Someone opens Spotify on their iPhone and takes out a subscription. Thirty seconds, two screens, a saved card. The question I want to ask is deliberately naive: who has a right to be paid for that subscription?
Spotify, obviously, because it produces the service. The bank, because it processes the payment. The state, through tax. And Apple, because it built the device the screen appears on, the operating system that draws it, the tools the application was written with, and the channel through which Spotify met that person. None of these claims is absurd. Each rests on something different.
The question looks like accounting. It actually contains an entire theory of digital power, and over the last sixteen months Europe has started answering it with numbers. On August 18, yesterday, Apple announced that from October 1 those numbers change again. This is a good moment to look not at the numbers, but at the fact that they have become something to be decided.
Apple Built the Road
Before getting to the thesis it is worth conceding almost everything to the opposing argument, because it is stronger than it is usually made to sound.
Apple did not put a toll booth in front of a road somebody else built. It designed the iPhone and the operating system that governs it. It wrote the development tools, the languages, the compilers, the libraries you build an interface with and the ones you sell a subscription with. It built the code signing and pre-clearance machinery that stops an arbitrary application from doing arbitrary things, the payment infrastructure, the fraud systems, the refund handling, the cataloguing and discovery of applications. Above all it built, over twenty years, a population of users accustomed to buying software with one gesture, which is the hardest and most valuable thing of all.
Seen this way a commission is not necessarily a rent. It can be the price of an enormous amount of value produced by others before you arrived. That is exactly how Apple describes the Core Technology Commission, the percentage it charges those who distribute applications outside its store: remuneration for ongoing investment in the technologies and services that let developers create and distribute software.
The idea that a platform should be forced to permit external payments is also less obvious than it sounds. If Spotify meets a user because that user has an iPhone and opens the App Store, why should Apple receive nothing when the subscription is then completed on Spotify’s own site? And if the infrastructure keeps being used after the purchase, every day, to listen to the music, why should the remuneration end the instant the card is processed elsewhere?
These are legitimate questions with no obvious answers. The European regulation, for its part, does not claim that Apple and Google no longer have a right to be paid. It claims something more precise and far more uncomfortable.
A Closed Door and a Door With a Price
Suppose Apple said: buying Spotify Premium outside the App Store is forbidden. The nature of the restriction is plain, visible to the naked eye, and any competition authority has known what to call it for a century.
Now suppose a second formulation: you may buy Spotify Premium outside the App Store, but for every customer who arrived through my ecosystem you still owe me a commission economically comparable to what you would have paid by staying inside.
Formally the door is open. Economically it may not be. And here an analysis of competition built only on formal rights becomes insufficient, because it records that the door opened and never measures the slope just outside it.
Article 5(4) of the DMA requires the gatekeeper, the platform the regulation designates as an obligatory passage to users, to allow business users to communicate and promote their own offers free of charge, including on different terms, to end users acquired through the platform, and to conclude contracts with them regardless of the channel used. It is a rule about communication and contract, not about prices. It sets no cap.
On April 23, 2025 the Commission fined Apple €500 million, finding that its terms prevented developers from fully benefiting from alternative channels and consumers from learning about potentially cheaper offers. The same day Meta took a €200 million penalty on a different count. Apple appealed in July.
The question that fine opens, rather than closes, is the only one that counts: how much may freedom cost before it stops being freedom? It is a deeply political question, because economic power does not operate only through prohibitions. It operates through incentives, commissions, switching costs, contracts, defaults, friction and fee structures. A system does not need to forbid you from leaving if it can make leaving irrational.
This Is Not Twentieth-Century Antitrust
One wrong way to read the DMA is as an updated version of classical antitrust. Its objective is not to break Apple up, to stop Google running a search engine or to forbid Meta from owning several services.
The Commission describes the regulation as a tool for making digital markets fair and contestable, and identifies gatekeepers precisely as the platforms occupying a particularly important intermediation position. The DMA completes European competition law rather than replacing it.
The difference is decisive. Classical antitrust tends to ask, case by case, whether a company abused a dominant position, and it demands proof of effect. The DMA starts from a prior observation: in some digital markets there are infrastructures so central that waiting each time for the abuse to be demonstrated systematically arrives too late, once the competitor who should have been protected has already shut down.
Hence obligations imposed ex ante, in advance rather than after proof of harm. There is no need to demonstrate each time that stopping a developer from showing a lower price on their own site eliminated a specific rival. It is established beforehand that a gatekeeper may not block that communication, must permit applications to be distributed by other routes, and may not necessarily impose its own payment system.
