(26) Vendor Financing Pulls Demand Forward

When the Seller Funds the Sale

NVIDIA's July 2026 quarter shows why an ordinary revenue number is no longer enough to describe AI infrastructure demand. The company disclosed $36 billion of cloud-service commitments to AI cloud providers that procure NVIDIA systems, alongside $29 billion of remaining service obligations and $25 billion of commitments to make equity investments. In August, it also entered guarantees capped at $105 billion for leases supporting an OpenAI campus being developed by SB Energy. The chip sale, the customer's financing, the supplier's investment and the future purchase of compute now sit in the same commercial system.

Part of the system is converting to cash more slowly. NVIDIA's accounts receivable reached $63.1 billion, while days sales outstanding rose from 45 to 60 because some large investment-grade customers received extended payment terms. The filing says the cloud providers can sell capacity to third parties and that NVIDIA's purchase commitments decline as they do. The guarantees also decline as OpenAI pays its leases.

They also reveal the question that headline orders cannot answer. A purchase measures how much infrastructure a customer has contracted to acquire. Durable demand requires unaffiliated users to generate enough cash for that customer to pay the supplier, service the financing and renew the capacity without another round of support. Vendor finance compresses the time between a forecast of that demand and the purchase made in anticipation of it. It brings the sale forward; the end user still has to arrive.

The Same Dollar Can Support Several Contracts

Traditional vendor financing is a loan or an extended receivable used to buy the lender's product. The AI version is broader. A chip supplier can invest in a model developer, guarantee a datacentre lease, commit to buy unused cloud capacity or help arrange third-party debt. A hyperscaler can invest in a model developer that pays for its cloud services. A model developer can sign a long-term capacity agreement with an infrastructure company whose equipment is supplied or supported by one of the developer's investors.

The transactions remain legally distinct, but their economic dependence can overlap. Microsoft reported $24.1 billion of revenue from commercial arrangements with OpenAI in its 2026 financial year, while funding $11.9 billion of its $13 billion investment commitment and carrying $6 billion of related receivables. NVIDIA has stated an intention to invest as much as $100 billion in OpenAI as successive gigawatts are deployed. AMD granted OpenAI warrants for as many as 160 million shares, with vesting linked to purchases scaling from one to six gigawatts and to AMD's share price.

Accounting standards already recognise that payment terms matter. IFRS 15 and the FASB's guidance require a significant financing component to be separated from the cash selling price in relevant contracts. Revenue recognition still answers a narrower question than market analysis: whether control transferred and consideration is expected under the applicable rules. An analyst must ask where the cash originated, which party retains the downside and whether the customer could have made the purchase on similar terms without the supplier's participation.

An investment can become the recipient's cash, pay a cloud bill and fund the cloud provider's chip purchase. Each entry can be accurate. The chain nevertheless contains less independent demand than separate transactions among unaffiliated, self-financing businesses would imply.

Suppliers Sometimes Make the Best Lenders

The strongest case for vendor finance begins with information, not promotion. Research on trade credit finds that suppliers may understand a customer's inventory, order flow and resale value better than a bank. They can recover and redeploy specialised equipment, earn product margin as well as financing income, and treat credit as an implicit investment in a distribution channel. A supplier may therefore fund a viable buyer that conventional lenders reject because the asset class is new or the borrower's operating history is short.

That case fits parts of the AI market. A lender assessing a young GPU cloud must estimate hardware resale values, power availability, software compatibility, utilisation and the pace of chip obsolescence. NVIDIA has more information about several of those variables than a generalist bank. Its arrangement with CoreWeave has an initial value of $6.3 billion and covers residual capacity that CoreWeave cannot sell through 2032. This support can let CoreWeave build before every rack has an outside tenant, while placing some demand risk with the company most able to assess and remarket the capacity.

Supplier support can also solve a coordination problem. Users will not commit to applications without available compute, while lenders will not finance compute without committed users. Capacity built ahead of demand can lower prices, shorten queues and make new applications possible. The resulting market may be valuable even though it required sponsorship at inception.

