Anthropic just tied itself to AWS for a decade

THE AI BRIEF

Today's operating signal: the compute layer is becoming a lock-in tier of its own. Plus four shorter reads on Apple, Chrome, NSA, and a new Anthropic Labs product.

THE READ

On Monday, Anthropic took another $5 billion from Amazon and agreed to spend $100 billion on AWS compute over the next ten years. That's up to 5 gigawatts of new Trainium capacity for Claude, against run-rate revenue reportedly past $30 billion.

If you run Claude in production, nothing about your Tuesday changes. The API is the same. Claude's quality, rate limits, and invoice all stay where they are. The near-term upside is real though: more compute means fewer capacity errors and more headroom for the next model, which is mostly good for anyone already in.

What does change is what you have to watch. Your model provider's cloud choice was something most operators could safely ignore last week. It isn't anymore. When a lab ties itself to one hyperscaler for ten years, your future options for where Claude runs get narrower. That matters if you're in a regulated industry with BAAs on a specific cloud, or if your long-term AI strategy assumed optionality across Azure, GCP, and AWS.

The one move worth making this week: on your next vendor review, add a single line asking your AI provider which cloud, which region, and which chip family the model runs on. You don't need to act on the answer yet. You just need it in a document somewhere, because a year from now you'll want to know when you wrote it down.

FREE WORKSHOP

I'm running a free 30-min Lightning Lesson on Claude Design tomorrow. RSVP here:

ALSO WORTH KNOWING

Tim Cook steps up, John Ternus takes Apple CEO
Planned transition, effective later this year. Ternus ran iPhone and iPad silicon programs. Apple's AI posture for the next decade is now his call, which matters for anyone building on the platform or selling into it.

OpenAI dropped GPT Image 2 at a live event today
Announced mid-day Tuesday on livestream, positioned as the best image generation model on the market. If your marketing or design team is currently standardized on Midjourney or Flux, expect a rebenchmarking cycle this week.

Perplexity Computer launches as a mainstream agent product
Computer-use AI moving out of developer previews and into consumer-tier packaging. If your team has been waiting for browser-agent reliability to cross a usability bar, Perplexity's launch is the right moment to run a pilot.

DeepSeek is raising $300M at a $10B valuation, its first outside round
Reports surfaced Tuesday. The core V3 team has left for Xiaomi and ByteDance, and V4 is delayed for Huawei chips. A $10B brand value during mid-transition is the kind of number that changes what "Chinese AI lab" means to your procurement team.

Sam Altman: OpenAI is going all-in on robotics
Altman framed the move as a national-security bet, alongside the news that a humanoid robot just broke the human half-marathon record in Beijing. Worth reading as the clearest signal yet that OpenAI's next lock-in fight is physical, not digital.

WATCHING TOMORROW

GPT 5.5 "Spud" is expected to drop on Wednesday or Thursday per OpenAI's own telegraphing. If it lands tomorrow, the Amazon-Anthropic take at the top of this issue gets an early stress test against an OpenAI counter-move. I'll have the operator read in Thursday's issue.

Back tomorrow,
Haroon