Nadella has published another long article discussing the economic reflexivity of AI. The concept of the Information Paradox describes the dilemma for information sellers: to prove the value of information, they must first disclose it; but once the information is disclosed, its value is captured for free by buyers. In the AI era, buyers who want to use good AI models instead risk handing over valuable know-how. The result is paying twice: once for tokens, and once for proprietary knowledge. Nadella directly quotes Karp's view in the article, "What technology customers want is control over their own computing power, models, data stack, and alpha. They need to be sure they own the means of production, not that it's being transferred to someone else."

Currently, it seems the new narrative for AI applications is a competition between two paths.

Camp A: The Token Economics of Model Vendors

Stronger models are better -> Higher usage is better -> Feedback for model training -> Shared weights

Form: Token consumption / Subscription-based

Camp B: Controlling AI Sovereignty / Learning Loop

Models are switchable -> Data/Logic stays within the enterprise -> The loop and data are the true AI Alpha

Form: Cloud platform / Sovereign stack / Heavy on-premise deployment

The current situation is that Camp B has been suppressed by Camp A for over half a year and is now starting to counterattack, vying for the right to define the narrative. Who will win?

$Microsoft(MSFT.US)$Palantir Tech(PLTR.US)