Off-balance-sheet commitments for AI infrastructure are rapidly expanding, and true leverage pressure is rising

Overview

The report analyzes off-balance-sheet commitments that have expanded rapidly in AI infrastructure buildout. Hyperscale cloud providers and Nvidia signed large amounts of long-term procurement, leasing, take-or-pay, and guarantee agreements to secure GPUs, memory, cloud computing capacity, colocation/data center shell leases, power, and other critical resources. These agreements provide financing visibility for suppliers and developers, but many obligations enter the balance sheet only when delivery, lease commencement, or payment becomes probable, which may cause investors to underestimate true economic leverage.

Core views

The core view is that AI capital spending is not fully reflected in traditional balance-sheet liabilities; increasingly, more risk is shifted to off-balance-sheet contractual commitments. The rapid growth of procurement and future lease commitments both supports supply-chain expansion and reduces flexibility for customers if AI demand underperforms. The report particularly notes that disclosures from META, GOOGL, AMZN, MSFT, ORCL, and NVDA should be analyzed together across procurement obligations, lease footnotes, derivatives, and guarantee footnotes.

Analysis framework

The report applies a combination of accounting standards, contract structure, and credit analysis to unpack off-balance-sheet arrangements in the AI ecosystem by category, including unconditional procurement obligations, inventory purchase commitments, compute capacity arrangements, take-or-pay contracts, not-yet-started leases, variable lease payments, renewal options, residual value guarantees, third-party lease guarantees, power purchase agreements, and contract-based chip financing structures.

Methodology notes

• Accounting standardsUS GAAP lease identification

  Identifying assets and control of use rights

  A compute capacity contract is more likely to be treated as a lease and brought onto the balance sheet only after lease commencement when a specific GPU or rack is designated and the customer obtains substantially all economic benefits and controls how it is used.

• Disclosure analysisUnconditional purchase obligation disclosure

  Firm or minimum purchase commitments

  SEC Regulation S-K and US GAAP require disclosure of certain material cash needs and unconditional purchase obligations, but these long-term commitments are typically not recognized as liabilities on the balance sheet prior to delivery.

• Credit analysisRating agency debt adjustments

  Contingent guarantees and lease endorsements

  When lease endorsements or guarantees are a central support for transaction-counterparty financing, rating agencies may treat them as potential liquidity calls and adjust leverage metrics when they become more specific or effective.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• META PLATFORMS INC

  AI infrastructure demand-side participant and lease/cloud capacity commitment issuer

  Strengths

  Locks in AI compute resources through long-term data center, cloud capacity, and power-related arrangements, supporting AI product expansion.

  Weaknesses

  Commitment levels are relatively high versus projected future operating cash flow, with variable lease costs and not-yet-started leases potentially understating current balance-sheet pressure.

  Comparison

  The report states that META commitments are about 1.7 times future operating cash flow, and discloses that variable lease costs are influenced by CPI, energy, maintenance, and utility costs.

  Risks

  AI demand softening, lease liabilities coming onto the balance sheet after commencement, rising variable costs, residual guarantees, or other contingent obligations becoming more likely.

• Nvidia

  Core AI chip supply-chain company and supplier of procurement/cloud service commitments

  Strengths

  Locks in foundry, memory, and cloud service resources in advance, helping meet high compute demand and supporting supplier financing.

  Weaknesses

  Inventory and procurement commitments have risen to around 32% of projected revenue, increasing downside risk if demand weakens.

  Comparison

  Historically, inventory and procurement commitments were around 15%-20% of forward revenue, rising to about 32% in Jan 2026.

  Risks

  Weaker than expected AI demand, prepayments and inventory risk in the supply chain, and fixed-cost pressure from cloud service commitments.

• GOOGL / Alphabet

  Hyperscale cloud provider and third-party lease endorsement provider

  Strengths

  Investment-grade credit can lower data center developers' financing costs and support expansion of TPU and AI infrastructure.

  Weaknesses

  Approximately $1.7 billion of lease endorsement remains mostly off-balance-sheet and may later be included in debt adjustments by rating agencies.

  Comparison

  The report describes Alphabet as the main case of third-party lease endorsement, at a scale larger than guarantees disclosed by Oracle and NVIDIA.

  Risks

  Low project utilization, endorsement effectiveness after lease commencement, debt upgrades by rating agencies, and liquidity-call risk.