The regulation’s first review, published on April 28, 2026, concluded that the DMA is fit for purpose and needs no legislative amendment, naming artificial intelligence and cloud as enforcement priorities for the coming years. It is an interested judgement, since it comes from the institution that enforces it, but it is also a choice: no reopening of the text, all the action moved to enforcement.
The economic policy implicit in this construction is remarkable. Europe is saying that some technological positions are so structural that they require rules about the form in which power may be monetised.
The Sentence the Commission Wrote About Google
On July 23, 2026 the Commission adopted two decisions against Google totalling €890 million: €460 million for placing its own services above rivals in search results, and €430 million for obstructing developers who wanted to steer Google Play users toward alternative purchase channels. Sixty days to comply, on pain of periodic payments of up to 5% of worldwide turnover. An appeal suspends neither the fine nor the deadline: comply first, litigate afterwards.
The second decision contains the most important sentence written about the DMA so far, and it has the anonymous look of a technical recital. Google, the Commission says, may receive a fee for having facilitated the initial acquisition of a new customer through Google Play. What goes beyond the permissible is the level of the steering-related fees, the fees tied to sending the customer out to pay, and the length of the period over which those fees remain due.
It is worth stopping on that sentence, because it contains three distinct claims and none of them is trivial.
First: the right to a fee exists. This is not a rhetorical concession, it is an explicit rejection of the argument that digital intermediation is worth nothing.
Second: that right has a measure. There is a level beyond which a commission, however legitimate in principle, becomes incompatible with an obligation to open up.
Third, and most radical: that right expires. There is a moment at which having secured the first meeting between a business and a customer stops generating a claim on the relationship that followed.
Taken together these three claims are something far more sophisticated than “big platforms must work for free.” They say: you may be paid for value you actually produced, and you may not turn your position indefinitely into a claim on value produced afterwards, by others. The boundary will be contested for years. But defining that boundary is the political act.
It also makes the DMA hard to reduce to European distaste for Apple. The problem is not Apple. It is the gatekeeper’s economic right of pursuit.
Price as a Technology of Power
It is worth widening the frame beyond app stores, because the mechanism is general.
We are in the habit of recognising power mainly in prohibition. A state exercises power when it forbids. A company when it prevents. A platform when it expels or censors. It is the mental model we have carried for centuries, and it has the advantage of being visible: a prohibition can be seen, documented, challenged.
In digital markets an enormous share of power is exercised differently. Not by telling you that you cannot do something. By arranging things so that it is not worth doing.
A platform can make a programming interface available and price it so that competitors cannot be born. It can permit data portability and make it slow, partial, in a format nobody reads. It can allow an alternative payment method while applying a fee structure that erases the benefit of the alternative. It can permit interoperability while reserving worse quality, higher latency and fewer features for external interactions. It can let you leave the ecosystem while making departure coincide with the loss of complementary services you cannot do without.
This is a less spectacular power than prohibition. It produces no headlines, generates no outrage, leaves no traces that are easy to hold up in front of a judge. Which is exactly why it often works better.
The DMA explicitly tries to act on this zone. Developers must be able to reach users outside stores, distribute applications through alternative marketplaces or straight from their own website, and pay lower commissions when they give up particular gatekeeper services. The regulator is not only opening the door. It is trying to govern the economics of the door.
Who Decides What Apple Is Worth
This is the strongest objection to the whole edifice, and it deserves room to breathe rather than dismissal.
If we accept that the Commission may establish, even indirectly, when a commission is too high, we are handing a public authority an extremely difficult task: valuing an infrastructure built by a private company. The risk of regulatory excess is real. A commission that looks high may fund components of the ecosystem developers never see and benefit from anyway. A legally imposed cut in the ability to monetise can cut future investment. Technical openness has real costs in security, support and complexity.
Apple argues precisely this, and argues it in writing. In September 2025 it published a document on the DMA’s effects on European users listing features that arrived late or never arrived, risks tied to distribution outside its store and to payment systems that do not meet its standards, and data-sharing requests that on its account extend to the full content of user notifications or the history of every WiFi network a user has joined. The conclusion is that the list of features delayed in Europe will probably get longer.
You do not have to believe every Apple argument is disinterested to recognise that the problem is real. Some of those objections are technically serious and deserve to be checked one by one, not filed away because of who is making them.
But the alternative is not the absence of a political decision. It is that the gatekeeper makes it unilaterally. Setting the commission at 30% is also a decision about how rent is distributed between the owner of the infrastructure, the producer of the service and the consumer. It is simply taken in private, by a board, with no published reasoning and no way to appeal.