The contractual exit matters more than the initial circularity. NVIDIA's service commitments reduce when partner clouds find third-party customers, and its new financing memoranda with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR contemplate more than $500 billion of independently underwritten capital. Oracle offers another useful variation: much of the increase in its remaining performance obligations came from large AI contracts in which customers prepaid for GPUs or supplied them directly. In both cases, someone outside the infrastructure provider is assuming more of the funding risk. That is stronger evidence than a larger order backed by a larger supplier guarantee, although it still leaves the credit and funding source of the end customer to be assessed.

Telecom Failed the Substitution Test

Telecom equipment vendors used customer finance for a similar reason in the late 1990s. New carriers needed large networks before they could acquire subscribers, and equipment suppliers understood the assets better than many lenders. Traffic growth was real, fibre was useful and vendor credit helped challengers enter markets dominated by incumbents. The financing mechanism served a rational commercial purpose.

It became dangerous when projected subscriber cash failed to replace supplier and capital-market support before the financing matured. Nortel's reported customer-finance exposure fell from $5.17 billion in 2000 to $2.08 billion in 2001 as it curtailed new commitments, increased provisions and faced customer defaults and bankruptcies. Its ability to place those loans with banks and capital markets weakened at the moment its customers needed refinancing most. The supplier had pulled equipment demand into a period when the ultimate buyers could not yet fund it.

Lucent adds a separate accounting warning. The SEC later alleged that undisclosed side agreements, credits and other practices had caused the company to recognise more than $1.1 billion of revenue improperly. That enforcement history should not be projected onto every financed AI sale. It does show why contract form is insufficient when the seller has preserved risk or altered payment obligations outside the reported terms.

The telecom lesson is about substitution. A financed network becomes sound when subscribers pay enough to support the carrier and outside creditors can refinance it without relying on the equipment vendor. Demand remains conditional when the vendor must keep extending credit, absorbing inventory or arranging the next buyer. Useful infrastructure can be overbuilt when its financing schedule assumes adoption will arrive sooner than it does.

AI Has Real Users and Financed Future Users

Current evidence supports neither a closed-loop caricature nor an assumption that every announced gigawatt has an independent buyer. The US Census Bureau found business AI use fluctuating between 17% and 20% from December 2025 to May 2026, rising to 37% among firms with at least 250 employees. Eurostat measured use by 20% of EU enterprises with at least ten employees in 2025. Stanford's survey-based compilation reports much higher organisational adoption, but its broader definitions also show agent use in individual business functions remaining in single digits. Measurement differs; incomplete diffusion does not.

Company disclosures show substantial activity within that incomplete adoption. Microsoft reported more than 30 million paid Microsoft 365 Copilot seats and Azure annual revenue above $100 billion. Alphabet said 70% of existing Google Cloud customers were using its AI products, 150 customers were each processing roughly one trillion tokens, and revenue from generative-model products had more than tripled. Amazon said Bedrock was used by more than 100,000 companies, while Trainium2 capacity was fully subscribed. These are credible signs of paying, distributed usage, even though the companies do not disclose a complete standalone profit-and-loss account for AI.

Physical consumption offers another check. The International Energy Agency estimates that datacentre electricity use grew 17% in 2025 and that consumption at AI-focused facilities rose about 50%. CBRE recorded 420 megawatts of European signings in the first half of 2026, up from 89 megawatts a year earlier, with providers increasingly requiring deposits or letters of credit from prospective tenants. Power draw, signed capacity and customer security all point to activity beyond accounting entries.

Scale remains the unresolved issue. OpenAI and NVIDIA described a path to at least ten gigawatts, Stargate announced an intention to invest $500 billion over four years, and hyperscalers continue to raise capital expenditure. The Bank of England, BIS, IMF and OECD all describe a funding shift toward bonds, private credit, leases, joint ventures and guarantees as the build-out outruns internal cash generation. AI debt is moving into private markets, where disclosure can become thinner even as structural dependence grows. Existing usage proves a market; it does not prove that its cash flow can absorb every financed project on the announced timetable.

Measure the Exit From Support

The useful metric is a substitution ratio: unaffiliated customer cash generated by the financed capacity divided by the vendor-linked capital, guarantees and purchase obligations supporting it. The exact calculation will vary by company, and public disclosure is not yet sufficient to produce one clean number. Its direction can still be observed through several accounts.

