AI’s own cheerleaders just hit the brakes

The vibe changed fast

When the people building the AI gold rush start talking like park rangers, the market pays attention. Anthropic CEO Dario Amodei called for the industry to “pace the frontier,” warned rogue AI could create serious damage in months, and said outside evaluators will get deeper access to systems. OpenAI, DeepMind, and even Elon Musk backed the general idea of more testing.

Chips got the memo

Wall Street did what Wall Street does: it turned a safety essay into a whole-sector haircut. Futures weakened, and the usual AI hardware suspects got hit first — from Applied Materials and Lam Research to Marvell, Micron, AMD, Intel, Vertiv, and NVIDIA. The logic is pretty simple: if model growth slows, the buildout for training chips, memory, networking gear, and data-center power may not need to sprint quite as hard.

The other trade woke up

On the flip side, cybersecurity stocks caught a bid. CrowdStrike, Palo Alto Networks, SailPoint, and ServiceNow all moved higher as investors re-priced the world where AI agents aren’t just clever assistants — they’re also potential headaches. That’s not exactly a fun thesis, but it is a real budget line.

Big picture

This doesn’t mean the AI capex party is over. It does mean the trade may be broadening: less pure “buy every chip with an AI sticker,” more “show me the picks-and-shovels that keep the robots from going feral.”