This is the heart of the matter. The market does not remove politics when a single infrastructure sets the conditions of access to a market. It privatises it.
Platforms as Private Legislators
An app store is not a shop. A shop decides what to put on the shelves. An app store decides who may distribute software, by what means, which functions are permitted, how payments may be received, which data may be collected, how you enter into a relationship with the user, what share of every transaction is withheld, and which behaviours get you expelled from the market.
Functionally it is a small legal order: substantive rules, procedure, penalty, jurisdiction. All it lacks is the name.
The historical peculiarity of digital platforms is that this private legislative function sits on top of ownership of the infrastructure the rules apply to. Apple writes the rules and owns the territory. Google defines the architecture of search and competes in the markets that search orders. Meta sets the technical conditions on which third parties reach a community of users that exists inside its own infrastructure. It is precisely this coincidence between legislator and owner that the regulation calls gatekeeping.
Seen this way the DMA is not only an expansion of European regulation. It is the return of public law into spaces we treated as private property for twenty years, and which had meanwhile started behaving like institutions.
The Precedent Is Not Antitrust, It Is Railways
The useful historical parallel is not the history of monopolies but the history of infrastructure.
When a technology becomes essential, societies have historically stopped treating every condition of access as a simple exercise of private property. It happened with railways, with telecommunications, with electricity grids, with payment systems. The question is not necessarily nationalisation. It is recognising that whoever controls the infrastructure holds a power exceeding the ordinary relationship between a seller and a customer, and that this power needs rules other than those of contract.
App stores, mobile operating systems, search engines and some digital intermediation services have reached that condition. The same caution applies that I wrote in June about cloud: sovereignty doesn’t live in the data center, and it does not live in a line of a fee schedule either. The DMA is interesting because it tries to regulate private platforms as market infrastructure without formally turning them into public utilities: no legally fixed tariff, no universal service obligation, no public ownership. Only constraints on the form in which infrastructural power may be monetised.
It is a fairly new institutional experiment, and there is no guarantee it works.
What Changed Yesterday
On August 18 Apple announced new terms for European developers, presenting them as the outcome of close collaboration with the Commission and as resolving the open disputes over business terms and alternative distribution. They take effect on October 1 and move every European developer onto a single set of terms.
The numbers. The standard commission for applications using the in-app purchase system drops from 30% to 26%, with 15% for those in the Small Business program, in the Mini Apps and Video partner programs, and for auto-renewing subscriptions after their first year. Developers processing payments themselves pay 20%, cut to 10% for the same categories. Those sending the user to complete the purchase on a website pay 15%, also cut to 10%. Those distributing the application outside the App Store pay a 5% Core Technology Commission on digital transactions, definitively replacing the old Core Technology Fee of 50 cents per install, which used to be owed even when an install produced no revenue at all. And for the first time in Europe a single app may offer Apple’s in-app purchase alongside an alternative, provided it keeps the configuration it picks for twelve months.
The direction is right and should be acknowledged. But it is worth doing the arithmetic, because this is exactly the kind of announcement that gets read as a capitulation and may not be one.
Until yesterday, sending a user to pay outside the App Store cost the sum of three items, all of them published by Apple in the addendum governing external links: 2% for initial acquisition, owed in the first six months after install, a store services charge of 5% or 13% depending on the tier chosen, and the 5% Core Technology Commission. In the first six months that came to between 12% and 20%; afterwards, once the 2% fell away, between 10% and 18%. From October it is one number: 15%.
Whoever sat at the top gains three to five points. Whoever sat at the bottom, meaning whoever had given up the optional services precisely in order to pay less, loses as many. And the distance that matters, the one between staying in and going out, moves from a range of ten to twenty points to a single value of eleven. For a share of developers, leaving has just become slightly less attractive relative to staying than it was.
Then there is a case worth isolating, because it is the one we started from. A mature subscription, meaning one into its second year, pays 15% if it stays inside and 10% if it sends the user to pay on a website: five points of difference, not eleven. And 10% is the same rate the previous addendum already reserved for that category at the base service tier. For the company this essay opened with, in other words, very little changed yesterday. The real winner is the large, ordinary developer, outside any reduced-rate program, who sat on the higher service tier.
I am not arguing that Apple outmanoeuvred the Commission: simplification has real value, predictability and legibility are economic goods, a single price is harder to use for discrimination, and being able to keep two payment systems in one app is a freedom that did not exist before. I am observing two things. First, that contestability is not measured by the number of percentage points announced but by the distance between the options, and that this distance, after sixteen months of penalties, decisions and investigations, remains at a level nobody has yet justified publicly with an argument about value produced. Second, that the twelve-month lock on the payment configuration is, in miniature, exactly the mechanism this essay is about: not a prohibition, a friction.
