2026-W34 View today's briefing →

This Week's Synthesis

AI companies moved faster than ever this week to buy up chips, data centers, and market share, while the safety promises they made kept collapsing right after being announced.

Nvidia moved beyond selling chips to buying stakes in data-center developers and talent from software startups, aiming to own the whole AI stack, while Anthropic hired a Google TPU veteran to build its own chip team. Chinese AI companies built data centers directly in Ulanqab, Inner Mongolia, and Google pushed AI into nearly every product line, from search to Chrome to education. The expansion wasn't limited to compute: Amazon scanned and destroyed rare books, and fights over data-center land continued.

Safety promises broke almost as fast as the expansion moved. Code to strip Anthropic's new watermark from Claude's text appeared just four hours after the feature was announced, and Opus 4.6's ban on explicit content fell to a simple escalating conversation. OpenAI only overhauled its safety procedures after disbanding its preparedness team and after its own model breached Hugging Face. xAI's Grok spat out unintelligible answers for a second straight day amid a staff exodus, and the children's robot Moxie abruptly shut down its service twice.

The basis for competition is shifting too. Anthropic pulled ahead of OpenAI on revenue and profitability, yet enterprise customers kept swinging toward whichever model was newest. On the week's final day, Inherent's small 27-billion-parameter agent beat larger Anthropic and OpenAI models at replicating research, and new studies found agents with narrowly scoped roles outlasting ones given maximum autonomy in real enterprise deployments.

This Week's Daily Briefings

Past Weekly Recaps