Amazon Just Put $25 Billion on Anthropic. This Isn’t About Claude.

$25 Billion on Anthropic.

4:25

Let’s be precise about what happened here, because the headline undersells it.

Amazon announced it will invest up to $25 billion in Anthropic — $5 billion immediately, with up to $20 billion more tied to commercial milestones. That brings Amazon’s total potential stake to $33 billion. In exchange, Anthropic commits to spending over $100 billion on AWS infrastructure over the next decade, including 5 gigawatts of compute powered by Amazon’s Trainium chips.

Read that again. One hundred billion dollars in cloud commitment. In return for twenty-five billion in equity. This is not a bet on a chatbot. This is Amazon buying a moat.

What Amazon Is Actually Buying

Microsoft locked in OpenAI years ago. That partnership — Azure as the exclusive cloud, GPT-4 baked into every Microsoft product, Copilot everywhere — became the defining enterprise AI distribution play of the last three years. Microsoft didn’t just invest in OpenAI. It became the distribution layer for OpenAI. Azure revenue grew on the back of it.

Amazon watched that happen. And now it’s doing the same thing with Anthropic.

The $25 billion isn’t charity toward Dario Amodei’s safety-first research lab. It’s AWS protecting its cloud revenue from the Azure flywheel. Claude running on AWS, Anthropic spending $100 billion on Trainium compute, enterprise customers choosing AWS because that’s where their Claude integration lives — that’s the actual product being purchased here. This is infrastructure economics dressed up as an AI investment.

The Numbers Are Genuinely Staggering

Anthropic’s revenue trajectory is one of the most aggressive in the history of enterprise software:

• Start of 2024: $100M ARR

• Start of 2025: $1B ARR

• End of 2025: $9B ARR

• February 2026: $19B ARR April 2026: $30B ARR

That’s 300x growth in 27 months. The company hit its full-year 2026 revenue target by the end of Q1. It did this with roughly 5,000 employees — a ratio of revenue-per-employee that almost nothing in the history of software has matched.

The current implied valuation sits somewhere between $380 billion (Series G post-money, February 2026) and $800 billion depending on which secondary market you’re reading. Polymarket is pricing a 52–54% chance of an IPO before 2027. That’s a coin flip — which tells you the market genuinely doesn’t know if Dario takes the $25 billion and stays private, or rings the bell.

My read: with $25 billion just walked in the door, the pressure to IPO decreases in the short term. Why dilute now?

The OpenAI Comparison Nobody Is Saying Out Loud

OpenAI’s revenue model is gross revenue — the headline number. Anthropic’s is net, after infrastructure costs. The comparison between the two is never apples-to-apples, and the AI media rarely flags this.

What we do know: both companies are on a trajectory toward numbers that would have been considered science fiction two years ago. The difference is that Anthropic now has $100 billion in committed infrastructure spending locked to one cloud provider, which changes the cost structure, the margin profile, and the strategic dependency in ways that will take years to fully understand. It’s not obviously good for Anthropic’s independence. It’s obviously good for AWS.

The Wildcard: Cerebras at 92%

The Polymarket IPO probability table is worth a second look. SpaceX at 94% is basically guaranteed. But Cerebras at 92% is the one I’d be watching. AI chip company. Wafer-scale architecture. One of the few credible inference-compute challengers to NVIDIA in specific workloads. If Cerebras goes first, it sets the pricing ceiling for everything that follows — including Anthropic — and gives the market a reference point for what an AI infrastructure company is actually worth when the rubber meets the road.

The Bigger Picture

What Amazon just did is declare, with $25 billion of clarity, that the AI infrastructure layer is the most valuable real estate in technology right now. Not the models. The pipes the models run on. The race isn’t who builds the smartest AI. It’s who owns the terrain the AI runs across. Amazon just bought a very large piece of it.