Google has announced a monumental investment plan for AI safety and research firm Anthropic, committing up to $40 billion in a deal that combines direct cash infusion with massive cloud computing credits. This move signals an intensifying arms race among tech giants to lock in the compute capacity necessary for the next generation of frontier AI models.
The Deal Structure
The investment is structured in two primary phases:
- Immediate Capital: Google is investing an initial $10 billion at a company valuation of $350 billion.
- Performance Milestones: An additional $30 billion is earmarked for future investment, contingent on Anthropic meeting specific performance and development targets.
This deal follows closely on the heels of Amazon's recent $5 billion investment in Anthropic, highlighting the company's unique position as a neutral player receiving support from multiple cloud rivals.
Focus on Compute and TPUs
A central component of the partnership is infrastructure. Anthropic relies heavily on Google Cloud for its training needs, specifically utilizing Google’s specialized Tensor Processing Units (TPUs).
The agreement guarantees Anthropic access to 5 gigawatts of TPU-based computing capacity over the next five years. This scale is essential for training models like the recently limited-released Mythos, which Anthropic describes as its most powerful model to date with advanced cybersecurity capabilities.
The Global AI Arms Race
The scale of this investment reflects the staggering costs of modern AI development. OpenAI has similarly been securing hundreds of billions in deals for chips, energy, and data centers.
For Google, backing Anthropic serves a dual purpose: it secures a massive, long-term customer for its Google Cloud and TPU business, while also ensuring that one of the world's most capable AI research teams remains deeply integrated into the Google ecosystem.
As Anthropic explores a potential IPO as early as October 2026, this backing from Google provides the financial and technical stability needed to compete at the absolute frontier of artificial intelligence.
Related Reading
Gemini Decision 1 Returns Probabilities, Not Text, at $0.10/M
Google's Gemini Decision 1 scores up to 1,000 typed questions with zero output tokens at $0.10 per million input tokens, leading 7 of 9 benchmarks.
Google's Gemini 4 Argon Scores 87% on ERQA, Controls Spot
Google reports Gemini 4 Argon scores 87% on ERQA and uses long prompt harnesses to drive Spot, Omni, and MuJoCo Real2Sim workflows.
Anthropic FlashProve: Proofs for 400B MoE Replies at 110 Tok/s
Anthropic's FlashProve uses zkSNARKs to prove a Llama 4 Maverick reply came from committed weights at about 110 tokens/s on one H100.