Off-balance-sheet commitments in AI infrastructure are becoming a new source of economic leverage

Overview

This report analyzes the rapid expansion of off-balance-sheet commitments in AI infrastructure buildout. Morgan Stanley notes that hyperscale cloud providers, Nvidia, and other high-credit-quality companies in the AI ecosystem are supporting financing for data centers, chips, storage, and power infrastructure through long-term procurement commitments, cloud compute capacity agreements, leases not yet commenced, third-party guarantees, power purchase agreements, and chip financing structures. These arrangements improve revenue visibility for suppliers and enable developers to finance through bank loans, public debt markets, and private credit; however, because many obligations are not yet recognized as liabilities in accounting, investors may underestimate the true economic leverage if they only look at the balance sheet.

Core views

The core view is that the financing foundation for AI buildout is expanding from the traditional balance sheet to long-term contractual commitments. Purchase obligations, lease commitments, take-or-pay contracts, PPAs, and guarantees can support suppliers building capacity in advance, but they also reduce flexibility for commitment providers when AI demand is below expectations. The report emphasizes that off-balance-sheet commitments are growing in size, tenor, and structural complexity, making it harder for investors to assess total potential leverage, cash flow stress, and credit risk.

Analysis framework

The report uses a combination of accounting rules, contract structure, and credit analysis: first identify the main off-balance-sheet arrangements in the AI ecosystem, then determine when these contracts are brought onto the balance sheet under US GAAP and ASC 842, and finally assess the implications for financing, leverage, rating-agency adjustments, and investor analysis. It focuses on comparing purchase obligations, compute capacity agreements, lease payments, variable lease costs, renewal rights, residual guarantees, third-party guarantees, PPAs, and chip financing structures.

Methodology notes

• Accounting StandardsUS GAAP purchase obligation disclosure

  Unconditional Purchase Obligations

  When commitments are non-cancelable, involve fixed or minimum amounts, have terms longer than one year, and are related to production-capacity financing, a company is required to disclose unconditional purchase obligations; these are typically not recorded on the balance sheet before goods or services are delivered.

• Accounting StandardsASC 842 lease accounting

  Lease Identification and Lease Liability Recognition

  A contract is a lease only when the customer controls the right to use a specified asset, receives substantially all economic benefits, and controls how the asset is used; after lease commencement, minimum lease payments are generally capitalized as a right-of-use asset and a lease liability.

• Contract Structuretake-or-pay contract analysis

  Minimum Purchase or Payment Obligation

  Take-or-pay contracts require the buyer to take a minimum volume or make minimum payments regardless of actual usage, improving supplier financing visibility while reducing flexibility for the customer to exit or renegotiate.

• Credit Analysisrating agency contingent guarantee treatment

  Credit Adjustment for Guarantees and Backstops

  When contingent guarantees are closely linked to counterparty financing, rating agencies may treat them as potential liquidity calls and reflect some guarantee exposure in liability-related adjustments.

Asset mapping & comparison

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

• META PLATFORMS INC (META.US)

  The report cites its AI-related contractual commitments, cloud capacity arrangements, and variable lease payment disclosures.

  Strengths

  Long-term commitments help ensure AI compute, data center, and power resources, supporting product expansion.

  Weaknesses

  Commitments are about ~1.7x forward operating cash flow, with variable lease costs making up about 25% of total lease costs; true economic leverage may be higher than what is reflected on the balance sheet.

  Comparison

  Compared with MSFT and AMZN, the report highlights that META has more pronounced growth in commitments and more prominent variable lease disclosures; compared with ORCL, META has a lower commitments-to-cash-flow multiple.

  Risks

  AI demand slowdown, underutilization of compute, rising lease and power costs, or residual-value guarantees and off-balance-sheet obligations moving onto the balance sheet.

• NVIDIA (NVDA)

  The report views Nvidia as a key participant in the AI supply chain, supporting upstream financing through procurement commitments and cloud services agreements.

  Strengths

  Procurement commitments help suppliers obtain financing and secure GPU, foundry, and memory supply.

  Weaknesses

  Inventory and procurement commitments account for about 32% of FY27 consensus revenue, above the historical 15%-20% range.

  Comparison

  Compared with cloud providers, Nvidia more clearly reflects procurement commitments and supply assurance on the chip supply chain side.

  Risks

  If compute demand cools, inventory and procurement obligations could amplify downside risk.

• Alphabet (GOOGL)

  The report focuses on its lease backstop, leases not yet commenced, and variable lease payments.

  Strengths

  Investment-grade credit can reduce data center project financing costs and may accelerate TPU ecosystem buildout.

  Weaknesses

  An estimated ~$17bn lease backstop is largely off-balance-sheet and may be adjusted as debt by rating agencies in the future.

  Comparison

  Compared with other cloud providers, Alphabet is more notable for third-party lease backstop cases.

  Risks

  Backstop trigger risk, low utilization, project default, or credit-rating debt adjustments.

• Microsoft (MSFT)

  The report cites its take-or-pay disclosures related to CRWV and cloud compute capacity arrangements.

  Strengths

  Long-term compute contracts help secure AI capacity and support supplier financing.

  Weaknesses

  The take-or-pay structure can reduce flexibility when demand changes.

  Comparison

  Together with CRWV, it forms a customer and financing support relationship; the contracts may support CRWV data center financing.

