Google guarantees a $15B bank loan backing Anthropic's Texas data-center buildout
Hyperscalers are increasingly financing AI-lab buildout through off-balance-sheet guarantees to third-party developers rather than direct capex or public bond issuance — Google backstopping Nexus Data Centers' bank debt for Anthropic's Texas campus is the same structure already seen this week from Meta/BlackRock (El Paso) and Nvidia/OpenAI. Read against the standing rate picture (capital-context.yaml: 30-year Treasury near 5.24%, highest since 2007), a guarantee lets a AAA-adjacent balance sheet lower a project's borrowing cost without the parent issuing debt itself or carrying the asset on its own books — cheaper capital, but the credit risk is real even if it's contingent rather than funded.
Scenarios 2
The pattern keeps spreading — more hyperscaler-guaranteed, third-party-developer financing structures appear as the default way to fund frontier-lab compute buildout, keeping headline capex figures from fully capturing the hyperscalers' true credit exposure.
Three such structures (Meta/BlackRock, Nvidia/OpenAI, Google/Anthropic) landing inside one week suggests this is becoming the standard playbook, not a one-off, especially while long-end borrowing costs stay elevated.
Precedent — Meta's El Paso JV with BlackRock (07-23/24, ~$10B/1GW) and Nvidia's guarantee structure backing OpenAI's compute commitments both used the same off-balance-sheet logic within the same week, establishing the pattern this item extends.
A single high-profile default or credit event on one of these guarantee structures forces rating agencies and lenders to reprice the whole category, tightening terms across all of them at once.
Contingent guarantees are, by construction, thinly disclosed until something goes wrong — a cluster of similar structures sharing correlated underlying risk (AI-compute demand) is exactly the setup where one stress event repricing the group is a known pattern in credit markets.
Context
Sits alongside this week's CoreWeave debt-repricing (spread widened to 5.5pp over benchmark, priced at 96-97 cents on the dollar) as a second data point on how AI-infrastructure credit is actually priced right now, not just how much capex is guided — the standing macro picture's elevated long-yield environment (capital-context.yaml rate_regime) is the backdrop both are pricing against.
This is generated reasoning, not a sourced fact — the mechanism and scenarios above are the model's read on what this item could mean for capital flows, tagged with its own confidence rather than stated as settled. Back to Global Capital.