OpenAI told investors in February its compute bill would be around $600bn. A July presentation puts it at $856bn.

OpenAI reportedly projects cumulative negative free cash flow of $278bn over 2026-2030, improved from an estimated $305bn in its May forecast. However, expected compute and infrastructure spending has increased sharply to about $856bn from roughly $600bn, underscoring the capital intensity and financing demands of scaling AI infrastructure.

Analysis

The key investable issue is not OpenAI’s operating loss profile itself, but who ultimately warehouses the compute obligation. If financing remains dependent on strategic partners, MSFT and ORCL face rising capital-intensity and depreciation risk before matching AI revenue is proven; this pressures FCF conversion and can compress hyperscaler multiples even while reported cloud growth remains strong. Conversely, NVDA and AVGO retain near-term revenue visibility, but their valuation risk increases if customer demand shifts from unconstrained build-out to financed, staged capacity commitments.

Over the next 1-3 months, the market will focus on whether cloud partners characterize AI infrastructure as externally monetized capacity versus internal/partner capacity. That distinction determines whether capex is rewarded as growth investment or penalized as a balance-sheet subsidy. Data-center infrastructure names VRT, ETN, DLR and EQIX are second-order beneficiaries only if signed power and colocation commitments are backed by creditworthy counterparties; a funding delay would hurt their order timing more than their long-run demand case.

The consensus may be too focused on GPU demand and too little on depreciation. A rapid upgrade cycle can leave cloud operators carrying assets whose economic lives are shorter than accounting lives, creating a margin headwind 6-18 months after the initial capex surge. The thesis is falsified if AI workload revenue and utilization rise fast enough to stabilize cloud gross margins despite accelerating depreciation, or if third-party infrastructure financing absorbs a meaningful share of the funding burden.

Key Decisions for Investors

• Maintain a 1-3 month pair: long NVDA / short MSFT in matched beta size. NVDA captures near-term shipment economics, while MSFT bears a larger risk of AI capex-to-revenue conversion disappointment; exit if Azure growth reaccelerates and management demonstrates stable or improving cloud gross margin.
• Avoid adding to ORCL on AI backlog headlines until its next earnings call clarifies customer prepayments, contract duration, and capex funding. A favorable setup requires backlog converting into operating cash flow rather than further debt-funded infrastructure; otherwise use ORCL as a relative short versus AMZN.
• Buy VRT selectively on 5-10% weakness rather than chase momentum, with a 6-12 month horizon. Its upside depends on power-dense data-center deployment, but order cancellations or extended customer financing terms would be an early warning that AI infrastructure demand is being deferred.
• Set an earnings-season alert for MSFT, ORCL, AMZN and GOOGL: any sequential deterioration in cloud gross margin alongside higher capex should trigger a sector de-risking, particularly in high-multiple AI infrastructure suppliers.