Why Is Google Paying $40 Billion for a Claude License?

There's a thought experiment worth sitting with for a minute.

You're the world's largest search company. You have one of the biggest AI research labs on earth. You published the paper that launched the modern AI era. You have your own large language model, which you've been developing for years and deploying across every product you own.

Would you write a $40 billion check to someone else's AI company?

Google would. Google did.

Last week, Google announced it's committing up to $40 billion to Anthropic — $10 billion now, at a valuation of $350 billion, with another $30 billion contingent on performance milestones. Anthropic is the company that makes Claude. Claude is, by most accounts, Google Gemini's most serious competitor in the AI assistant market.

Let that sentence breathe for a second.

The Official Explanation Makes Sense. The Full Picture Is Stranger.

Google's investment is structured to ensure Anthropic keeps running its workloads on Google Cloud — specifically on Google's Tensor Processing Units. The deal includes access to five gigawatts of compute capacity over five years. Whoever provides that compute earns significant infrastructure revenue. If Anthropic's models run on Google Cloud, Google benefits twice: from the equity stake, and from the hosting fees.

So the logic goes: Google isn't just backing a rival. It's locking in a major customer.

Fine. That makes sense as a business strategy.

But here's where it gets interesting. Amazon has already committed $5 billion to Anthropic in a separate arrangement, with estimates suggesting total compute spending could reach $100 billion over time. Both Google and Amazon have their own AI products. Both are now among Anthropic's largest financial backers.

Both of Claude's two biggest competitors are also its two biggest investors.

This Is Not a Mistake. It's a Hedge.

The cloud infrastructure business runs on workloads. If a frontier AI company picks one cloud provider exclusively, the others lose not just the customer, but a market signal that their infrastructure is inferior. Google cannot afford for Anthropic to run exclusively on Amazon Web Services. Amazon cannot afford for Anthropic to run exclusively on Google Cloud. Both would rather fund the company — and both are.

Anthropic is now, somewhat improbably, a neutral beneficiary of the Google-Amazon cloud war. It is taking money from both sides, using their infrastructure, and remaining technically independent. It's the Switzerland of AI companies, except Switzerland has never been valued at $350 billion.

That valuation deserves a moment. Three hundred and fifty billion dollars is more than FedEx, Delta Air Lines, and Marriott combined. Anthropic was founded in 2021 by former OpenAI researchers. It is reportedly considering an IPO as soon as October. If it goes public near that valuation, it will be among the largest tech IPOs in recent memory. All of that from a company whose cybersecurity model was breached on its first day.

But Something Nags at Me About the Google Side Specifically.

Google published the "Attention Is All You Need" paper in 2017, the transformer architecture that underpins essentially every modern AI model including Claude, GPT-4, and Gemini itself. Google built the research foundation. Google had DeepMind. Google had the data, the talent, the compute, and a significant head start.

And it is now paying $40 billion for a $350 billion company it did not build.

To be fair, that company built something ambitious. Anthropic's Mythos model represented a genuine attempt at purpose-built AI for cybersecurity, something few labs had attempted at that scale. The breach on day one was unfortunate. But the ambition was real.

There are a few ways to read Google's position. Charitable interpretation: Google is executing a sophisticated dual-track strategy, developing Gemini for consumer distribution while securing infrastructure revenue from Anthropic's workloads. The two bets hedge each other. If Gemini wins the consumer market, Google wins. If Claude dominates enterprise, Google still wins on compute fees.

Less charitable interpretation: Google looked at the benchmarks, the adoption curves, and the enterprise contracts, and decided it wasn't winning the AI model race outright. So it pivoted to securing the infrastructure relationship instead.

I'm not sure those two interpretations are actually different.

What This Means for You: Honestly, Not That Much.

Whether Claude is running on Google Cloud or AWS, your experience doesn't change. If Anthropic IPOs at $350 billion and the stock pops, it won't make Claude better at summarizing documents or worse at writing code.

But there's a pattern worth noticing. Microsoft has OpenAI. Google has Anthropic. Amazon has Anthropic. Apple is reportedly in discussions with multiple labs. The companies with the most resources to build their own AI are the same companies signing the biggest external checks to other AI companies.

You're allowed to wonder what that says about confidence in the internal bets.

Or you can just appreciate the irony. Google invented the architecture. It trained its own model. It has its own product.

And it is writing a $40 billion check to use someone else's.

---

Interested in how the AI industry actually works, not just the press releases? Subscribe to About.chat Weekly for our newsletter covering the chatbot and AI landscape every week.