OpenAI and Anthropic now sit at the center of Big Tech’s AI cloud backlog
OpenAI and Anthropic have become the fulcrum of a structural accounting dynamic where Big Tech giants simultaneously fund AI startups, book their cloud spending as revenue, and recognize paper gains on inflated equity stakes, raising hard questions about revenue quality and cash sustainability for institutional investors evaluating these cloud giants’ earnings. The $2 trillion in future cloud commitments tied to these two firms now represents the earnings visibility story that consensus valuations depend on, yet operates on a circular cash flow mechanism rather than genuine customer demand.
- Microsoft’s $13 billion OpenAI investment consisted primarily of Azure credits that generated $60 billion annual cloud revenue for OpenAI, exceeding its $25 billion revenue twofold.
- Anthropic spent $2.66 billion on AWS in nine months, nearly matching its total revenue in the same period, with Amazon recognizing $16.8 billion in mark-to-market gains on its stake.
- OpenAI and Anthropic account for more than half of the nearly $2 trillion in forward cloud commitments recorded by Microsoft, Oracle, Alphabet, and Amazon combined.
- $60B OpenAI annual cloud bill against $25B annual revenue, or 2.4x coverage ratio
- $16.8B Amazon Anthropic equity gains versus $30.3B total Q1 2026 earnings, or 55% of profit
- 95% Amazon free cash flow decline to $1.2 billion, reflecting infrastructure investment velocity
The artificial intelligence investment cycle has crystallized around a structural problem that institutional investors have largely overlooked: the world’s largest cloud providers have engineered a self-reinforcing loop in which they fund AI startups, those startups deploy that funding exclusively on the same providers’ infrastructure, and the providers then recognize both the funding as customer cloud revenue and the startups’ rising valuations as unrealized gains on their balance sheets.
OpenAI and Anthropic sit at the center of this mechanism, with fresh regulatory filings revealing that these two companies alone account for more than half of the nearly $2 trillion in future cloud spending that Microsoft, Oracle, Alphabet, and Amazon have booked as committed revenue.
The arrangement is not illegal, but it operates on a circular cash flow model that obscures whether genuine economic demand exists or whether capital is simply moving through a series of related-party transactions disguised as separate business activities.
Microsoft Books $60 Billion OpenAI Cloud Revenue From Its Own $13 Billion Investment
The Microsoft-OpenAI relationship illustrates the mechanics of the structure with unusual clarity. Microsoft deployed approximately $13 billion into OpenAI through a funding agreement, yet the vast majority of that capital took the form of Azure compute credits rather than cash.
OpenAI has no choice but to consume those credits by purchasing cloud infrastructure from Microsoft to train and run its models. This requirement transformed what might appear as a capital investment into a pre-committed customer contract, where the “customer” is the very company receiving the funding.
The consequence is a doubling of OpenAI’s infrastructure costs relative to its revenue. OpenAI’s annual cloud bill has now climbed above $60 billion per year, while its total revenue remains around $25 billion, creating a cost-to-revenue ratio of 2.4 to 1.
In conventional enterprise software or SaaS contexts, such a ratio would signal unsustainable unit economics and trigger immediate strategic review. Within the AI financing narrative, however, Wall Street has reframed the same numbers as evidence of scaling and growth velocity.
Microsoft simultaneously benefits on two lines of its income statement. The Azure division recognizes $60 billion in annual cloud revenue attributed to OpenAI’s infrastructure consumption, bolstering cloud segment growth metrics that drive institutional valuations.
Independently, Microsoft’s corporate investments division marks up its equity stake in OpenAI as the startup’s valuation rises, creating unrealized paper gains that flow through non-operating income. Neither mechanism requires OpenAI to convert its cloud spending into actual product revenue or customer cash payments.
Anthropic Spends Nearly All Revenue Back to AWS Within Nine Months
Anthropic operates under an analogous arrangement with Amazon Web Services. Filings show Anthropic spent approximately $2.66 billion on AWS infrastructure during a nine-month period, nearly matching its total revenue for the same window.
