Anthropic is said to have committed $200 billion to Google Cloud services and chips over five years, a pact that would turn one of the world’s fastest-growing artificial intelligence developers into a cornerstone customer for Alphabet’s infrastructure business.The reported agreement underlines how the economics of generative AI are being reshaped by a small group of companies capable of securing vast compute capacity. Anthropic, the maker of Claude, is racing to support heavier enterprise use, more capable models and wider product distribution. Google, meanwhile, is using its cloud platform, tensor processing units and semiconductor partnerships to challenge Amazon Web Services and Microsoft Azure in the most expensive phase of the AI build-out.
The commitment would account for more than 40 per cent of Google Cloud’s revenue backlog, which stood at about $462 billion at the end of March. Google Cloud revenue rose 63 per cent to about $20 billion in the first quarter of 2026, helped by demand for AI infrastructure, enterprise AI products and core cloud services. The scale of the Anthropic commitment indicates that a large portion of future cloud growth is now tied to a small number of frontier model developers.
The deal is expected to involve Google Cloud services and Google-designed TPUs, with Broadcom playing a role in supplying chip capacity. Anthropic had already announced an expanded arrangement with Google and Broadcom for multiple gigawatts of next-generation TPU capacity, expected to begin coming online from 2027. That followed an earlier plan to expand use of Google Cloud technologies, including access to up to one million TPUs and well over a gigawatt of capacity.
Anthropic’s compute strategy is deliberately diversified. The company uses Google TPUs, Amazon’s Trainium chips and Nvidia graphics processors, reducing dependence on a single hardware supplier while maximising access to constrained capacity. That approach also gives Anthropic leverage as cloud providers compete for its long-term spending and prestige as a leading AI customer.
Alphabet’s relationship with Anthropic extends beyond cloud services. Google has built a major financial position in the company and is expected to invest up to $40 billion in cash and compute support, subject to performance milestones. That arrangement strengthens the bond between Anthropic’s model development and Google’s infrastructure ambitions, while raising questions about how investment, cloud sales and future revenue recognition are becoming intertwined across the AI sector.
Amazon remains another crucial partner. Anthropic has secured up to 5 gigawatts of capacity from Amazon, including Trainium2 and Trainium3 infrastructure. The company has also disclosed a strategic partnership with Microsoft and Nvidia involving $30 billion of Azure capacity, as well as a $50 billion investment in US-based AI infrastructure with Fluidstack. The pattern shows that frontier AI companies are no longer buying cloud services in ordinary increments; they are locking up power, chips and data centre capacity years in advance.
For Google, the Anthropic commitment strengthens the case that its custom-chip strategy can become a serious revenue engine. TPUs were developed internally to accelerate machine-learning workloads and have become a key differentiator for Google Cloud. Demand from Anthropic provides validation at a time when hyperscalers are under pressure to prove that capital spending on AI will translate into durable revenue and profits.
The risks are substantial. AI data centres require enormous capital expenditure, long construction timelines, energy supply guarantees and specialised chip procurement. Google’s own filings show hundreds of billions of dollars in purchase commitments and infrastructure obligations linked to technical capacity, energy and long-term supply agreements. A large customer concentration in AI also exposes cloud providers to shifts in model economics, pricing pressure and the possibility that today’s leading platforms may not retain their current momentum.
Topics
Technology