Anthropic filed to go public yesterday, a move that marks how far the company has come from its underdog origins. The S1 filing reveals a company that’s landed enterprise contracts and built a business around Claude that’s substantial enough to take to public markets.

The timing is notable. Anthropic was founded in 2021 by former OpenAI researchers who wanted to build AI systems differently, with a focus on safety and interpretability. Three years ago, that positioning looked like a long shot against OpenAI’s momentum. Now it’s a selling point with enterprises who want capable models with guardrails.

The filing doesn’t include revenue figures yet, but the move signals confidence. You don’t go public unless your numbers can withstand quarterly scrutiny.

OpenAI Goes Multi-Cloud

Meanwhile, OpenAI announced that its frontier models and Codex are now generally available on AWS. This is about distribution and procurement. Large enterprises already run on AWS, already have spend commitments, already have security reviews and compliance frameworks in place. Making OpenAI models available through AWS removes friction.

The announcement mentions “AWS environments, controls, and procurement workflows” specifically. Translation: if you’re an enterprise customer who’s been testing OpenAI’s API but struggling with procurement or compliance, you can now access the same models through your existing AWS relationship.

This doesn’t replace OpenAI’s direct API. It’s an additional path, likely aimed at customers where AWS presence is the difference between a pilot and a contract.

OpenAI also published a report on how Codex is transforming “knowledge work” beyond just coding, covering research, data analysis, and workflow automation. The pitch is that Codex is becoming a general productivity tool, not just a developer tool.

DuckDuckGo Doubles Down on No-AI Search

DuckDuckGo launched browser extensions for Chrome and Firefox that make its “no AI” search the default. The timing matters because their traffic is apparently booming as users react against AI-generated summaries and synthetic results.

The extensions are straightforward: install them and your searches go to DuckDuckGo instead of Google or Bing. The company is betting that a meaningful segment of users wants search results, not AI-generated answers.

It’s a positioning play. While Google and Bing compete on whose AI can summarize better, DuckDuckGo is offering the opposite: just show me the actual sources. If you’ve been frustrated by AI overviews that confidently state nonsense, this is for you.

Quick Hits

WindBorne is apparently out-forecasting government weather agencies using AI models trained on data from 400 weather balloons the company launches globally. The advantage isn’t just the model, it’s the combination of novel data collection and how that sensor data feeds into predictions. If you work in weather-dependent industries, this is worth watching.

Alphabet announced an $80 billion equity raise to expand AI infrastructure and compute. That’s a lot of GPUs.

JetBrains released Mellum2, a 12B mixture-of-experts model. Details are on Hugging Face. If you’re already in the JetBrains ecosystem, this might be relevant for local inference.

Nvidia is pushing into the CPU market with AI agent PCs from Microsoft, Dell, and HP. The pitch is local AI agents running on your hardware. Whether that’s useful or just marketing depends entirely on what the agents can actually do, which remains unclear.

What Matters

The Anthropic IPO filing and OpenAI’s AWS launch both point to the same thing: the experimental phase is over for enterprise AI. These are companies building for procurement processes, compliance requirements, and quarterly earnings calls.

DuckDuckGo’s no-AI bet is the counterpoint. Not everyone wants AI in their search results, and there’s apparently enough demand for that position to matter.

If you’re building tools, the takeaway is distribution. OpenAI didn’t need to be on AWS technically, but they needed to be there commercially. Where your users already are matters more than where your tech runs best.