⚠ The Data Edge CommoditizesLow threat

Meta's advertising advantage rests on a marriage of proprietary data and the models to exploit it, and there is a scenario in which the second half of that marriage erodes the value of the first. If foundation models grow so capable that excellent targeting and creative can be produced from modest data, then owning the most behavioral signal on earth matters less, and rivals with good models but smaller datasets could close much of the gap.

The danger is that AI, which has so far widened Meta's lead, could in principle narrow it by lowering the data threshold for good advertising. A world where any competent model turns a little data into sharp targeting is a world where Meta's ocean of signal is less of a differentiator — where the edge shifts from who has the data to who has the model, a game more contestable by the other AI giants.

But data and models are complements, not substitutes: the same powerful model applied to Meta's vastly richer data still outperforms it applied to a rival's thinner data, so better models tend to widen Meta's lead rather than erase it. And Meta owns both a frontier model and the data, an unusual combination its pure-data or pure-model rivals lack.

Treat it as low-to-moderate and speculative. It is conceivable that advancing models commoditize targeting enough to devalue Meta's data hoard — but data and models reinforce each other, and Meta's ownership of both — it trains its own frontier Llama models atop its data¹ — makes it more likely to compound the advantage than to lose it.