Anthropic Plans to Spend $518 Billion on Cloud and Data Centers. More Than $100 Billion Is Already Promised to Amazon.

Anthropic's draft IPO prospectus reportedly outlines $518 billion of cloud, computing and infrastructure obligations, including roughly $110 billion payable to AWS from May 2026 through April 2036. Anthropic revenue accelerated from nearly $4.6 billion in all of 2025 to $4.73 billion in Q1 2026 and more than $11.5 billion in Q2, supporting the affordability of an AWS commitment exceeding $10 billion annually. Key risks are Anthropic's more than $8 billion operating loss in 2025 and customer concentration, with nearly one-quarter of 2025 revenue coming from two customers whose spending is not secured by long-term contracts. Amazon's Anthropic holdings were marked at about $190 billion as of June 30, while investment-related gains accounted for most of its $53.4 billion in Q2 pre-tax non-operating income.

Analysis

The relevant AMZN read-through is less incremental AWS demand than a conversion of speculative AI capacity into long-duration, take-or-pay backlog. That improves utilization visibility and supports data-center depreciation absorption over the next 12-36 months, but it also pulls forward AWS capex: the cash-flow benefit will lag reported cloud revenue. AMZN therefore has a more favorable revenue-duration profile than MSFT and GOOG, while AVGO benefits indirectly if contracted capacity translates into sustained accelerator/networking procurement rather than merely reserved power.

The non-obvious risk is circular counterparty exposure. AMZN is simultaneously a major infrastructure creditor/vendor and a holder of a large, privately marked financial asset; a slowdown in Anthropic enterprise consumption could impair AWS booked demand, force capacity underutilization, and reduce the investment mark in the same cycle. This makes headline EPS a poor valuation anchor for AMZN until the investment gain is separated from operating earnings and the IPO establishes an independently price-discovered value.

Near term, a prospective IPO can validate AMZN's stake and compress perceived counterparty risk, but the larger catalyst over 1-3 months is disclosure on cancellation rights, prepayment/security, annual payment timing, and the share of commitments that are genuinely irrevocable. Consensus may be too optimistic in annualizing exceptional revenue growth: with meaningful customer concentration and limited customer contract duration, enterprise churn can surface faster than infrastructure obligations can be resized. Over 6-18 months, the winners are cloud vendors with diversified AI demand and monetizable reserved capacity; the loser is whichever supplier carries excess power and GPU capacity if Anthropic's revenue conversion normalizes.

Key Decisions for Investors

• Maintain a tactical long AMZN only through the IPO/disclosure catalyst, sized modestly: the contracted AWS backlog can support cloud revenue visibility, but use operating-income/FCF valuation rather than reported EPS. Reassess if AWS capex rises faster than revenue for two consecutive quarters or if disclosed Anthropic protections are weaker than take-or-pay economics imply.
• Prefer a relative long AVGO / short AMZN basket over 3-6 months for AI infrastructure exposure: AVGO participates in continued scale-out with less direct exposure to a single customer's private-valuation mark. Exit if hyperscaler capex guidance decelerates materially or Broadcom's AI semiconductor revenue guide misses by more than 10%.
• Do not add to AMZN on a valuation uplift from its Anthropic holding alone. Treat an IPO valuation above private marks as a tradable upside catalyst, but take profits if the stock rerates primarily on non-operating gains rather than AWS margin expansion.
• Set an event alert for the public prospectus: initiate no new cloud-capacity trade until payment schedules, termination provisions, collateral, and Anthropic customer-retention metrics are disclosed. Evidence that obligations are cancellable or contingent would be a direct negative for AMZN, MSFT, GOOG, and AVGO AI-demand assumptions.