Anthropic Just signed a Multibillion Dollar Deal With Google

Anthropic announced Thursday it signed a multibillion-dollar deal with Google to acquire up to 1 million of Google's Tensor Processing Units (TPUs)—the specialized AI chips that will power Claude through 2026 and beyond. The deal is "worth tens of billions of dollars" and will bring "well over a gigawatt of capacity online in 2026," enough to power roughly 350,000 homes.

This is being framed as a capacity expansion. What it actually is: Anthropic cementing its strategic dependence on Google, the company that owns 10% of Anthropic's equity and competes directly with Claude through Gemini.

This isn't a partnership. It's a controlled dependency disguised as infrastructure investment.

The Numbers: Massive Scale, Murky Economics

A deal "worth tens of billions of dollars" suggests at least $20-30 billion in committed compute spend over multiple years. For context, Anthropic's most recent valuation is $183 billion—meaning this single infrastructure commitment represents 10-16% of the company's total market value locked into Google's hardware.

One million TPUs is an extraordinary volume. Google's TPU v5 chips, the latest generation, are designed for training and inference of large language models at scale. Anthropic is essentially pre-purchasing enough compute to train multiple generations of Claude models and serve inference at massive scale through 2026.

But here's the problem: Anthropic doesn't manufacture chips. It doesn't own data centers. It doesn't control the infrastructure stack. It's entirely dependent on cloud providers—primarily Amazon (via AWS), and now Google (via TPUs). This deal shifts the balance decisively toward Google.

According to a 2024 analysis from SemiAnalysis, Google's TPUs are optimized for Google's own AI workloads and integration with Google Cloud services. They're not hardware-agnostic. Using TPUs at this scale means Anthropic's infrastructure is increasingly architected around Google's ecosystem, making migration to alternative providers (Nvidia GPUs, AWS Trainium, Microsoft Azure) progressively more expensive and technically complex.

This is lock-in by design.

Google Already Owns 10% of Anthropic—Now It Controls Infrastructure

Google invested $2 billion in Anthropic in 2023, giving it approximately 10% equity and a board observer seat. That investment made Google both a financial backer and a strategic partner—but also a competitor, since Google's Gemini directly competes with Claude in the enterprise AI market.

Now, Anthropic is committing tens of billions in compute spending to the same company that owns part of it and competes with it. That's not a neutral vendor relationship—it's structural dependency on a competitor who has financial upside if Claude succeeds and strategic upside if Claude's infrastructure costs become unsustainable.

Google wins either way. If Claude becomes the dominant enterprise AI assistant, Google profits from its equity stake and from selling compute. If Claude struggles to monetize fast enough to cover infrastructure costs, Google gains market share through Gemini while still collecting revenue from Anthropic's TPU spending.

Anthropic, meanwhile, is locked into a relationship where its primary infrastructure provider is also a major shareholder and direct competitor. That's not a balanced partnership. That's asymmetric leverage.

The Amazon Factor: Hedging or Fracturing?

Anthropic's announcement explicitly notes that its AI systems also run on chips from Nvidia and Amazon Web Services, "Anthropic's first big investor and its primary cloud provider." Amazon invested $4 billion in Anthropic in 2023-2024 and made AWS the default cloud platform for Claude.

So now Anthropic has:

• Amazon as its primary cloud provider and largest investor ($4B)
• Google as a major investor ($2B), equity holder (10%), and now primary chip supplier (tens of billions)
• Nvidia GPUs for training and inference

This looks like strategic diversification—spreading infrastructure risk across multiple providers. But it's actually fragmentation. Anthropic is now operationally dependent on three vendors who don't interoperate seamlessly, compete with each other, and have divergent strategic interests.

Amazon wants Anthropic locked into AWS. Google wants Anthropic locked into TPUs and Google Cloud. Nvidia wants Anthropic buying more GPUs. Each relationship creates switching costs that make the others more expensive to exit.

The result: Anthropic's infrastructure costs are higher, its operational complexity is greater, and its strategic flexibility is lower than if it controlled its own compute or had a single, neutral provider.

The Valuation Question: $183 Billion on What Foundation?

Anthropic's most recent funding round valued the company at $183 billion. That's higher than most publicly traded tech companies and positions Anthropic as one of the most valuable private companies globally.

But what's the underlying business? Anthropic doesn't manufacture hardware. It doesn't own proprietary data at scale. It doesn't have an advertising business, a cloud platform, or a consumer operating system. Its core asset is Claude—a language model that's technically impressive but commercially unproven at the scale required to justify a $183 billion valuation.

For comparison:

• OpenAI is valued at $157 billion with ChatGPT's 200 million weekly active users and enterprise contracts with Microsoft, Salesforce, and others
• Nvidia is valued at $3.4 trillion and owns the GPU supply chain that powers most AI training
• Anthropic is valued at $183 billion with Claude usage numbers that remain undisclosed

That valuation is a bet on future dominance, not current market position. And it's a bet that requires Anthropic to outcompete OpenAI, Google, and Meta while remaining dependent on Google for compute and Amazon for cloud infrastructure.

The multibillion-dollar TPU deal increases Anthropic's burn rate, locks it into Google's ecosystem, and creates a scenario where the company needs to grow revenue dramatically just to cover infrastructure commitments—all while competing with the same company supplying its compute.

That's not a sustainable strategic position. That's a high-wire act.

What This Signals: The Infrastructure Wars Are Over

Anthropic's TPU deal confirms what's been obvious for months: the companies that control AI infrastructure—Google, Amazon, Microsoft, Nvidia—have won the first phase of the AI race. The model builders—OpenAI, Anthropic, Mistral, Cohere—are all dependent on infrastructure providers who are also building competing models.

This creates a structural disadvantage. Google can subsidize Gemini's development with revenue from Google Cloud, Search, and YouTube. Amazon can subsidize its AI investments with AWS profits. Microsoft can leverage Azure and Office. These companies own the full stack.

Anthropic, OpenAI, and other model builders are tenants. They pay market rates (or above-market rates with long-term commitments) for compute that their competitors provide. And as their models get larger and inference costs grow, they become more dependent on those providers—not less.

The TPU deal is Anthropic accepting that reality. They can't build their own chip foundry. They can't compete with Nvidia's ecosystem. They can't afford to walk away from Google or Amazon. So they're locking in supply at massive scale, hoping that Claude's revenue growth can outpace infrastructure costs.

If that bet fails, Anthropic becomes a high-cost operation subsidizing Google and Amazon's cloud businesses while losing market share to Gemini and AWS's Bedrock models.

The Real Risk: Strategic Capture

The most dangerous aspect of this deal isn't the cost—it's the strategic capture. Once Anthropic commits tens of billions to Google's TPUs and architects Claude's infrastructure around them, migrating away becomes prohibitively expensive. Google effectively controls Anthropic's cost structure, performance ceiling, and competitive positioning.

If Google decides to prioritize Gemini over third-party customers, Anthropic has no recourse. If Google raises TPU prices, Anthropic pays or exits at massive cost. If Google integrates Gemini deeper into its ecosystem—Android, Chrome, Search, Workspace—Anthropic competes from a position of infrastructure dependency on the same company it's trying to beat.

That's not a partnership. That's vassalage with a valuation.

If your team is evaluating AI platforms and trying to understand which ones have sustainable business models versus which ones are burning infrastructure spend they can't recoup, we can help. Winsome Marketing works with growth leaders to separate viable AI strategies from valuation theater. Let's talk.