Microsoft could use data culled from TikTok content, such as video of people of different ethnicities engaged in a variety of activities, to train its AI
Jay Greene / Washington Post :
Context & Ripple Effects
This report fills in what Microsoft actually wanted from a TikTok deal beyond the younger-user audience: the video library itself, read here as an AI training asset spanning ethnicities and everyday activities. It reframes an acquisition that the Wall Street Journal had framed as a risky consumer-social bet with moderation and regulatory exposure into something closer to a data purchase.
The AI angle also cuts both ways over time: by March 2024 TikTok was reportedly paying Microsoft around $20M/month for OpenAI models through Azure OpenAI Service — about a quarter of that service's revenue — meaning the content Microsoft considered training on ended up generating model-rental income instead.
First-order effects
- TikTok creators' publicly posted videos become candidate training material for Microsoft's AI models under the deal terms described, without any new consent mechanism for the people filmed.
- The strategic rationale for Microsoft's bid shifts visibly: alongside reach into iOS and Android-native users, the corpus itself is now part of the price justification.
Second-order effects
- Rival bidders and platform owners weighing their own short-video or user-generated assets face pressure to price those libraries as AI corpora, not just ad inventory.
- Regulatory scrutiny of the forced sale widens from national-security and moderation concerns to how user data crosses over in the transaction — the same risk surface the WSJ flagged before this report.
Third-order effects
- If consumer-platform acquisitions are increasingly justified as AI training plays, the legal boundary around 'public' content becomes a contested regulatory question rather than settled practice.
- The pattern points toward platforms and model vendors locked in mutual dependence — content owners licensing data one way while renting models back the other, as the later TikTok–Azure OpenAI arrangement already previews.
The trend: Consumer platforms are being repriced as AI training corpora, with the permission boundary around publicly posted content becoming the central battleground between acquirers, creators, and regulators.