Refusal to Deal and Essential Facilities: When Must Dominant Firms Share?

Refusal to deal and essential facilities is the doctrine that decides when a powerful firm must share — its network, its platform, its IP, its data — with the rivals it would rather starve. It sits on competition law’s deepest tension: forced sharing corrects bottleneck power but dulls the incentive to build bottlenecks worth owning. This guide maps the conditions, the canonical cases and the modern access battles, as part of the dominance pillar of our Competition & Antitrust hub.

Why is there any duty to deal at all?

Because some inputs cannot be replicated — the port, the national local-loop network, the copyrighted programme listings, the operating system’s APIs — and control of them converts one market’s dominance into every adjacent market’s. Where duplication is impossible or economically irrational, refusal is not competition on the merits but leverage of a bottleneck; ordering access restores the downstream contest at the cost of supervising terms.

The counter-argument is dynamic: investment builds bottlenecks, and compelled sharing taxes the winners’ returns, teaching firms to underbuild or free-ride. Every legal system prices this trade-off differently — which is why the doctrine’s conditions are strict everywhere, strictest in the US, and why regulation (telecoms access regimes, open banking, the DMA) increasingly does the work adjudication is too slow for. The full abuse-law context is in our dominance overview.

What do the canonical EU cases hold?

Commercial Solvents (1974): a dominant input maker cutting off a downstream customer to reserve the market for itself abuses — the founding authority. Magill (1995): exceptionally, refusing to license IP (TV listings) abuses where it blocks a new product with consumer demand, the refusal is unjustified and it reserves a secondary market. Bronner (1998): the discipline — a newspaper delivery network need not be shared, because indispensability means duplication is unviable for an equally efficient rival, not merely costly; Advocate General Jacobs’ opinion on incentives remains the doctrine’s intellectual anchor.

IMS Health (2004) confirmed the IP conditions; Microsoft (2007) applied them at scale — interoperability information ordered disclosed where refusal risked eliminating workgroup-server competition and stifled technical development, with ‘new product’ read broadly as innovation harm. Slovak Telekom (2021) then rationalised the field: strict Bronner applies to outright refusals to grant access never given; degraded or unfair terms on existing/mandated access (‘constructive refusal’) and margin squeeze are judged on general effects standards. The result is a two-track doctrine — hard to open the pipe, easier to police the pipe once open.

How did the US close the door — and what remains open?

Aspen Skiing (1985) — a monopolist terminating a profitable joint venture to injure its rival, forsaking short-run profit — remains the only Supreme Court-endorsed refusal liability, and Trinko (2004) confined it to its facts: no general duty to deal, no essential-facilities doctrine endorsed, and regulatory access regimes displace antitrust supervision. linkLine extinguished margin squeeze. The policy: courts are poor price regulators, and forced sharing chills the investments monopoly profits reward.

What survives: refusals that abandon profitable existing courses of dealing to exclude (the Aspen sliver), conditional-dealing theories reframed as exclusive dealing or tying, and — importantly — regulatory regimes doing the access work (telecom interconnection, standard-essential-patent FRAND obligations enforced through contract and patent law). The transatlantic architecture is thus inverted: in the US, access duties are sector-regulatory or nothing; in the EU and Türkiye, competition law itself carries a live, if narrow, compulsion power — a difference every platform and infrastructure owner must internalise regime by regime.

What counts as constructive refusal in practice?

Everything short of “no” that achieves “no”: interminable negotiation and technical stalling; degraded quality or delayed provisioning for rivals versus the incumbent’s own downstream arm; pricing the access so high, or the process so opaque, that entry dies in procurement (Slovak Telekom’s fact pattern); tying access to unrelated conditions; and information asymmetries — giving your own unit roadmaps and specifications rivals receive late or never.

Authorities read these patterns through discrimination and effects lenses rather than the Bronner ladder, which matters strategically: incumbents that have granted access — voluntarily or under regulation — carry an equal-treatment burden that outright refusers paradoxically escape, and internal service levels become the benchmark rivals are entitled to approximate. For infrastructure and platform operators the compliance architecture is measurable parity: recorded SLAs, comparable APIs, audit trails demonstrating that your downstream business queues like everyone else. The same parity logic drives self-preferencing enforcement — refusal’s modern, softer sibling.

Where are the access battles now — data, interoperability, AI?

Three fronts. Data access: business users’ claims to the data their activity generates on platforms (DMA Article 6 grants it to gatekeepers’ business users), automotive and IoT telematics access for independent service providers, and financial-data regimes (open banking) as sector templates. Interoperability: messaging interop under the DMA, device-ecosystem openings (the Apple specification decisions), and standard-essential patents’ FRAND regime as the long-running model of compelled licensing.

AI inputs: compute, foundation-model access and training data are being analysed in exactly these terms — the US search remedies’ data-syndication orders are essential-facilities relief in modern dress, and authorities studying AI verticals ask openly whether model access will need Bronner-style treatment; our AI and competition analysis tracks the front. The direction of travel is consistent: adjudicated compulsion remains narrow, while regulation generalises access duties for designated bottleneck holders — the DMA’s wager that ex ante sharing beats decade-long Microsoft-style litigation.

How do commitments and regulation now deliver most access relief?

Quietly, the compulsion doctrine’s work has migrated into negotiated and legislated instruments. Commitment decisions deliver access without infringement findings: Apple’s NFC opening, energy-network capacity releases, Amazon’s Buy Box and data commitments — each a Bronner-shaped remedy obtained in a fraction of litigation time. Sector regulation generalises the pattern: telecom access regimes, payment-system access rules, open banking, and now the DMA’s per-se duties for gatekeepers.

The division of labour is rational: adjudication sets the outer principles and handles novel bottlenecks; regulation industrialises access where bottlenecks are stable and known. For strategy, the routing matters — a dependent business may get relief faster from a commitments-minded authority or a sector regulator than from an Article 102 case, while an infrastructure owner should assume that today’s voluntary access terms are tomorrow’s regulated baseline, and design them survivably.

What should bottleneck owners do before the complaint arrives?

Govern the asset as if access were already regulated. Publish objective, stable eligibility and pricing criteria; keep internal-use and third-party service levels measurably comparable; document refusals against pre-set criteria with reasons; and quarantine the downstream business from access decisions — the organisational firewall that defeats discrimination narratives. Where IP is the bottleneck, calibrate licensing posture to the Magill risk: blanket refusal across an entire dependent market is the fact pattern that loses.

Above all, price the incentive argument honestly: courts credit investment-protection justifications that match actual investment records and risk, not rhetorical ones. An access policy built on these lines rarely faces compulsion — and when it does, it litigates from the high ground of demonstrated reasonableness rather than reconstructing justifications under subpoena.

How does the doctrine intersect with data protection and IP policy?

Access remedies increasingly collide with other legal regimes. Data-access orders must reconcile with GDPR and Türkiye’s KVKK: business-user data flows under the DMA are engineered around personal-data minimisation, and ‘privacy’ is now a contested justification — genuinely load-bearing in some refusals, pretextual in others, and authorities have grown skeptical of privacy arguments that exclusively disadvantage rivals (the German Facebook saga ran the interaction to the EU Court, which confirmed competition authorities may consider GDPR compliance in abuse analysis).

On the IP side, compulsory licensing under Magill coexists with patent policy’s FRAND machinery and with trade-secret protection for interface information — Microsoft settled where the line runs: protocol information needed for interoperability can be ordered disclosed against reasonable royalties, innovation core stays protected. Counsel structuring either side of an access dispute must brief all three regimes; single-lens arguments lose to tribunals that now read them together.

Frequently Asked Questions