Anthropic May Spend $750 Billion on AI Compute, Locking in $450 Billion in Non-Cancellable Cloud Contracts Ahead of IPO

Anthropic’s push to dominate the frontier AI market comes with an unprecedented price tag: the Claude developer could spend as much as $750 billion on raw computing capacity over the coming years, according to details from its draft IPO prospectus first reviewed by The Wall Street Journal.

The disclosure lays bare the staggering scale of capital required to train and run foundation models at commercial volume. In the first six months of 2026 alone, Anthropic burned through nearly $9 billion on compute and data center capacity—almost quadrupling the $2.3 billion it spent during the same stretch in 2025 and outpacing its total infrastructure outlays for all of last year.

The $450 Billion Take-or-Pay Bet To ensure it doesn't get boxed out of scarce silicon, Anthropic has entered massive long-term commitments across the cloud triumvirate, locking down approximately $450 billion in compute contracts over a seven- to 10-year horizon. Crucially, the draft prospectus notes that the bulk of these deals are non-cancellable, requiring the startup to pay regardless of actual server utilization:

• Google Cloud: $111 billion in binding infrastructure agreements.

• Amazon Web Services (AWS): $110 billion in dedicated capacity commitments.

• Microsoft Corp: $31 billion in specialized compute access.

Beyond existing agreements, ongoing talks could push the company’s forward-looking compute tab up to the full $750 billion mark.

Public Listing as a Capital Lifeline Anthropic’s anticipated initial public offering is no longer just a milestone for early backers; it has become an operational necessity to fund these obligations.

The public debut aims to refill the company’s war chest to bankroll specialized AI chips, bespoke data center buildouts, and elite engineering talent. While Anthropic's top-line revenue has accelerated sharply—hitting $16.3 billion in the first half of 2026—the sheer volume of fixed cloud spending means the firm must raise massive public market liquidity to avoid cash-flow choke points.

Mounting Bottlenecks and Political Friction Despite its willingness to spend at historic scale, Anthropic's filings outline critical operational hurdles:

• Hyperscaler Competition: As Amazon, Google, and Microsoft invest heavily in their own competing in-house AI agents and foundational models, Anthropic faces the risk of constrained allocations from the very partners hosting its workloads.

• Domestic Grid and Community Pushback: Most of Anthropic’s contracted capacity sits within the United States, where proposed data center campuses are colliding with local resistance over strained electrical grids, water consumption, and land use.

The scale of Anthropic’s compute commitments shows that the foundation model race is no longer won on algorithmic cleverness alone—it is an industrial war of financial endurance.