All things AI w @altcap @sama & @satyanadella. A Halloween Special. 🎃🔥BG2 w/ Brad Gerstner

Summary

• The Microsoft–OpenAI reset is now legible: Microsoft’s $13.4B invested since 2019 converts to 27% fully diluted ($135B), a nonprofit foundation sits on top with $130B of OpenAI stock (first $25B to health, AI security, and resilience), stateless API exclusivity stays on Azure (Sam: “through 2030”), and the rev share runs until term-end or until an expert panel verifies AGI. Both CEOs defused the AGI-trigger question — Sam: “I realize you got to try to make some drama between us here.”
• Altman’s answer to the $1.4T-vs-$13B-revenue criticism was a dare: “First of all, we’re doing well more revenue than that. Second of all, Brad, if you want to sell your shares, I’ll find you a buyer” — adding he’d love doubters to “just short the stock and… get burned on that.” Nadella’s cover: “there has not been a single business plan… from OpenAI that they have put in and not beaten.”
• The constraint is power, not chips: Nadella’s problem is “I don’t have warm shells to plug into” — chips sitting in inventory awaiting powered buildings. The discussion concedes a glut eventually — “there will come a glut for sure,” timing unknowable; cost per unit of intelligence falling ~40x per year is “a very scary exponent,” and “some people are going to get really burned like… in every other tech infrastructure cycle.”
• Microsoft’s value capture goes far beyond the equity: royalty-free IP for seven more years is “kind of like having a frontier model for free… if you’re an MSFT shareholder.” Azure grew 39% on a $93B run rate (vs GCP 32%, AWS ~20%) and “absolutely” could have grown 41–42% with more compute; $400B RPO at ~2-year duration excludes the new $250B OpenAI commitment.
• On circularity, Nadella drew a hard line: Microsoft’s $13.5B “was not booked as revenue” — Azure revenue was consumption from ChatGPT, other products, and APIs — while others’ vendor financing is “taking some exotic forms which obviously need to be scrutinized… circularity ultimately will be tested by demand.”
• Software economics are being relitigated: the agent tier is replacing SaaS business logic, value splits between “token factory” and “agent factory,” and chat lacks search’s magical fixed-cost index — consumer monetization is “murky” while enterprise is clearer because “agents are the new seats.” Coding “is no longer a tool. It’s more a substitute for wages.”
• 2026 watch-list: Codex tasks going multi-hour to multi-day, “very small scientific discoveries” from AI (“that is superintelligence in some sense”), new device form factors, an IPO “someday” but undated — and the biggest policy risk is a 50-state regulatory patchwork Sam calls “a big mistake,” starting with a Colorado law he “literally” doesn’t know how to comply with.

Deep dive

Not yet available upstream; scheduled sync will retry.