This spending ratio means Anthropic is returning nearly 100 cents of every dollar earned directly back to AWS in the form of compute bills, creating a closed loop in which Amazon funds Anthropic, Anthropic purchases AWS capacity, and Amazon records the purchase as customer cloud revenue while simultaneously marking up its equity position in Anthropic as the company’s valuation inflates.
The equity gains have become material to Amazon’s reported profitability. In the first quarter of 2026, Amazon posted $30.3 billion in total earnings, of which Anthropic-related mark-to-market gains accounted for $16.8 billion, or 55 percent of consolidated profit.
That ratio signals a dangerous convergence: Amazon’s headline earnings increasingly depend on paper gains from AI startup equity stakes rather than on cash generated from actual customer relationships or products sold.
The mismatch between reported earnings and cash generation has begun to surface in Amazon’s cash flow metrics. The company’s free cash flow fell by 95 percent to $1.2 billion, a dramatic contraction that reflects the enormous capital outlays required to build the infrastructure that Anthropic and other AI firms are consuming.
The gap between earnings (boosted by equity mark-ups) and cash flow (pressured by real infrastructure investment) creates accounting asymmetry that institutional investors must now reconcile.
The $2 Trillion Cloud Backlog Depends Entirely on AI Startup Funding Cycles
The structural risk extends across the entire cloud sector. Microsoft, Oracle, Alphabet, and Amazon have collectively booked nearly $2 trillion in future cloud revenue commitments, with OpenAI and Anthropic accounting for more than half of that total.
These figures appear in quarterly disclosures as contracted future revenue, contributing directly to analyst models and consensus earnings forecasts that underpin institutional capital allocation decisions.
Yet the visibility on which those forecasts rest is not rooted in independent customer demand or market-tested pricing; it exists because two AI startups received funding from the same companies now counting that funding as customer contracts.
If OpenAI or Anthropic encounter valuation pressure, face regulatory restrictions on their operations, or experience slower-than-expected scaling in actual AI product adoption, the cloud spending commitments that justify $2 trillion in recorded future revenue would evaporate.
The companies would still own the infrastructure, but the revenue visibility that supports current cloud division valuations would disappear. This creates a hidden leverage point in the cloud sector’s investment thesis: institutional investors are implicitly betting not on cloud infrastructure demand broadly, but specifically on the continued growth and funding of two private AI startups.
The arrangement also creates a perverse incentive structure. Cloud providers benefit from funding AI startups regardless of whether those startups achieve profitable operations or reach genuine product-market fit. The funding itself generates immediate cloud revenue recognition and equity upside, independent of any external validation of the AI business model.
This decouples infrastructure investment from the underlying demand signals that normally govern capital allocation in competitive markets.
Mark-to-Market Gains Replace Cash Profits as the Driver of Cloud Giant Earnings
The shift from cash earnings to equity mark-ups as the primary source of reported profit represents a material change in how institutional investors should model these companies.
Alphabet reported $62.6 billion in earnings for the first quarter of 2026, with $28.7 billion attributed to gains on its Anthropic stake, meaning nearly 46 percent of consolidated earnings came from equity valuation increases rather than operations. Amazon’s situation was more extreme at 55 percent. These are not minority contributors; they are the dominant earnings drivers.
Equity mark-ups are inherently unstable. A downturn in AI sentiment, slower-than-expected model capability improvements, or competitive pressure on AI pricing would immediately compress AI startup valuations, erasing these paper gains. Unlike revenue from actual customer contracts, mark-up gains can reverse sharply and offer no accompanying cash benefit.
An institutional investor receiving earnings per share attributable primarily to mark-ups has less durable earnings than one whose profits derive from customer cash flows, yet the financial statements present both identically.
The accounting treatment also creates a visibility cliff. As long as AI startups remain in private markets
