Anthropic Could Raise Up to $100 Billion in Its November IPO

Anthropic is reportedly considering an IPO that could raise up to $100 billion at a roughly $2 trillion valuation, but the proceeds would cover only about 24% of its approximately $413.7 billion in largely fixed infrastructure commitments. About 82.6% of 2025 revenue was usage-based, while roughly 80% of its $518 billion compute plan is effectively fixed, creating material downside risk if customers reduce Claude spending. The proposed valuation implies about 31x its $65 billion July 2026 annualized revenue run rate, despite customer concentration and dependence on Amazon and Google, which generated 47% of 2025 sales through cloud marketplaces.

Analysis

The relevant public-market issue is not near-term AI demand but credit quality embedded in the compute buildout. A large equity raise would improve liquidity optics without eliminating the duration mismatch between volatile inference revenue and long-dated capacity obligations; that mismatch can convert modest usage deceleration into a sharp cash-burn inflection. This is most consequential for suppliers whose backlog is being valued as quasi-sovereign hyperscaler demand rather than as demand ultimately dependent on a single AI application vendor.

NVDA's prospective investment should be treated partly as demand-financing rather than a clean third-party validation signal. If supplier capital, cloud credits, and customer commitments increasingly support one another, the market will eventually discount reported AI infrastructure bookings through a higher cancellation/renegotiation probability. AVGO has a more nuanced exposure: contracted equipment economics can protect near-term revenue recognition, but any customer liquidity stress raises residual-value, receivable, and renewal risk over a 6-18 month horizon.

AMZN and GOOG retain the best downside protection because they can redeploy scarce compute to internal products and other enterprise tenants, but marketplace concentration creates bargaining leverage for them rather than for the model provider. The contrarian view is that a public listing could initially expand AI multiples by creating a scarce pure-play benchmark; the more investable second-order trade may emerge 1-3 months later, when IPO disclosures reveal minimum-purchase terms, collateral requirements, and whether capacity commitments are gross or net of partner subsidies.

Key Decisions for Investors

• Maintain NVDA exposure but reduce incremental buying into any IPO-driven sympathy rally; use a 3-6 month NVDA put spread funded by an upside call sale if valuation expands without a corresponding upward revision to data-center revenue guidance. Thesis fails if disclosed customer diversification and cash prepayments demonstrate that demand is not vendor-financed.
• Prefer long AMZN / short AVGO as a 6-12 month relative-value expression of AI-capex quality. AMZN has cloud utilization optionality and diversified cash flow, while AVGO is more exposed to the market continuing to capitalize long-duration AI equipment commitments at low credit risk; exit if AVGO backlog converts with stable receivables and no customer-concentration disclosure deterioration.
• Do not initiate an Anthropic position at listing. Place an alert for the first registration statement: consider a post-lockup short or put spread only if fixed commitments materially exceed committed financing, top-customer concentration remains above 20%, and the IPO trades above 25x forward run-rate revenue; the risk is a scarcity premium and accelerating enterprise inference adoption.
• Monitor AMZN, GOOG, and MSFT quarterly capex commentary for any shift from capacity scarcity to utilization or monetization language. A reduction in capex growth or an increase in AI workload discounting would be the earliest public signal that infrastructure suppliers' revenue assumptions require de-rating.