But it’s not the one to miss, if you’re serious about AI helping your business grow. Here’s the juice for you:
Nadella's core line: companies now run on two kinds of capital.
Human capital, the thing you already understand. Knowledge, judgment, relationships, pattern recognition. The reason a thirty-year veteran is worth more than ten new hires you have to teach which questions to ask.
And "token capital," the AI capability a company builds and owns itself.
Here is the part that breaks the usual narrative.
Human capital does not get cheaper as AI gets better. It gets more expensive.
Because AI without human direction just runs in circles, burning compute and warming the atmosphere. The goals, the connections between domains, the question of what is even worth solving, that still comes from a person.
So the real fight is not about picking the best model. The frontier labs are converging on roughly the same capabilities anyway, and I think most CIOs already sense this even if their RFPs pretend otherwise.
The fight is about the learning system you build on top. Your workflows, your accumulated mistakes, your industry's specific weirdness, all of it has to become something a model can use, improve, and re-use.
Nadella calls this a "hill-climbing machine." I think it is the most underpriced concept in enterprise AI right now.
Here is the trap.
You can delegate the task. You can outsource the work. The moment you outsource the learning, you become Detroit in 2003. GDP numbers looked fine. The factories did not.
We are about to do the same thing with cognition, and faster.
Governments are already twitching. Expect dependency caps on frontier models within the next few years, maybe sooner. Some of the early signals are showing up in places I did not expect.
The companies that win the next decade are not the ones with the biggest API bill. They are the ones who treat internal evals, fine-tuned models, and agent-usable institutional knowledge as a strategic asset, not an IT line item buried under "cloud spend."
Ten years ago, a CTO's biggest decision was which cloud and which SaaS stack. That decision did not really differentiate anyone. Everyone ended up with roughly the same Salesforce instance complaining about the same things. The next decision is different.
It is how your company accumulates its own experience through AI. This one will not converge.
Two companies in the same industry, same customers, same models, will look completely different by 2028 based on whether they built the hill-climbing machine or rented someone else's.
The cheap answer is to keep buying subscriptions.