On May 5, 2026, The Information reported that AI startup Anthropic has committed to spend about $200 billion on Google Cloud and related chip services over the next five years (approximately RMB 1.37 trillion). The massive pledge stems from a multi-gigawatt TPU compute agreement signed in early April between Anthropic, Google and Broadcom, to be phased in starting in 2027. Once disclosed, the news sparked market debate: is this a routine commercial cooperation in AI infrastructure, or a 'left-hand giving to the right-hand' cycle of capital and compute between tech giants and AI firms?

This unprecedented-sized deal is not unique. In 2025, OpenAI signed a cloud compute procurement agreement with Oracle worth up to $300 billion. Anthropic has also previously reached multi-billion-dollar investment and cloud-service arrangements with Amazon. Google pledged in April to invest up to $40 billion in Anthropic (an initial $10 billion, with a further $30 billion contingent on milestones). Together, these deals sketch an "investment-plus-procurement" ecosystem unique to the AI era.

Deal details: investment for compute, binding for growth

According to public information, the core of the Anthropic-Google agreement is the provision of multiple gigawatts of next-generation Tensor Processing Unit (TPU) compute for training and inference of the Claude model family. Anthropic's $200 billion commitment will be mainly spent on Google Cloud services and related chip purchases, accounting for more than 40% of Alphabet's cloud business backlog.

The arrangement is highly reciprocal:

For Google: the large investment secures a strategic partnership with Anthropic while locking in long-term, high-value cloud revenue. Alphabet's share price rose noticeably after the disclosure.
For Anthropic: it secures stable, large-scale compute to support rapid expansion of the Claude models and enterprise applications. Anthropic's valuation has reached tens of billions of dollars, but the compute costs of training frontier models are extremely high.

Similar models recur across the industry. The OpenAI-Oracle $300 billion agreement likewise used an "cloud-giant investment + startup procurement" structure. These deals have swollen cloud providers' revenue backlogs; it is estimated that agreements involving Anthropic and OpenAI contribute roughly $2 trillion in future revenue commitments across Amazon, Google, Microsoft and Oracle.

Underlying logic of the "left-hand-to-right-hand" pattern

The public has nicknamed such deals a "left-hand giving to the right-hand" pattern: cloud giants first back AI startups through equity investments, and startups then route large sums back to the investors' cloud infrastructure businesses. This cycle has several drivers:

Compute as a core barrier: training and deploying frontier AI models requires massive compute resources; pure market procurement cannot fully meet requirements for security, customization and scale. Binding agreements ensure priority access.
Shared risk and aligned incentives: combining investment and procurement deeply ties both parties, reducing unilateral risk and creating joint incentives in regulation, standards and ecosystem building.
Capital efficiency: for cloud providers, direct investment offers greater strategic value than merely selling compute; for AI firms, securing funding while locking compute is more efficient than independently financing and building data centers.

Anthropic's continued partnerships with Amazon, Google and others also reflect a "multi-cloud and diversification" strategy to avoid vendor lock-in.

Broader impacts on the AI industry and market

1. Accelerating an AI infrastructure arms race

A $200 billion agreement further raises the compute threshold. Smaller AI firms will find it harder to access top-ti