Anthropic Plans to Spend $518 Billion on Cloud and Computing Power

Anthropic reported $4.6 billion in 2025 revenue, while its annual revenue run rate exceeded $65 billion by July 2026; the company is still losing money and is seeking a $2 trillion IPO valuation. Its prospectus outlines $518 billion in cloud and computing spending over the next 10 years, while TrendForce projects high-bandwidth memory prices will rise 121% year over year in 2027. Rapid AI adoption and constrained data-center capacity support supplier pricing power, but Anthropic’s valuation case depends on sustained growth and eventual profitability.

Analysis

The investable signal is not AI demand by itself, but who captures returns after the build-out bill. Hyperscalers can monetize scarce compute through cloud contracts, yet incremental revenue must outrun power, depreciation and financing costs; contracted demand does not establish attractive returns. For Amazon.com and Alphabet, watch cloud growth alongside capex, depreciation and free cash flow—not revenue growth in isolation. The second-order winners may be constrained inputs: memory makers, power providers, grid equipment and data-center cooling suppliers. Higher component prices are a margin headwind for compute buyers even as they support supplier pricing power; capacity additions could eventually reverse that pricing leverage.

Nebius is a higher-beta scarcity trade: rising capacity prices help only if it can bring megawatts online, secure power and fund construction without excessive dilution or customer concentration. Short-term contract pricing should not be extrapolated to long-duration economics. Anthropic’s revenue run rate is not equivalent to recognized revenue, cash generation or proof of customer retention; verify the prospectus assumptions, committed versus optional cloud spend, and capex obligations before treating the spending figure as a firm order book.

Near term, sentiment can lift AI infrastructure names, but the 1–3 month test is hyperscaler earnings commentary on capacity, capex and cloud monetization. Over 6–18 months, power availability, utilization and depreciation determine returns. Contrarian risk: investors may be pricing component scarcity as permanent while underpricing the cost of converting demand into profitable capacity. The thesis weakens if hyperscaler free cash flow deteriorates without corresponding cloud acceleration, or if component lead times and prices roll over before new capacity is absorbed.

Key Decisions for Investors

• Avoid buying the AI-capex theme indiscriminately. Track AMZN and GOOG for cloud growth versus capex, depreciation and free-cash-flow conversion; reconsider exposure if spending rises while cloud monetization stalls.
• Treat NBIS as a tactical, high-volatility capacity-pricing exposure rather than a proven compounder. Wait for evidence on commissioned megawatts, utilization, power access and funding; cut the thesis if delivery delays or dilution undermine growth.
• Watch memory, power and data-center infrastructure suppliers as potential relative beneficiaries of scarcity, but require evidence that pricing gains translate into earnings rather than being offset by capacity investment or customer pushback.
• Over the next 1–3 months, verify whether Anthropic’s stated commitments are binding and how much of its run rate converts to collected revenue and gross profit. Without that evidence, do not use its growth trajectory alone to justify a broad AI valuation premium.