Supplier exposure per deployed gigawatt should decline as a market matures. Equity investments, lease guarantees, residual-capacity purchases and extended receivables belong in the numerator of support, even when they sit in separate legal entities. A shift from a direct vendor guarantee to a special-purpose vehicle funded by private credit does not complete the substitution if the vendor still protects residual value or commits to buy unused output. The BIS calls some of these lease, guarantee and joint-venture structures shadow borrowing because they move financing away from conventional debt without removing the economic obligation.

Cash collection provides a second test. Rising revenue accompanied by lengthening payment terms, faster receivables growth or repeated contract modifications suggests that delivery is outrunning the customer's cash. NVIDIA's move from 45 to 60 days sales outstanding is not decisive on its own: large quarterly transactions and investment-grade terms can change the measure. It is a reason to follow whether DSO normalises as capacity reaches outside users. Cash paid by a customer that was recently funded by the supplier also deserves separate treatment from cash earned from unaffiliated operations.

Utilisation and renewal supply the operational test. A long-term capacity contract establishes credit exposure, not productive use. Lenders therefore underwrite tenant concentration, actual utilisation, re-leasing prospects, power costs and hardware residual value. The decisive evidence arrives when supported capacity is sold to unrelated customers, then renewed or re-leased at economic prices without another guarantee. S&P and industry credit research identify overcapacity, tenant renewal and refinancing as central risks for exactly this reason.

The final test occurs at refinancing. As datacentre projects move from construction loans into bonds, capital recycling can make development repeatable, but it cannot establish the quality of the tenant's revenue. A refinancing funded by creditors relying on unaffiliated cash flow is evidence of substitution. One that requires a larger sponsor backstop extends the forecast. The maturity of AI adoption therefore matters as much as the maturity of the debt; management systems determine whether experiments become recurring enterprise use.

A Market Must Outgrow Its Sponsor

Boards and investors should separate three questions that large announcements tend to combine. The first is whether the infrastructure will be built. The second is whether a contracted counterparty will pay. The third is whether that counterparty earns cash from customers outside the financing circle. Guarantees can answer the first two while leaving the third unresolved.

This framing avoids two mistakes. Treating every reciprocal arrangement as fabricated ignores real customers, scarce powered sites and the information advantage of specialised suppliers. Treating every contract as independent demand ignores who supplied the customer's capital and who retains the unused-capacity risk. The market can contain genuine adoption and excess financed capacity at the same time.

The best future disclosure from an AI supplier may therefore look superficially less ambitious: fewer dollars of guarantees and capacity purchases for each dollar of third-party cloud revenue, shorter collection periods, more unaffiliated tenants and renewals completed without support. Supplier exposure can still grow in absolute terms during a rapid expansion. It should grow more slowly than outside cash if the financing is doing its job.

Vendor finance gives a young market time to become self-supporting. In quarterly disclosures, success should appear as falling guarantees per unit of capacity, faster collections, more outside tenant revenue and renewals completed without supplier support. Those figures will show whether today's financing accelerated adoption or shifted revenue into an earlier period.

Sources

Company Filings and Disclosures

1. NVIDIA Corporation, "Quarterly Report on Form 10-Q for the Quarter Ended July 26, 2026" https://www.sec.gov/Archives/edgar/data/1045810/000104581026000075/nvda-20260726.htm

2. NVIDIA Corporation, "CFO Commentary on Second Quarter Fiscal 2027 Results" https://www.sec.gov/Archives/edgar/data/1045810/000104581026000073/q2fy27cfocommentary.htm

3. NVIDIA Corporation, "NVIDIA Partners With Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to Establish AI Compute Infrastructure Financing Platforms" https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Partners-With-Apollo-BlackRock-Blackstone-Brookfield-Goldman-Sachs-and-KKR-to-Establish-AI-Compute-Infrastructure-Financing-Platforms-to-Mobilize-Over-500-Billion-of-Third-Party-Capital/default.aspx

4. CoreWeave, Inc., "Current Report on Form 8-K, September 9, 2025" https://www.sec.gov/Archives/edgar/data/1769628/000176962825000047/crwv-20250909.htm

5. CoreWeave, Inc., "Quarterly Report on Form 10-Q for the Quarter Ended June 30, 2026" https://www.sec.gov/Archives/edgar/data/1769628/000176962826000366/crwv-20260630.htm

