Sora died. Atlassian cut 1,600 engineers. Anthropic got blacklisted. The thread that connects them runs through your org.

March 2026 was one of the densest months in AI history. Three frontier models shipped in a single month — GPT-5.4, Gemini 3.1 Ultra, and Grok 4.20. GTC happened. The SaaSpocalypse continued. A major AI product was killed. A frontier lab got blacklisted by the government. The takes were written. The benchmarks were compared. You’ve already read those takes.

I don’t want to repeat them. I want to show you what happened underneath.

While everyone was comparing benchmarks, OpenAI’s flagship video product was burning $15 million a day in inference costs against $2.1 million in total lifetime revenue. The first real ad dollar entered a ChatGPT conversation and converted at 1.5x the rate of search. Atlassian had just reported cloud revenue up 26% — and its stock was still down 84% from its peak — because Wall Street wasn’t pricing the present, it was pricing a future where AI agents don’t need software seats. And a frontier AI lab got blacklisted by the U.S. government for refusing to build autonomous weapons, then watched its app go to number one on the App Store the same week.

None of those stories are about which model won a benchmark. All of them are about something harder to see and more consequential: the AI industry is leaving the capability phase and entering the economics phase. The question is no longer what can we build. It’s what can we sustain.

Here’s what’s inside:

• Inference became the kill metric. Sora died because serving it cost more than anyone would pay — and the physics of why apply to every AI product you’re building on.

• The first real ad dollar entered the AI interface. Criteo’s ChatGPT integration produced 1.5x conversion rates and the first credible threat to Google’s $300 billion ad model.

• The physical path for AI is closing. The White House cleared the regulatory lane, but 12 states, 50+ local governments, and Iranian drones are blocking where the infrastructure actually gets built.

• The SaaS business model broke. Per-seat pricing, the most durable model in enterprise tech for twenty years, is being repriced in real time — and the stocks are pricing it in before the seat counts even decline.

• Safety posture became a market position. The Anthropic-Pentagon standoff sorted AI labs into categories that governments and enterprise buyers are now using to make real procurement decisions.

• The Open Brain skill that does this for you. Weekly News Analysis skill tracks 30 companies across 10 categories, re-ranks using what your Open Brain already knows about you, and produces a personalized structural diff every week — plus 6 prompts that work in any AI client. For paid subscribers.

Each shift operates at a different altitude — physics, monetization, geography, business models, geopolitics — but they all feed the same underlying story. Let me show you what I found.