  Risks

  Minimum payments may still be required if demand is below expectations, and exiting or renegotiating contracts may be difficult.

• Oracle (ORCL)

  The report references its cloud capacity arrangements, power PPAs, and lease guarantee disclosures.

  Strengths

  Long-term capacity and power arrangements can support AI cloud infrastructure expansion.

  Weaknesses

  Commitments exceed 7x of future operating cash flow, indicating higher financial flexibility pressure.

  Comparison

  The commitments-to-cash-flow multiple is materially higher than META.

  Risks

  High commitment multiple, rising financing costs, and execution risk in power and data center projects.

• Amazon (AMZN)

  The report includes it in the comparison of hyperscaler procurement obligations, lease commitments, and variable lease costs.

  Strengths

  Large-scale cloud infrastructure and long-term commitments help lock in AI foundational resources.

  Weaknesses

  Variable lease costs exceed 10% of total lease costs, and could become more important after AI data center lease commencements.

  Comparison

  Procurement commitment growth appears comparatively more moderate versus some peers, but it remains within the industry trend of increasing off-balance-sheet commitments.

  Risks

  Rising power, maintenance, tax, and utilization-linked variable costs.

Key data

• Total AI-Related Long-Term Commitments>$1.3tnMade up of procurement and lease commitments from hyperscale cloud providers and Nvidia, supporting supplier and developer financing.
• Purchase Obligations>$640bnAs of the latest disclosures, purchase obligations for hyperscale cloud providers and Nvidia total more than $640bn, most of which relate to data centers and technology infrastructure.
• Lease Payments Not Yet Commenced~$675bnThese leases have not commenced, so they remain off-balance-sheet commitments before commencement.
• META Commitments / Forward Operating Cash Flow~1.7xShows a clearly higher commitment scale relative to cash flow.
• ORCL Commitments / Forward Operating Cash Flow>7xShows a higher commitment burden among the sample companies.
• Nvidia Cloud Services Agreement Commitments$27bnSize of cloud services agreement commitments disclosed by Nvidia.
• Oracle Compute Capacity Arrangements$10bnSize of compute capacity arrangements disclosed by Oracle.
• Meta Contractual Commitments$131bnMeta did not separately quantify third-party cloud capacity arrangements, but stated they are included within the $131bn contractual commitments.
• Hyperscale Cloud Provider On-Balance-Sheet Finance Lease Liabilities$82bnFinancing lease liabilities already reflected on the balance sheet.
• Hyperscale Cloud Provider On-Balance-Sheet Operating Lease Liabilities$175bnOperating lease liabilities already reflected on the balance sheet.
• Year-over-Year Increase in Future Lease Commitments+$435bnFuture lease commitments increased from about $240bn to more than $670bn.
• Variable Lease Cost ShareMETA 25%, GOOGL 30%, AMZN >10%Variable lease payments are often not included in lease liabilities but can still be material operating costs.
• AI Data Center Power Cost Example~$2.5bnEstimated using ~ $80/MWh, 200MW critical IT load, and a 15-year lease term; no growth was assumed.
• Alphabet Lease Backstop~$17bnAlphabet provides a lease backstop to bitcoin mining customers with data centers under construction to reduce construction financing costs and support TPU adoption.

Impact & implications

For investors, the implication is that traditional balance-sheet leverage may not fully capture economic obligations from AI infrastructure expansion. Long-term commitments can lock in capacity, support supply-chain financing, and create competitive advantage when demand is strong; however, if AI demand, pricing, or utilization are below expectations, commitment providers may face higher fixed costs, difficult contract renegotiation, credit-rating adjustments, and cash flow pressure. Investors should include off-balance-sheet commitments, variable lease costs, PPAs, and guarantees in leverage and valuation analysis.

Risks

• Rapid expansion of off-balance-sheet commitments can make true economic leverage higher than balance-sheet leverage.
• Long-term procurement, leasing, take-or-pay, and PPAs reduce financial flexibility when AI demand changes.
• Many obligations are recognized on balance sheet only when delivery, lease commencement, or payments become probable, so investors may underestimate future cash flow pressure.
• Financing for data centers, power, and chip supply chains depends on high-credit-quality customer commitments; if expectations change, contract renegotiation can be difficult.
• Variable lease payments, renewal options, and residual-value guarantees may be understated in reported lease liabilities.
• Third-party lease backstops and guarantees may be treated as debt by rating agencies, affecting credit metrics.
• AI ecosystem circular transactions and greater contractual complexity increase the difficulty of interpreting disclosures.

What to watch

• Purchase obligations, contractual commitments, lease commitments, and guarantee footnotes in each company’s 10-K and 10-Q filings.
• The delivery schedules of data center leases not yet commenced, and the pace of bringing commitments on balance sheet across 2026-2031.
• Variable lease cost shares, especially payments linked to power, maintenance, taxes, insurance, and compute usage.
• PPA tenor, fixed pricing, minimum purchase volumes, and whether they support project financing.
• The ratio of Nvidia’s inventory and procurement commitments to future revenue.
• Changes in commitments-to-operating cash flow multiples for META, GOOGL, MSFT, AMZN, and ORCL.
• How rating agencies adjust leases and guarantees, including backstops, as debt for companies such as Alphabet.
• How AI compute demand, utilization, and price changes affect the economics of long-duration contracts.