• ORCL

  Cloud capacity and AI infrastructure commitment provider

  Strengths

  Participates in AI compute expansion through cloud capacity and infrastructure commitments.

  Weaknesses

  Commitments are more than 7 times projected future operating cash flow, making financial flexibility more sensitive.

  Comparison

  In the report sample, ORCL has the highest commitment multiple relative to cash flow.

  Risks

  Contract rigidity, higher capital costs, and underutilization of commitments due to changes in AI demand.

• Data center developers and suppliers

  Financing beneficiaries of off-balance-sheet commitments

  Strengths

  Once they have long-term support from hyperscale cloud providers or Nvidia, they can more easily obtain bank loans, bonds, or private credit.

  Weaknesses

  Financing is dependent on core customer credit quality and contract performance, and project-level leverage can rise.

  Comparison

  The report explains the leverage transfer from customer-side to supplier-side through cash-flow channels among supplier balance sheets, banks, and private credit balance sheets.

  Risks

  Changes in customer demand, difficulty renegotiating contracts, project financing defaults, and cost overruns in data centers or power.

Key data

• Total AI-related commitments>$1.3 trillionIncludes approximately $640 billion in procurement commitments and around $675 billion in lease commitments.
• Procurement commitments>$640 billionCovers hyperscale cloud providers and Nvidia, primarily related to data centers and technology infrastructure.
• Future lease payment commitmentsabout $675 billionMany leases have been signed but not yet commenced, so they remain off-balance-sheet before lease commencement.
• META commitments versus cash flowabout 1.7x future operating cash flowThe report notes that META's commitments have increased materially versus cash flow.
• ORCL commitments versus cash flow>7x future operating cash flowIndicates that some companies' AI infrastructure-related contractual commitments are already far above near-term cash-flow scale.
• Nvidia inventory and procurement commitmentsabout 32% of FY27 consensus expected revenueThe historical range is about 15%-20%, indicating an increased level of supply pre-locking at Nvidia.
• Nvidia cloud services agreement commitments$2.7 billionFrom the latest disclosed cloud services agreement commitments.
• Oracle cloud capacity commitments$1.0 billionCloud capacity commitment disclosed by Oracle.
• Alphabet lease endorsementabout $1.7 billionUsed to support financing for Bitcoin miner-to-data-center conversion projects, and is a credit-enhancement structure emphasized in the report.
• AI data center power cost exampleabout $2.5 billionEstimated using a 200MW critical IT load, 15-year lease term, and a national average electricity price of around $80/MWh, excluding growth.

Impact & implications

For investors, traditional leverage metrics may be insufficient to measure the real financial risk during the AI build cycle. Long-term commitments can provide an advantage in financing suppliers and securing scarce compute for cloud providers, but they can also magnify operating leverage and cash-flow stress when demand weakens, technology roadmaps change, or capital costs rise. Credit analysis should incorporate off-balance-sheet procurement, leasing, guarantees, variable payments, and renewal options into an overall leverage assessment.

Risks

• Off-balance-sheet obligations can understate balance-sheet leverage, making it difficult for investors to assess true economic leverage.
• Long-term procurement, take-or-pay, and lease commitments reduce flexibility when AI demand slows.
• Not-yet-started leases may enter the balance sheet in a concentrated manner after delivery, increasing lease liabilities.
• Variable lease payments, power costs, maintenance, insurance, and taxes may become persistent operating costs that are not fully capitalized.
• Residual value guarantees and third-party lease endorsements depend on management judgment, and may trigger liability recognition if conditions change.
• Project financing reliant on contract support can be harder to renegotiate or exit, especially when contracts already support bank, bond, or private credit financing.
• The AI ecosystem has circular financing and mutually reinforcing demand, increasing opacity and complexity for credit analysis.

What to watch

• Quarterly changes in each company's procurement obligations, required liquidity needs, and lease footnotes.
• The schedule for not-yet-started leases, especially the pace at which they move onto the balance sheet during 2026 to 2031.
• Changes in the share of variable lease costs in total lease costs for META, GOOGL, and AMZN.
• Whether Nvidia's inventory and procurement commitments relative to forward revenue remain above the historical range.
• Whether disclosures of lease endorsements, guarantees, and contingencies from GOOGL, ORCL, NVIDIA, and peers are expanding.
• Whether rating agencies make debt adjustments for Alphabet or other hyperscale cloud providers' endorsements and guarantees.
• Whether AI product revenue, compute utilization, and data center power costs can cover long-term fixed commitments.