Gatekeepers in AI: Data and Infrastructure Are the Moats
The EU's High-Level Group on AI says the Digital Markets Act can hit the two levers that matter most in AI competition: access to infrastructure and access to exclusive data. Days before the analysis became public, the European Commission opened a formal case into Google's use of publisher and YouTube content for AI Overviews and AI Mode.
If you build AI products or own distribution inside a platform, this is your signal: incentives are shifting. Integration that felt "smart" yesterday may look like foreclosure tomorrow.
Why this matters for engineers and product teams
- AI infra and distribution are consolidating under a few platforms, making interoperability and neutrality a real constraint, not a nice-to-have.
- First-party user interaction data is the scarce input. Who collects it, who can reuse it, and who can't-that's the competitive edge regulators will scrutinize.
- Search UX changes (AI Overviews/Mode, Preferred Sources) will move traffic and monetization. Your telemetry and contracts need to catch up.
Two structural advantages the DMA is targeting
1) Access to AI infrastructure and distribution
The HLG paper flags concentration in cloud and fast integration of AI across gatekeeper services. When AI features are embedded across search, mobile, work suites, and app stores, third parties must pass through those surfaces to reach users.
This setup invites self-preferencing and makes switching hard without proper interoperability. The paper also notes a direct impact on content providers: embedded AI can reduce traffic and visibility, aligning with publisher complaints around AI-generated search features.
2) Exclusive data access and partnerships
The group highlights the compounding value of user interaction data-first-party, often personal, and rarely shared. Combined with exclusive content deals, this creates feedback loops that improve models and widen the gap.
Public pretraining sets help, but the document is blunt: without comparable interaction data, contestability suffers. That's where the DMA could push for fairer access or limits on exclusivity.
The Google case: what Brussels is testing
The Commission is investigating whether Google breached EU competition rules by using web publishers' content in AI Overviews and AI Mode without fair terms or a real ability to refuse, and by giving itself privileged access to YouTube data while blocking rivals from training on it.
Case AT.40983 has no legal deadline. Outcomes could include fines and behavioral or structural remedies if abuse is proven. The core questions: can a platform reuse dependent partners' content for AI without compensation or opt-out, and can it reserve that data exclusively for its own models?
What Google shipped right after
- Preferred Sources in Top Stories went global, letting users bias news modules toward chosen outlets. Google says this doubles clicks to selected sources.
- Links from your news subscriptions get more prominent placement (starting in the Gemini app), with plans to expand to AI Overviews and AI Mode.
- Inline links in AI Mode increased, with short explanations on why a link is useful.
- New AI partnership pilots with major publishers and wires to test features like AI-powered article overviews and audio briefings, with attribution and links.
Translation for teams: expect query result pages to keep evolving. Some outlets may get more visibility via partnerships; others may rely on Preferred Sources adoption and subscription signals.
Interplay of EU rules you now need to account for
- DMA: self-preferencing limits, interoperability duties, ad transparency, data portability obligations for gatekeepers.
- GDPR: lawful bases, consent, purpose limitation, and explainability for automated decisions.
- DSA: transparency for recommender systems and restrictions around dark patterns.
- AI Act: transparency duties and risk management, including explainability for certain systems.
- Data Act: portability and switching for cloud services; contract terms for data access.
- AVMSD/EMFA: ad targeting constraints (minors) and editorial independence interacting with AI transparency.
The HLG pushes for cross-regulatory coordination so the same practice isn't over- or under-enforced across different laws. For builders, that means harmonizing your privacy, model training, and product UX controls instead of treating them as separate tracks.
Practical playbook for tech teams
- Architecture: reduce platform risk with multi-cloud or cloud-agnostic design where feasible; document portability paths and switching costs.
- Interoperability: expose and consume open interfaces; keep adapters for alternative search, feeds, and identity providers ready.
- Data governance: separate training, fine-tuning, and inference logs; track consent and usage rights at the record level; block ingestion where rights are unclear.
- Contracts: inventory all content and data licenses; flag "exclusive" or "AI training" clauses; set alerts for renewal windows to avoid lock-in.
- Telemetry: segment traffic by query intent and SERP feature; capture CTR, dwell, and scroll depth when AI Overviews/Mode appears.
- Publisher engineering: validate structured data; ensure clear attribution paths; monitor referrer patterns for AI surfaces; test paywall/subscription signals where available.
- Policy controls: document your positions on AI scraping, embeddings, and model training; publish and enforce them via robots, headers, and legal notices.
- Security & privacy: apply data minimization for interaction logs; implement subject rights tooling for portability and deletion.
- Risk reviews: run cross-functional reviews for new AI features against DMA/DSA/GDPR/AI Act checklists; log decisions and rationales.
Metrics to watch
- CTR delta with AI Overviews/Mode present vs not present
- Share of queries covered by AI Overviews/Mode by topic
- Click share from Preferred Sources users and from subscription-linked users
- Publisher referral mix: Top Stories vs AI modules vs organic blue links
- Content reuse detections and takedown/opt-out response times
Model builders: key risks
- Asymmetric access to premium or first-party data via platform policies
- Licenses that allow platform training but block competitors
- Using content without clear rights, creating liability and model contamination
- Explainability and logging gaps for GDPR/AI Act obligations
Marketing and monetization impact
Publishers report notable CTR declines when AI summaries appear, translating into fewer impressions and weaker ad yield. If you run growth for content products, plan for blended acquisition: search, direct, newsletters, and partnerships-not one dependent channel.
Expect remedies that push for better interoperability, clearer opt-outs, and possibly compensation mechanisms. That shifts how you negotiate data partnerships and how you design APIs and attribution.
Timeline
- 2010: EC opens investigation into Google's search practices
- 2017: €2.4B fine for Shopping case
- May 2, 2023: DMA designates six gatekeepers
- March 2024: DMA obligations become binding
- April 10, 2025: HLG AI subgroup meets on regulatory interplay
- June 30, 2025: Independent Publishers Alliance files AI Overviews complaint
- September 5, 2025: €2.95B fine for ad tech abuse
- November 13, 2025: Google rejects structural ad tech remedies
- December 9, 2025: EC opens AI content investigation (AT.40983)
- December 10, 2025: Google launches Preferred Sources globally and new publisher pilots
What happens next
The Commission will run an in-depth probe with no fixed deadline. Possible outcomes: fines, mandated interoperability, stronger opt-outs, data access commitments, or limits on exclusive deals.
For engineering and product leaders, the move is the same either way: audit your dependencies, secure your data rights, and build for flexibility across infra, distribution, and attribution.
Resources
- EU Digital Markets Act
- European Commission - Competition Policy
- Upskill your team: AI courses by job role (Complete AI Training)
Your membership also unlocks:
