Nadella: Enterprise AI runtimes must solve memory, entitlements, and action space
Satya Nadella · Satya Nadella describes how lessons from Microsoft’s history apply to today’s boom · Nov 18, 2025 · at 7:34
Microsoft CEO Satya Nadella outlines the core architectural components required outside foundational models to make AI functional in enterprise environments.
0:00 / 0:55exact quote · 55.5s
▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark
“There, at least there are three things to me that are outside of the model at runtime that I think you kind of have to crack, right? One is memory, and all forms of memory, right? Short, long right? Even like this, these big challenges of, you know, humans are great at long-term credit assignment, right? Which is how does intuitively, like somebody said to me, hey, the day, you know, AI models can both reward and remember that you know, how to punish for some mistake, because they have the ability to do long-term credit assignment, that's when you'll know they have real memory, but in any case, memory is one. The second one is entitlements, right? So, which is, they have to really respect all of the permissioning system at runtime, right? Because this is where, because there are roles, what access do I have? And so the model needs to meet that, and then the action space all has to work.”
quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →
More from Satya Nadella
Nadella: There is no open web, only the Google web
“Because the organizing layer of the web you know, I always say, like, there's no such thing as the open web. There's the Google web. And just because they just dominated it.”
Nadella: Microsoft has zero AI utilization issues; capacity is sold out
“I don't have a utilization problem. I may have a PUE, like, I mean, I want higher utilization because it's mostly because it's memory bottleneck or what have you, but that is a different, I mean, like, there is not a thing that I have that's not sold out. In f…”
Nadella: Neural networks and compute solve enterprise data, not complex schemas
“In fact, one of the longest time we used to always obsess about, oh, how complex do the relationships have to be, or the data model needs to be to capture the essence of an enterprise, right? And it turns out, It's lots of parameters in the neural network with…”
Nadella: Separating cognitive core from knowledge enables continual learning
“If you kind of have the model's cognitive core separated from its knowledge, then you have essentially the continual learning, you know, formula, so to speak, or the algorithm, and then you just unleash it.”
Nadella: AI agents will create their own RL environments
“Right now, for the RL loop, they are simulating our production environment, but they will ultimately be more efficient in creating their own production environments to kind of RL themselves.”
Nadella: Next IDE will be born in autonomous PR triage
“One of the things that we are experimenting with this mission control in GitHub Copilot is that, right, which is, well, the idea is you go have five, six different branches in which you fire off all these autonomous agents, they all do their work, they come ba…”