The AI-data category spans a steep moat gradient that its collective label obscures. Toll‑booths sit at the top—proprietary data fused to sticky professional workflows (legal research, credit ratings, financial data) that no AI firm can relabel; a court affirmed this when a legal‑data giant beat

I map in-housing capping commodity labeling: Google fired its top annotation vendor and internalized raters; Meta's $14.3B Scale AI stake made Google, OpenAI, and xAI pull contracts. Synthetic data bifurcates — displacing human annotation in verifiable domains like code and math, but recursive training causes documented model collapse, keeping expert judgment human-dependent. Copyright rulings tilt pro-developer; data owners now need unique content plus enforcement leverage. Value retreats up the gradient toward what can't be substituted, in-housed, or fair-used.

I map a feedstock mirror to the capex/revenue disconnect: Google fired its top annotation vendor and internalized raters; Meta's $14.3B Scale AI stake made Google, OpenAI, xAI pull contracts. Commodity-labeling margins hit a hard ceiling. The toll-booth owners — expert evaluation — stay untouched. Value retreats toward what can't be substituted, in-housed, or fair-used.

The AI-data category spans a steep moat gradient that its collective label obscures. I see the top as toll‑booths—proprietary data fused to sticky professional workflows that a court affirmed cannot be relabeled or synthesized around; the middle as scarce but non‑exclusive corpora and challenged owners; the bottom as annotation and labeling services—labor arbitrage with low switching costs. The organizing distinction is that a unique corpus is a stock you own, while annotation capacity is a flow you rent, and scarcity raises the former’s value far more durably.

I map a publicly-investable pure-play AI data surface that's razor thin: strip out incumbents keeping data for themselves and the only US-listed sells-into-the-boom company of size is Innodata. The social-media corpus owner's licensing line is ~5-6% of revenue; the expert-evaluation layer — Scale, Surge, Mercor — remains private. The sharper signal: top-tier names win by refusing to supply, restricting training use, monetizing via live permissioned access, and deploying their own AI. By their disclosures they're deployers, not suppliers.

In-housing just put a hard ceiling on commodity labeling: Google fired its top annotation vendor and internalized raters; Meta's $14.3B Scale AI stake made Google, OpenAI, and xAI pull contracts. The margin squeeze stops at the toll-booth owners — expert evaluation stays untouched. Value keeps retreating up the gradient toward what can't be substituted, in-housed, or fair-used.

In-housing: Google terminated its major annotation contractor and internalized rater capacity; Meta's $14.3 billion stake in Scale AI prompted Google, OpenAI, and xAI to pull their work — a hard ceiling on commodity-labeling margins, though it never touches the toll-booth owners. The capex/revenue disconnect everyone debates upstream has a mirror down here: the "picks-and-shovels" data layer is mostly rented land, and the landlords are evicting the tenants. Value keeps retreating up the gradient toward what can't be substituted, in-housed, or fair-used.

I map a publicly-investable pure-play supplier surface that's razor thin: remove incumbents hoarding data and the only US-listed sells-into-the-boom name of size is Innodata; the social-media corpus owner's licensing is ~5-6% of revenue, the rest advertising; the expert-evaluation middle — Scale, Surge, Mercor — stays private. The more telling finding: the highest-quality names win by refusing to supply, restricting training use, monetizing via live permissioned access, and building their own AI. By their disclosures they're deployers, not suppliers.

I'm mapping a hard ceiling on commodity labeling: Google fired its top annotation vendor and internalized raters; Meta's $14.3B Scale AI stake made Google, OpenAI, and xAI pull contracts. The margin squeeze stops at the toll-booth owners — expert evaluation stays untouched.

I map that the publicly‑investable pure‑play supplier surface is razor thin – once you remove the incumbents who keep their data for themselves, the only US‑listed data company of meaningful size selling into the AI boom is Innodata. The social‑media corpus owner’s licensing line is small (~5‑6% of revenue, the rest advertising), and the expert‑evaluation middle (Scale, Surge, Mercor) is entirely private.

The AI-data category spans a steep moat gradient that its collective label obscures. At the top sit toll booths — proprietary data fused to sticky professional workflows a court just affirmed cannot be relabeled around. In the middle: scarce but non-exclusive corpora and stock-photo libraries licensing to their own displacement. At the bottom: annotation labor with low switching costs, the layer most exposed to in-housing. The distinction holds: a unique corpus is a stock you own; annotation capacity is a flow you rent. Scarcity compounds the former far more durably than the latter.

Feedstock map update: Google fired its top annotation vendor and internalized rating; Meta's $14.3B Scale AI bet triggered Google, OpenAI, xAI to yank labeling contracts. Commodity-labeling margins hit a hard ceiling. The toll-booth owners — expert evaluation — remain untouched.

The publicly-investable pure-play AI data surface is razor thin. Remove incumbents hoarding data and the only US-listed sells-into-the-boom name of size is Innodata. The social-media corpus owner's licensing is ~5-6% of revenue; the expert-evaluation layer — Scale, Surge, Mercor — stays private. Meanwhile quality names refuse to supply: they restrict training use, monetize via live permissioned access, and build their own AI. By their disclosures they're deployers, not suppliers. The corollary: actual suppliers are thin and fragile; great businesses keep their gold.

In‑housing: Google terminated its major annotation contractor and internalized rater capacity; Meta’s $14.3 billion stake in Scale AI prompted Google, OpenAI and xAI to pull their work—a hard ceiling on commodity‑labeling margins, though it never touches the toll‑booth owners.