6. CoreWeave, Inc., "CoreWeave and NVIDIA Expand Collaboration to Accelerate Buildout of AI Factories" https://www.sec.gov/Archives/edgar/data/1769628/000176962826000044/ex991pressrelease_final.htm

7. OpenAI, "OpenAI and NVIDIA Announce Strategic Partnership to Deploy 10 Gigawatts of NVIDIA Systems" https://openai.com/index/openai-nvidia-systems-partnership/

8. OpenAI, "Scaling AI for Everyone" https://openai.com/index/scaling-ai-for-everyone/

9. OpenAI, "Announcing the Stargate Project" https://openai.com/index/announcing-the-stargate-project/

10. Advanced Micro Devices, Inc., "Current Report on Form 8-K, October 6, 2025" https://ir.amd.com/financial-information/sec-filings/content/0001193125-25-230895/d28189d8k.htm

11. Amazon, "Amazon and Anthropic Deepen Their Shared Commitment to Advancing Generative AI" https://www.aboutamazon.com/news/company-news/amazon-aws-anthropic-ai

12. Amazon.com, Inc., "Quarterly Report on Form 10-Q for the Quarter Ended June 30, 2026" https://www.sec.gov/Archives/edgar/data/1018724/000101872426000026/amzn-20260630.htm

13. Amazon.com, Inc., "2025 Annual Report" https://www.sec.gov/Archives/edgar/data/1018724/000101872426000004/amzn-20251231.htm

14. Amazon.com, Inc., "Fourth Quarter 2025 Financial Results" https://www.sec.gov/Archives/edgar/data/1018724/000101872426000002/amzn-20251231xex991.htm

15. Microsoft Corporation, "2026 Annual Report" https://www.sec.gov/Archives/edgar/data/789019/000119312526323660/msft-20260630.htm

16. Microsoft Corporation, "Fourth Quarter and Fiscal Year 2026 Results" https://www.sec.gov/Archives/edgar/data/789019/000119312526323632/msft-ex99_1.htm

17. Microsoft, "The Next Phase of the Microsoft-OpenAI Partnership" https://blogs.microsoft.com/blog/2026/04/27/the-next-phase-of-the-microsoft-openai-partnership/

18. Oracle Corporation, "2026 Annual Report" https://www.sec.gov/Archives/edgar/data/1341439/000119312526277521/orcl-20260531.htm

19. Oracle Corporation, "Fourth Quarter and Fiscal Year 2026 Results" https://www.sec.gov/Archives/edgar/data/1341439/000119312526265848/orcl-ex99_1.htm

20. Meta Platforms, Inc., "2025 Annual Report" https://www.sec.gov/Archives/edgar/data/1326801/000162828026025534/meta-12312025x10kars.htm

21. Alphabet Inc., "Quarterly Report on Form 10-Q for the Quarter Ended March 31, 2025" https://www.sec.gov/Archives/edgar/data/1652044/000165204425000043/goog-20250331.htm

22. Alphabet Inc., "2025 Fourth Quarter Earnings Call" https://abc.xyz/investor/events/event-details/2026/2025-Q4-Earnings-Call-2026-Dr_C033hS6/default.aspx

23. Alphabet Inc., "2025 Third Quarter Earnings Call" https://abc.xyz/investor/events/event-details/2025/2025-Q3-Earnings-Call-2025-4OI4Bac_Q9/default.aspx

Public Institutions and Adoption Data

24. Bank of England, "Financial Stability Report - July 2026" https://www.bankofengland.co.uk/-/media/boe/files/financial-stability-report/2026/financial-stability-report-july-2026.pdf

25. Bank of England, "Financial Stability Report - December 2025" https://www.bankofengland.co.uk/-/media/boe/files/financial-stability-report/2025/financial-stability-report-december-2025.pdf

26. Bank for International Settlements, "Financing the AI Boom: From Cash Flows to Debt" https://www.bis.org/publications/bulletin-120-financing-ai-boom-cash-flows-debt

27. Bank for International Settlements, "Financing the AI Infrastructure Boom: On- and Off-Balance Sheet Borrowing" https://www.bis.org/publications/financing-ai-infrastructure-boom-on-and-off-balance-sheet-borrowing