On October 1 we will not know whether the DMA worked. We will know that the price of leaving has become a quantity negotiated between a private company and a public authority. That is already a historical novelty: three years ago it was a fact communicated to the market.
The Opposite Risk
A counterweight is needed, because otherwise this becomes a celebration, and there is nothing to celebrate.
If the DMA works badly it produces a market in which every platform innovation becomes a matter for negotiation with Brussels. We could move from ecosystems too tightly controlled by companies to ecosystems too heavily designed by the regulator, and the second condition is not obviously better than the first: it is merely less familiar. That Apple presents its new terms as the fruit of very close collaboration with the Commission can be read two ways, and one of them is that the price of applications in Europe is now agreed in a room between two parties, neither of which is the developer who will pay it.
The Commission must therefore take care not to decide which competitors should win. Its task should be narrower: making sure that whoever controls the ground cannot keep changing its gradient to stop others from running. Regulating contestability should not choose the outcome of the market. It should stop the owner of the market from choosing in advance who may compete.
That distinction is easy to state and extremely hard to hold, because every decision about the level of a commission advantages someone in particular. But it is the only line separating the regulation of contestability from the administration of a market.
From it follows the criterion for judging, and it is not the fines. A fine is the system’s intermediate failure: it says the obligation did not work on its own. The desired outcome is that economically viable alternatives exist. The serious test is not whether Apple paid €500 million. It is whether in five years it will be economically plausible to build a European digital company able to reach hundreds of millions of users without granting one party simultaneous control over distribution, payment and the commercial relationship. If the answer is no, the DMA will have failed having collected billions.
The Next Toll Booth Is the Agent
Today we argue about the commissions of an application store. The same structure of problem is already moving elsewhere, and with artificial intelligence it will get harder to see, not easier.
On July 16, 2026 the Commission adopted, under Article 6(7) of the regulation, binding specification decisions, the instrument through which it dictates concrete technical measures instead of merely stating the obligation, requiring Google to give competing AI assistants the same Android access Gemini enjoys. They cover eleven operating-system features and four chapters: the ability of a third-party assistant to be invoked by its own wake word and from system entry points, access to what the user has on screen and the ability to execute tasks across applications, access to the hardware resources, camera and microphone included, needed to be as responsive as Gemini, and access to the on-device data flows that make sense of context. The measures land with Android 18 by August 1, 2027 and Android 19 by August 1, 2028: infrastructure timescales, not software ones.
Tomorrow the questions will be these. What it costs a third-party assistant to be invoked like the system one. What it costs to reach the user’s context. What it costs an agent to take an action on a person’s behalf, and what it costs another agent to accept it. What it costs to appear inside the answer rather than in the tenth position of a list nobody scrolls any more. What compute costs.
I argued a few days ago that with open models the useful question is not whether a model is open, but who may inspect it, run it, modify it, and on what terms. It is the same question as this essay, with a price written beside every line.
The next gatekeeper may not be whoever controls the store. It may be whoever controls the agent through which a user searches, buys, reads, books and decides. And there too the central problem will not be forbidding it from offering its own service. It will be establishing how much it may cost to choose somebody else’s.
Politics Was Already There
For a long time we imagined the internet would remove a growing share of economic life from geography, and therefore from politics. The digital market seemed able to organise itself through protocols, contracts and private property, while states were left the residual task of chasing its abuses a few years late.
The platforms demonstrated the opposite, and they demonstrated it through prices before anything else. When a company controls the place where others do business, the price it sets is no longer only a price. It becomes a rule about how value is distributed, and therefore a norm.
The Digital Markets Act is not interesting because Europe has finally decided how much Apple ought to earn. That would be a poor conclusion, and fortunately nobody has decided it. It is interesting because it forces us to acknowledge something we had hidden under the neutral vocabulary of digital markets: there is a point at which technical architecture becomes an institution, a commission becomes a form of government, and the contractual freedom of whoever owns the infrastructure becomes the economic constraint of everyone else.
At that point the question is no longer whether politics should intervene. Politics is already there, and has been for years, written into the terms of service. The real choice is who exercises it, under which rules, and with what concrete possibility, for everyone else, of actually leaving.
