on reading purchase obligations

For readers less familiar with these disclosures, a purchase obligation is a contract to buy goods or services in the future on specified terms. Some are firm and non-cancellable; others can be cancelled with penalties. Companies generally disclose the total in their filings because the amounts can be large and are not recorded as liabilities until delivery. When demand falls, companies sometimes record charges for excess purchase commitments, which is why analysts watch the ratio of commitments to expected sales. The CFO commentary does not break down how much of the $390 billion is cancellable, which is one of the details a fuller filing could clarify.

What it means for crypto and tokenized-asset readers

There is no direct crypto exposure in these disclosures. The relevance is indirect and runs through three channels. First, AI infrastructure has become a competing use for power and data center capacity that bitcoin miners also rely on; a Bernstein note reported by The Block on November 17, for example, valued miners Riot and TeraWulf partly on implied AI data center values, and long-term commitments by the largest chip buyer shape that market. Second, equity markets have treated Nvidia as a bellwether for AI spending, and swings in that sentiment have spilled into crypto prices at times in 2026. Third, the disclosures are a reminder of how much modern balance-sheet risk sits in commitments and contingent guarantees rather than in reported debt, a theme that applies equally to crypto firms, stablecoin issuers and tokenization platforms.

Open questions the 10-Q may answer

  • How the SB Energy guarantees are structured, including any collateral or recourse.
  • Whether any third-quarter revenue came from customers party to neocloud agreements or guarantees, and how much.
  • How much of the data center lease portfolio Nvidia expects to assign to third parties.
  • The size of supplier prepayments and funding commitments, including SPIL.
  • Customer concentration figures, which Nvidia typically discloses in its quarterly filings.

Frequently asked questions

How much has Nvidia committed to spend?

The CFO commentary lists $518 billion of future commitments as of October 25, 2026, including $390 billion of supply and capacity obligations, $28 billion of cloud service agreements, $74 billion of data center leases not yet commenced, $20 billion of equity investments and $6 billion of capital expenditures.

What is the $105 billion guarantee?

It is the maximum exposure under credit support Nvidia agreed to provide for the land, power and shell buildout of about 4.25 gigawatts at SB Energy’s PORTS-Pike Technology Campus in Ohio, which will host Nvidia infrastructure under 20-year leases to OpenAI. The guarantees become effective building by building, starting in fiscal 2029.

Has Nvidia paid anything under its guarantees?

The company says it has not made any payments related to its guarantees to date.

What are neocloud agreements?

Contingent arrangements totaling $32 billion under which Nvidia commits to purchase compute capacity from neocloud partners once those partners finance, deploy and Nvidia accepts the related infrastructure.

Why did supply commitments rise so much?

Nvidia attributes the increase from $279 billion to $390 billion primarily to multi-year memory procurement.

Sources and research limits

Primary sources: Nvidia’s CFO Commentary on Third Quarter Fiscal 2027 Results (Exhibit 99.2 to the November 17 Form 8-K) for commitments, neocloud agreements, guarantees and balance sheet commentary; the third-quarter press release for the condensed financial statements; “NVIDIA Expands Commitments With Supply Chain Partners to Support AI Infrastructure Buildout”; and “NVIDIA Board of Directors Establishes National Security and Export Control Oversight Committee.” Related reading on this blog: AI capex in 2026 across the largest buyers and CoreWeave’s third-quarter results.

Limits: the CFO commentary is unaudited. The Form 10-Q, which may provide further detail on these obligations, is not included in this review. Ratios in this article are the blog’s arithmetic for scale only. The article does not evaluate the creditworthiness of any counterparty, and it does not infer the purpose of transactions the company has not explained. This article is for information only and is not investment advice.

Disclaimer: This content was generated with the assistance of artificial intelligence (AI) and has been reviewed by our editorial team. It is intended for informational purposes only and should not be construed as financial, investment, or legal advice. Cryptocurrency investments involve significant risk.
KCEX BLOGKCEX BLOG
Previous 9 hours ago
Next 9 hours ago

Related Posts

SHARE
TOP