28. International Monetary Fund, "Global Financial Stability Report, April 2026" https://www.imf.org/-/media/files/publications/gfsr/2026/april/english/text.pdf

29. OECD, "Corporate Debt Market Outlook in a Transforming World" https://www.oecd.org/en/publications/global-debt-report-2026_e9d80efd-en/full-report/corporate-debt-market-outlook-in-a-transforming-world_cf86a220.html

30. International Energy Agency, "Key Questions on Energy and AI: Executive Summary" https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary

31. International Energy Agency, "Energy and AI" https://www.iea.org/reports/energy-and-ai/

32. United States Census Bureau, "Large Firms With at Least 20 Employees Biggest AI Users" https://www.census.gov/library/stories/2026/05/ai-use-businesses.html

33. United States Census Bureau, "The Microstructure of AI Diffusion: Evidence from Firms, Business Functions, and Worker Tasks" https://www.census.gov/library/working-papers/2026/adrm/CES-WP-26-25.html

34. Eurostat, "20% of EU Enterprises Use AI Technologies" https://ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20251211-2

35. Stanford Institute for Human-Centered Artificial Intelligence, "The 2026 AI Index Report: Economy" https://hai.stanford.edu/ai-index/2026-ai-index-report/economy

Accounting, Trade Credit and the Telecom Precedent

36. Financial Accounting Standards Board, "Revenue Recognition Implementation Q&As" https://storage.fasb.org/Rev_Rec_Implementation_QAs.pdf

37. IFRS Foundation, "IFRS 15 Revenue from Contracts with Customers" https://www.ifrs.org/issued-standards/list-of-standards/ifrs-15-revenue-from-contracts-with-customers/

38. United States Securities and Exchange Commission, "Staff Accounting Bulletin No. 101: Revenue Recognition in Financial Statements" https://www.sec.gov/interps/account/sab101.htm

39. United States Securities and Exchange Commission, "Commission Statement About Management's Discussion and Analysis of Financial Condition and Results of Operations" https://www.sec.gov/rules-regulations/2002/01/commission-statement-about-managements-discussion-analysis-financial-condition-results-operations

40. United States Securities and Exchange Commission, "SEC Settles Enforcement Action Against Lucent Technologies" https://www.sec.gov/news/press/2004-67.htm

41. United States Securities and Exchange Commission, "SEC Settles Financial Fraud Charges Against Nortel Networks" https://www.sec.gov/enforcement-litigation/litigation-releases/lr-20333

42. Nortel Networks Corporation, "2001 Annual Report" https://www.sec.gov/Archives/edgar/data/72911/000113031902000168/t06646e10-k.htm

43. Mitchell A. Petersen and Raghuram G. Rajan, "Trade Credit: Theories and Evidence" https://www.nber.org/papers/w5602

44. Bruce J. Sherrick and Robert W. Lubben, "Economic Motivations for Vendor Financing: Theory and Evidence" https://ageconsearch.umn.edu/record/131334

Credit and Infrastructure Research

45. S&P Global Ratings, "AI Infrastructure Investment to Exceed $1.3 Trillion by 2027" https://press.spglobal.com/2026-08-27-AI-Infrastructure-Investment-To-Exceed-1-3-Trillion-By-2027,-S-P-Global-Ratings-Says

46. KBRA, "Data Centers: Developments and Trends in Project Finance" https://www.kbra.com/publications/wrRRvRmM/kbra-releases-research-data-centers-developments-and-trends-in-project-finance?format=web

47. JLL, "2026 Global Data Center Outlook" https://www.jll.com/content/dam/jllcom/en/global/documents/reports/research-reports/26-research-global-data-center-outlook-new.pdf

48. CBRE, "AI Demand Drives Record Data Centre Signings Across Europe" https://www.cbre.co.uk/press-releases/ai-demand-drives-record-data-centre-signings-across-europe

49. Western Asset, "Financing the AI Infrastructure Boom" https://www.westernasset.com/us/en/research/whitepapers/financing-the-ai-infrastructure-boom.cfm

50. CRE Finance Council, "Data Center E-Primer" https://www.crefc.org/common/Uploaded%20files/Learn/DataCenters-Eprimer_Final.pdf