Anthropic commits $200bn to Google Cloud and TPU chip capacity
TL;DR:
- Anthropic has committed to spend $200bn (£148bn) with Google Cloud over five years as part of a recent agreement, The Information reported, citing a person with knowledge of the matter.
- The commitment suggests Anthropic accounts for more than 40% of the cloud revenue backlog Google disclosed to investors last week. Contracts involving Anthropic and OpenAI now account for more than half of the $2tn (£1.48tn) backlog at AWS, Microsoft Azure and Google Cloud combined.
- Resultsense view: at this scale, AI infrastructure capex is no longer a sub-line of cloud — it is the cloud. UK enterprise procurement teams should rethink multi-cloud strategy on the assumption that hyperscaler pricing power, capacity allocation and reliability are now substantially shaped by frontier-lab commitments.
Anthropic signed a deal in April with Google and chip partner Broadcom for multiple gigawatts of tensor processing unit (TPU) capacity, expected to come online from 2027. Alphabet is also investing up to $40bn in Anthropic, deepening a partnership with the AI start-up that is also a Google rival.
Why the backlog matters
The cloud “revenue backlog” is the running total of contractually committed customer spend that hasn’t yet been recognised as revenue. The fact that two AI labs — Anthropic and OpenAI — together account for more than half of the $2tn backlog at the three big US hyperscalers is structural, not anecdotal. It means hyperscaler capacity planning, chip procurement, power provisioning and pricing models are now dominated by AI-lab demand for the foreseeable future.
For everyone else — UK retailers, banks, public-sector tenants, SaaS vendors — the practical effects are: capacity allocation may tilt toward the labs in periods of constraint, reserved-instance and committed-use discount terms may become harder to renegotiate, and reliability/SLA commitments may get repriced as hyperscalers concentrate operational attention on lab workloads.
The Anthropic compute strategy
Anthropic has signed a series of major compute agreements over the past quarter to keep up with demand for its Claude models. Last month it struck a multi-year deal with cloud-infrastructure firm CoreWeave, and is set to secure nearly 1 gigawatt of capacity via Amazon’s Trainium chips by year-end. Anthropic has previously stated that it trains and runs Claude on a range of AI hardware including AWS Trainium, Google TPUs and Nvidia GPUs — explicitly multi-vendor.
Alphabet shares rose around 2% in extended trading on Tuesday following the report, and Alphabet is now closing in on overtaking Nvidia as the world’s most valuable company.
UK relevance
Three implications for UK enterprises. First, AI infrastructure capacity available to UK customers is being negotiated against US-domiciled lab demand at a scale that dwarfs the entire UK enterprise market. UK CIOs should assume capacity-tightness risk when planning 2027/28 AI builds. Second, the practical reach of UK sovereign-AI ambitions is constrained by this concentration: any “build it here” alternative needs to be sized against a market where the largest single buyers are already committed multi-billions to US hyperscalers. Third, multi-cloud is now structurally more expensive: every hyperscaler is repricing to recover the capex commitments AI labs are driving.
Looking forward
The next data point to watch is the Q3 2026 cloud-revenue disclosures from Microsoft Azure and AWS, which will show whether their backlogs are similarly lab-concentrated. If the pattern holds across all three, UK enterprise pricing renegotiations from late 2026 onwards will face a materially harder negotiating environment. CFOs should consider locking longer-dated reserved capacity commitments while supplier flexibility remains.