A regulator signals before it strikes
On June 30, 2026, the Bundeskartellamt published its Annual Report 2025/26, and buried in the usual accounting of cartel fines and merger reviews was a notable forward-looking line: artificial intelligence is now shaping the authority's competition enforcement. The report states that "the changes to the economic landscape brought about by artificial intelligence are already playing a role in various proceedings, both in terms of digital platforms strengthening their market power and in terms of companies collaborating to pool data in order to train AI models" (Bundeskartellamt, June 30, 2026). No AI-specific case has been opened. The signal is the story: Germany's cartel office is telling firms, well before any complaint lands, that both AI-driven entrenchment and AI-training data pools sit inside its Section 19a GWB toolkit.
That toolkit is not hypothetical. Section 19a, added to the Act against Restraints of Competition in 2021, lets the Bundeskartellamt designate a company as having "paramount significance for competition across markets" and then police specific practices — self-preferencing, data leveraging, interoperability barriers — through a fast-track abuse regime, rather than the years-long dominance case a standard Article 102 TFEU claim requires. Five companies now carry that designation: Alphabet/Google (January 2022), Meta/Facebook (May 2022), Amazon (July 2022), Apple (April 2023), and Microsoft (September 2024) (Bundeskartellamt digital economy proceedings overview). Every one of them is also a frontline AI infrastructure provider — cloud compute, foundation models, or the distribution chokepoints AI products need to reach users. Naming AI a priority inside a regime that already covers all five is less a new front than an expansion of the same one.
The Apple probe is the live test case
The clearest evidence Section 19a has teeth, and the backdrop against which the AI language should be read, is the Bundeskartellamt's ongoing case against Apple's App Tracking Transparency Framework (ATTF). On February 13, 2025, the authority sent Apple a preliminary legal assessment arguing the ATTF likely violates Section 19a(2) GWB and Article 102 TFEU. The complaint is specific: Apple defines "tracking" narrowly enough to exempt its own cross-app data combination, third-party apps face up to four consecutive consent prompts versus Apple's two, and the two sets of dialogues are designed with opposite defaults — nudging users to allow Apple's data use while steering them to refuse everyone else's (Bundeskartellamt ATTF press release).
That case only has legal force because Germany's top civil court cleared the underlying framework first. On March 18, 2025, the Bundesgerichtshof confirmed Apple's paramount-significance designation, rejecting Apple's argument that it is merely a hardware seller rather than a platform operator, and affirming that Apple retains "broad and deep access to data" across its ecosystem (SCiDA Project case analysis). With that ruling in hand, the Bundeskartellamt has a court-tested template for exactly the kind of ecosystem-level, data-access argument an AI case would require — a foundation model maker favoring its own downstream product, or a cloud platform tying AI compute access to its other services.
The steelman, and where it breaks
The case for treating AI as a Section 19a priority is genuinely strong. Foundation models are trained on scraped and licensed data at a scale only a handful of firms can assemble; the same five companies designated under 19a already control the cloud infrastructure, distribution channels, and user data most AI products depend on. If dominance compounds — data advantage begets better models begets more data — waiting for a completed monopoly before intervening, as traditional abuse-of-dominance law does, could lock in an oligopoly before any regulator acts. The Bundeskartellamt's own February 2026 action against Amazon shows the tool isn't rhetorical: the authority prohibited Amazon's algorithmic price-control mechanisms on its Marketplace and ordered a €59 million disgorgement of profits — the first such order under the reformed GWB provisions (Jones Day summary of the ruling). Regulators who wait for harm to fully materialize in fast-moving markets often end up unwinding it instead of preventing it.
But naming a technology category a priority before any case exists carries real cost too. The same report that raised AI concerns also touts the Bundeskartellamt's own expansion of "AI-supported applications" for cartel screening and dawn-raid targeting — the regulator wants the efficiency gains from the same technology it's flagging as a market-power risk when firms use it (Annual Report 2025/26 press release). Data pooling to train models is also, in many configurations, exactly what antitrust law elsewhere encourages: smaller firms combining resources to compete with incumbents that can already assemble proprietary datasets alone. A prosecutorial posture toward AI-training collaborations, absent a demonstrated harm, risks penalizing the pooling that's pro-competitive right alongside the pooling that entrenches a leader — and does so under a designation regime that already gives the Bundeskartellamt unusually fast-moving powers against five of the world's largest technology companies.
What to watch
The Apple ATTF case remains the near-term signal. A final decision — not merely the February 2025 preliminary assessment — will show how far the Bundeskartellamt is willing to extend Section 19a's ecosystem-dominance theory, and whether that reasoning gets explicitly repurposed for AI-specific conduct. Given the five-year duration of each paramount-significance designation and the fact that all five run through 2027–2029, expect the AI language in this report to surface as a specific proceeding, not a standalone AI statute, well before Germany's next annual report.