Forty billion dollars. That is the staggering sum Google is dangling in front of AI powerhouse Anthropic, a deal structured to fundamentally reshape the artificial intelligence landscape.

The search giant is handing over $10 billion immediately at a $350 billion valuation. Anthropic can unlock the remaining $30 billion only if it hits specific, closely guarded performance milestones.

This cash solves a very specific problem: hardware. Building modern AI requires a terrifying amount of power.

Anthropic recently locked down a joint agreement with Google and Broadcom to secure multiple gigawatts of next-generation Tensor Processing Unit (TPU) capacity. Under the expanded partnership, Google explicitly guarantees 5 gigawatts of computing capacity for Anthropic servers by 2027.

The Compute Arms Race

The modern AI industry runs on a clever, closed-loop financial system. Cloud providers dump unprecedented capital into AI labs, which immediately turn around and spend those exact funds renting servers from their benefactors.

This circular economy pads Google's bottom line while keeping a powerful AI player locked firmly inside its cloud ecosystem. Antitrust regulators will almost certainly scrutinize the arrangement, but right now, it is the standard Silicon Valley playbook.

Anthropic ran the exact same play with Amazon earlier this month. Amazon promised up to $25 billion, dropping $5 billion upfront and keeping $20 billion in reserve for future commercial milestones. In return, Anthropic must utilize Amazon's Trainium chips for its AI models.

Add up previous funding rounds, and Anthropic's total available war chest now tops $100 billion. The company expects to burn through practically all of it establishing 5 gigawatts of compute capacity across multiple cloud networks.

Strategic Stakes and New Models

Google is not new to the Anthropic cap table. The relationship kicked off in 2023 with a $300 million check for a 10% stake, followed by an additional $2 billion injection. Before this week's mega-deal, Google already owned approximately 14% of the startup.

Google's massive check clears just as Anthropic rolls out Mythos, its most capable—and dangerous—AI model yet.

Because of severe cybersecurity risks, Anthropic restricts public access to the system. The company works exclusively with carefully vetted partner organizations to evaluate and patch the model's vulnerabilities.

Despite the heavy security, the Mythos model has already fallen into unsanctioned hands. Running a behemoth like Mythos at scale drains computing resources incredibly fast, making the Google and Amazon infrastructure deals a matter of survival rather than mere strategy.

Meanwhile, everyday demand for the company's tools is stretching servers to their breaking point.

The massive popularity of Claude Code, Anthropic's software programming tool, forces the startup to hoard every scrap of compute it can find. They need the hardware just to keep the lights on and stay in the ring with OpenAI.