Anthropic's $50B infrastructure pledge reshapes IPO strategy with owned data centres

Anthropic, the frontier AI lab behind Claude, is pivoting from renting hyperscaler capacity to owning its physical infrastructure, a strategy that will shape its confidential US listing at a reported $965B–$2T valuation. The shift became concrete in June when Apollo and Blackstone structured a $35B chip-backed special purpose vehicle (SPV) for Anthropic, with the first gigawatt delivered through UK neocloud provider Fluidstack rather than Amazon, Google, or Microsoft. Anthropic has also pledged $50B toward US computing infrastructure, building purpose-designed sites with Fluidstack in Texas and New York—creating 800 permanent jobs and 2,400 construction roles. CEO Dario Amodei says the spend is necessary to support continued frontier AI development. This complements existing commitments: a $150B chip agreement with Google, an $11B AWS campus in Indiana, and multi-cloud access to Google TPUs and Microsoft Azure. By owning capacity, Anthropic gains delivery certainty and custom-built facilities over balance-sheet rental relationships.

The financing landscape behind this build-out is increasingly circular. SpaceX's IPO documents revealed Anthropic pays $1.25B per month for 300MW at xAI's Colossus data centres through May 2029, alongside a separate $30B compute deal between SpaceX and Google. Nvidia formalized a $500B financing platform with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, treating GPUs as collateral—what Jensen Huang calls an "investable asset class." Power supply is now a critical bottleneck: analysts expect Anthropic's S-1 to list compute and power availability as risk factors, pricing grid interconnection risk for the first time. Neocloud precedent looms large—CoreWeave's debt hit $35B at end of Q2, driven by $9.4B capex. Anthropic's move toward asset-backed debt is likely next as it transitions to owning facilities.