Which is why regulation becomes politics the moment it stops asking power not to do something and starts asking what it may cost not to depend on it. The DMA does not try to make platforms less powerful by decree. It tries to stop every form of autonomy from their power having to be bought at a price the platforms set themselves.
The difference is small only in appearance. Inside it sits much of European digital policy for the next ten years, and on October 1 we get the first instalment.
Key takeaways
On July 23, 2026 the Commission fined Google €890 million, €460 million for self-preferencing in search and €430 million for restrictions on Google Play, with sixty days to comply. The reasoning contains the most important sentence written about the DMA so far: Google may receive a fee for having secured the initial acquisition of a customer, but the level of the fees and the length of the period over which they are charged went beyond what the regulation permits.
The DMA is not twentieth-century antitrust. Antitrust asks whether a company abused its position; the DMA starts from the premise that some infrastructures are so central that waiting each time for proof of abuse arrives too late, and imposes obligations in advance. The regulation’s first review, on April 28, 2026, judged it fit for purpose with no need for legislative amendment.
Digital power is rarely exercised by prohibition. It is exercised by making the alternative not worth taking: an API priced so competitors cannot be born, data portability that is slow and incomplete, an external payment method permitted and simultaneously taxed until the benefit is gone. A system does not need to forbid you from leaving if it can make leaving economically irrational.
On August 18 Apple repriced its European terms: 26% inside the App Store, 20% with your own payment processing, 15% for sending the user out to pay, 5% for distributing outside the App Store, effective October 1. Leaving used to cost between 10% and 20% depending on which services you kept and how old the install was. A single number simplifies, but for developers who sat at the low end leaving became more expensive, and the gap against staying did not widen. For a subscription past its first year that gap is five points, not eleven.
The test of the DMA is not the fines, which are its intermediate failure. It is whether in five years it will be economically plausible to build a European digital company reaching hundreds of millions of users without granting one party simultaneous control over distribution, payment and the commercial relationship. If the answer is no, the DMA will have failed even with billions collected.
Questions & answers
What exactly does the Digital Markets Act prohibit about commissions?
It does not prohibit commissions, and this is the point most often lost. The DMA requires the gatekeeper, the term the regulation uses for platforms that function as obligatory passages, to let developers communicate and promote their own offers free of charge to users acquired through the platform, and to conclude contracts with them regardless of the channel used. That is Article 5(4). The regulation sets no cap on commissions: it is the non-compliance decisions that establish, case by case, when the level and duration of a commission make that right unusable.
Why does the July 2026 Google decision matter more than the 2025 Apple fine?
Because it contains a criterion, not only a penalty. On April 23, 2025 the Commission fined Apple €500 million, finding that its terms prevented developers from benefiting from alternative channels and consumers from learning about cheaper offers. On July 23, 2026, fining Google €890 million, the Commission added the missing piece: it explicitly recognised that Google may charge for having secured the first customer, and objected instead to the level of the fees and the length of the period over which they remain due. It does not deny the right to a fee. It argues about its measure.
What changes for developers under Apple's new terms of August 18, 2026?
Apple announced a single set of terms for all European developers, effective October 1, 2026. The standard commission for apps using its in-app purchase system drops from 30% to 26%, with 15% for those in the Small Business, Mini Apps and Video partner programs. Developers processing payments themselves pay 20%, those sending users to complete a purchase on a website pay 15%, and those distributing outside the App Store pay a 5% Core Technology Commission on digital transactions, replacing the old Core Technology Fee of 50 cents per install. Apple presents the new terms as the outcome of close collaboration with the Commission and as resolving its outstanding disagreements.
Isn't it a problem for a public authority to decide what a private platform is worth?
It is the strongest objection to the whole approach, and it should stay open. A commission that looks high may fund parts of the ecosystem developers never see, and cutting the ability to monetise by law can cut future investment. Apple has argued for years that the openness the DMA imposes lowers protections and delays features in Europe. But the alternative is not the absence of a political decision: it is that the gatekeeper makes it alone. When a single infrastructure sets the conditions of access to a market, the market does not remove politics. It privatises it.
Will the DMA reach artificial intelligence too?
It already is. On July 16, 2026 the Commission adopted binding specification decisions requiring Google to give competing AI assistants the same Android access Gemini enjoys, across eleven operating-system features: wake words, understanding what is on screen, executing tasks across apps, access to hardware resources, on-device data flows carrying user context. The measures land with Android 18 by August 1, 2027 and Android 19 by August 1, 2028. The regulation’s first review, on April 28, 2026, named AI and cloud as enforcement priorities. The next obligatory passage may not be the store, but the agent through which a user searches, buys and books.
The author
