Meta drops $8B on data labeling. Yes, labeling.
What happened
Meta took an $8 billion stake in a fast-growing data-labeling startup, doubling down on the least glamorous layer of the AI stack.
Why this matters
Frontier models are increasingly bottlenecked by high-quality human data, not compute. The boring infrastructure is where the leverage moved.
The slightly cynical read
When you can't out-research your rivals, you buy the supply chain they all depend on.
What to watch next
Whether OpenAI and Google quietly cut their contracts with the startup — nobody trains on data their competitor can see.