Exclusive-Anthropic IPO prospectus lays bare deep dependence on Big Tech partners

Anthropic's confidential IPO filing shows 47% of its 2025 revenue, or $2.16 billion, was routed through Amazon and Google cloud marketplaces, underscoring growing reliance on investors, suppliers and AI competitors. Revenue surged 12-fold to nearly $4.6 billion in 2025, but operating losses more than doubled to over $8 billion and long-term computing commitments exceeded $417 billion by early 2026. The company is targeting an approximately $2 trillion valuation, while warning that partner conflicts, concentrated customers and potential disruption to third-party cash collections could impair compute access and cash flow.

Analysis

For AMZN, GOOG and MSFT, the relevant read-through is not simply AI demand: marketplace distribution converts third-party model usage into high-margin cloud-adjacent revenue while improving utilization of reserved GPU capacity. The offset is customer concentration and circularity—if Anthropic moderates its compute buildout or faces a funding reset, cloud revenue, deferred capacity monetization and AI-capex returns could all be questioned simultaneously. AMZN has the greatest asymmetric exposure because AWS is both a strategic channel and a major capacity provider; GOOG benefits similarly but has more internal-model substitution risk.

Over the next 1-3 months, the IPO process should force investors to normalize reported AI revenue for pass-through marketplace economics and assess cash conversion rather than headline growth. This is potentially negative for a premium IPO multiple, but constructive for the hyperscalers if the filing validates durable infrastructure commitments without incremental pricing concessions. A weaker-than-expected IPO reception would be a second-order negative for SMCI and other AI hardware proxies: it would raise the market's required return on GPU infrastructure before demand is fully contracted with end customers.

Consensus may be understating the option value held by the cloud platforms: distribution data, billing control and compute allocation create leverage over an AI vendor that cannot easily replicate their enterprise routes to market. Conversely, that leverage makes Anthropic's economics less independent than a software valuation framework implies. The key falsifier for the cloud-positive thesis is evidence of falling marketplace take rates, a material rise in receivables days, or any reduction in dedicated-capacity commitments rather than continued nominal AI demand growth.

Key Decisions for Investors

• Maintain an overweight AMZN versus SMCI for the next 3-6 months: AWS can monetize model demand through infrastructure and marketplace fees, while SMCI is exposed to a sharper capex-duration reset. Reassess if AWS commentary indicates AI capacity is becoming customer-funded only at lower returns.
• Use a long AMZN / short SMCI pair around the IPO marketing window, sized modestly; the expected catalyst is greater scrutiny of AI infrastructure payback. Exit if the issuer discloses contracted, take-or-pay hardware demand materially beyond current capacity commitments or SMCI raises backlog visibility with improved gross-margin guidance.
• Do not chase GOOG or MSFT solely on this development. Set alerts for disclosed marketplace fee rates, cloud receivable aging and any exclusivity or capacity-priority provisions; improved economics would support adding to GOOG, while deteriorating collections would challenge the cloud AI-revenue multiple.
• For IPO participation, require valuation work based on net revenue after channel costs, cash collection timing and committed-compute coverage—not gross consumption revenue. Avoid a directional allocation until the prospectus quantifies renewal behavior and the funding plan for long-dated infrastructure obligations.