Scale AI — AI supply-chain exposure

Scale AI occupies an unglamorous but critical spot: it supplies the labeled training data, human feedback and evaluation harnesses that frontier labs and governments use to build and grade models. It does not train the models itself — it produces the raw material and the measuring instruments that make training work. As models scale, the bottleneck shifts from raw compute toward high-quality, well-curated data, which is exactly what Scale sells.

Its structural hook is being a shared supplier to competing labs — a picks-and-shovels position that benefits from the race without having to win it. Meta's large minority stake underscores how strategic the data-and-evaluation layer has become. The exposure is customer concentration and the risk that labs bring data operations in-house over time. The model places it mid-chain as an enabling applications-layer supplier: essential to model quality, but dependent on the labs it serves.