Public transcript

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

By: Bg2 Pod
Length: 1 h 14 min
Language: English
Text from: the video's captions
Transcribed: 12 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

This Halloween episode of BG2 features Brad Gerstner in conversation with Sam Altman and Satya Nadella, recorded the week of the OpenAI–Microsoft partnership restructuring announcement. They break down the deal terms, the new $130 billion nonprofit, compute constraints, AGI definitions, Azure growth, and the future of software, agents, and jobs. Both speakers express optimism about AI-driven productivity gains and the reindustrialization of America, while acknowledging risks of overbuilding and regulatory patchwork.

Key points

Microsoft has invested roughly $134 billion in OpenAI and holds about 27% ownership on a fully diluted basis, with Satya Nadella calling it a potential '100 bagger.'
The restructured OpenAI creates a nonprofit capitalized with $130 billion of OpenAI stock, with the first $25 billion directed to health and AI security and resilience.
OpenAI's frontier models remain exclusive to Azure as 'stateless APIs' through 2030, while other products can be distributed on other platforms; a revenue share runs until 2032 or until AGI is verified.
Sam Altman says OpenAI is doing well more than the reported $13 billion in revenue and defends the massive compute commitments, saying the company is held back by lack of compute.
Satya Nadella says Microsoft's current constraint is not chips but power and 'warm shells' to plug them into, and Azure could have grown faster than 39% with more compute.
Both Altman and Nadella warn against a 50-state patchwork of AI regulation, with Altman saying he doesn't know how to comply with the Colorado AI act.
Nadella frames Microsoft's value as running an efficient 'token factory' and an 'agent factory,' noting agents are the new seats in enterprise monetization.
Altman hopes AI makes a novel scientific discovery in 2026 and expects coding agents to move from multi-hour to multi-day tasks.
Nadella says Microsoft will grow headcount, but with far more leverage per employee than pre-AI, citing an internal example of agents automating DevOps pipelines.
Microsoft's $400 billion in remaining performance obligations has an average two-year duration and is diversified across first-party and third-party demand.

Questions it answers

02:34The deal terms

What are the key terms of the restructured OpenAI–Microsoft partnership?

Microsoft has invested about $134 billion for roughly 27% ownership. OpenAI's frontier models remain exclusive to Azure as stateless APIs through 2030, a revenue share runs until 2032 or AGI verification, and a nonprofit capitalized with $130 billion of OpenAI stock sits on top of a public benefit corp.

First $25 billion of nonprofit funds go to health and AI security/resilience.
Satya Nadella calls it one of the great tech partnerships and a potential 100 bagger.

09:45AGI and the expert panel

How will AGI be determined and does it matter for the deal?

If OpenAI claims AGI, an expert panel selected by the parties makes a relatively quick decision. Both speakers downplay drama: Altman expects surprising twists, and Nadella says nobody is close to AGI, but the process is a good one.

Exclusivity and rev share both end early if AGI is verified.
Altman: even with superintelligence tomorrow, they'd still want Microsoft's help distributing.

11:45Compute commitments

How can OpenAI afford $1.4 trillion in compute commitments on ~$13 billion of revenue?

Altman says revenue is well above the reported figure and growing steeply, and that the company is compute-constrained — with 10x more compute, revenue would be far higher. Nadella adds OpenAI has beaten every business plan it has put forward.

Altman offers to find buyers for anyone worried about OpenAI's finances.
The bet is that ChatGPT, an AI cloud, consumer devices, and AI-driven science will generate the revenue.

18:00Glut or shortage?

Will there be a compute glut in the next few years?

Nadella says the real problem today is power and warm shells, not chip supply. Both agree a glut will come eventually but not in 2–3 years, and Altman warns that breakthroughs in cost per unit of intelligence could burn people holding long-term contracts.

Nadella: chips sitting in inventory he can't plug in.
Altman: if everyone can run a personal AGI on a laptop, some centralized builders will get burned.

37:40Azure growth and RPO

How fast is Azure growing and how solid is Microsoft's backlog?

Azure grew 39% on a $93 billion run rate and could have grown faster with more compute. Microsoft has $400 billion in remaining performance obligations with about a two-year average duration, diversified across first-party and third-party demand.

Nadella says Microsoft sometimes says no to demand, including OpenAI's, to keep the fleet fungible.
He notes investors are overly fixated on the Azure number versus the high-margin Copilot businesses.

46:00SaaS and agent economics

Does AI disrupt traditional SaaS and where does value accrue?

Nadella says the agent tier replaces the old business logic tier, and that Microsoft's low-ARPU, high-usage products are well positioned. Value splits between the token factory (efficient compute) and the agent factory (applications that use tokens smartly), and agents are the new seats in enterprise.

GitHub Copilot's auto mode picks the right model per task using eval and feedback loops.
There is now a real marginal cost to software, so business models must adjust.

1:04:00AI and jobs

Will AI be a net job creator at Microsoft?

Nadella says headcount will grow but with far more leverage per employee, citing a network operations lead who built agents to automate DevOps pipelines instead of hiring. Altman (earlier) and Gerstner frame it as companies growing revenue faster than headcount, which aggregates into economy-wide productivity gains.

Nadella: everyone at Microsoft gets M365 and GitHub Copilot as standard issue.
Learning how to work with agents is like relearning workflows when Excel first arrived.

Notes

The Deal

Microsoft invested roughly $134 billion in OpenAI since 2019 and holds about 27% on a fully diluted basis.
The restructuring creates a nonprofit capitalized with $130 billion of OpenAI stock, which Satya Nadella calls one of the largest nonprofits ever, alongside the Gates Foundation.
The first $25 billion of nonprofit funds go to health and AI security and resilience.
OpenAI's frontier models stay exclusive to Azure as 'stateless APIs' through 2030; a revenue share runs until 2032 or until AGI is verified by an expert panel.

Compute and Demand

Sam Altman pushes back on the reported $13 billion revenue figure, saying revenue is growing steeply and defending the ~$1.4 trillion compute commitments.
Altman compares compute to energy: demand depends on price per unit of intelligence; if cost fell 100x, usage would rise far more.
Nadella says Microsoft's real constraint is power and 'warm shells,' not chips — he has chips sitting in inventory he can't plug in.
Both expect a compute glut eventually, but not in the next 2–3 years; Altman warns some people will get burned on existing contracts if cheap energy arrives.

Azure and Microsoft's Position

Azure grew 39% on a $93 billion run rate; Nadella says it could have grown faster with more compute.
Microsoft has $400 billion in remaining performance obligations with roughly a two-year average duration, diversified across first- and third-party demand.
Nadella emphasizes the 'token factory' (efficient compute) and 'agent factory' (SaaS that knows how to use tokens efficiently); agents are the new seats.
Microsoft has royalty-free access to OpenAI IP for seven more years, which Nadella likens to having a frontier model for free.

Regulation

Altman says he literally doesn't know how OpenAI is supposed to comply with the Colorado AI act and calls a 50-state patchwork a big mistake.
Nadella argues federal preemption is critical, otherwise startups get crushed and the EU will set the framework instead.

Looking Ahead

Altman expects coding agents to move from multi-hour to multi-day tasks in 2026 and hopes for a small novel scientific discovery that year.
Nadella says headcount will grow but with far more leverage per employee, citing agents automating DevOps pipelines for fiber operations.
Both are optimistic about reindustrialization of America, with data centers, fabs, and power buildouts driving investment.

Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.