The Digital Markets Act (DMA) was built to police digital gatekeepers. The European Commission now wants to test how far that metaphor can stretch—past app stores, social networks, and marketplaces, and into the server racks.
The Commission has reached the preliminary view that Amazon Web Services (AWS) and Microsoft Azure should be designated as gatekeepers under the DMA. That would mark two firsts. It would be the first time cloud computing falls within the DMA’s reach, and the first time the Commission uses Article 3(8)—the provision that allows it to designate firms that do not meet the law’s numerical thresholds after conducting a market investigation.
AWS and Azure are the two largest cloud providers operating in the European Union. But the DMA’s user-number thresholds were built for consumer-facing platforms, not cloud computing, which is overwhelmingly a business-to-business service. Both providers therefore fall outside those thresholds. To designate them anyway, the Commission must do more than invoke the DMA’s built-in presumptions. It must show, with actual evidence, that AWS and Azure serve as “important gateways” and enjoy “entrenched and durable” market positions.
That makes these designations far more interesting than another lap around the DMA enforcement track. Cloud services do not obviously operate as “gates” in the way two-sided platforms do. They do not sit between business users and end users in the familiar app-store or marketplace sense. One of the DMA’s core rationales—increasing contestability by prying open bottlenecks controlled by gatekeepers—appears, at least at first glance, to be missing.
So the question is not merely whether AWS and Azure are large. Plainly, they are. The question is whether “gatekeeper” remains a meaningful legal category that separates firms with durable bottleneck power from firms that are big in competitive markets. Or is it just Brussels-speak for size, with the statutory criteria serving as the ceremonial chant before the inevitable designation?
Until now, the Commission has relied on the quantitative presumptions in Article 3(2) to designate gatekeepers. The qualitative criteria have never had to carry the load on their own. With AWS and Azure, they finally do.
Can Cloud Be a Gate if No One Walks Through It?
It is tempting to argue that cloud computing simply does not belong in the DMA at all. After all, cloud is infrastructure. It rarely touches consumers directly. On that view, it cannot be an “important gateway for business users to reach end users.”
As a matter of market reality, that intuition has some force. As a matter of law, it is much harder to sustain.
Cloud computing appears expressly in Article 2 as a core platform service. Recital 14 explains why: the listed services have “the capacity to affect a large number of end users and business users, which entails the risk of unfair business practices,” and therefore should fall within the DMA’s scope.
That verb matters. Recital 14 does not say cloud providers “have” a large number of end users. It says they have “the capacity to affect” them. Had the drafters meant the former, they could have said so. The qualitative gateway criterion in Article 3(1)(b) would then do no independent work; it would collapse into the numerical test in Article 3(2).
The choice of “affect” suggests the legislature contemplated—and accepted—that a service could be designated even when it reaches end users only indirectly, through the businesses that build on top of it. In other words, the qualitative thresholds exist precisely for services like cloud, which can affect end users without serving as their direct intermediary.
But “can be designated” does not mean “should be designated.” Eligibility says nothing about whether these providers, in this market, actually satisfy the substantive criteria. Indeed, the very reason the DMA’s recitals make cloud eligible—that it can affect end users without directly intermediating with them—should put us on guard.
When a core platform service lacks the usual features that support a finding of gatekeeper power—when it is not a traditional multi-sided, consumer-facing platform—the remaining criteria deserve more scrutiny, not less. The key question is whether those criteria establish a genuine competitive bottleneck, or merely dress up firm size in statutory clothing.
Size Is Only the First Box
Start with the structure of the test. Article 3(1) makes designation depend on three cumulative conditions. The undertaking must have a “significant impact on the internal market.” It must provide a core platform service that serves as an “important gateway” for business users to reach end users. And it must enjoy an “entrenched and durable position,” either now or in the foreseeable near future.
“Cumulative” is doing real work here. Each element must stand on its own. Evidence that supports one cannot simply be recycled to prop up another. A decision that lists indicators of scale—turnover, capitalization, investment, and vertical integration—and then presses them into service under every heading has really established only the first condition: that the company has a “significant impact on the internal market.”
For AWS and Azure, that first condition is not seriously in dispute. Firms of their size clear it easily. The fight is over the two remaining criteria: whether they are “important gateways,” and whether their positions are “entrenched and durable.”
Big Is Not a Bottleneck
The DMA’s gateway concept is built around intermediation. A core platform service sits between business users and the end users they want to reach. As the General Court held in ByteDance:
[I]n order to consider that business users of a CPS [core platform service] “depend” on it in order to reach their end users, it is not necessary for that CPS to be the only channel through which those undertakings can reach those users. It is sufficient for it to be an important channel for that purpose, which those business users can access only if they have an account on that CPS. (ByteDance, §210)
That distinction matters. The first and third limbs of Article 3(1), taken alone, describe nothing distinctively digital—and nothing distinctly gatekeeper-like. Plenty of firms have a “significant impact” on the internal market. Plenty occupy “entrenched and durable” positions. What separates a gatekeeper from a company that is merely large and long-established is the middle limb: whether the firm provides an “important gateway” for business users to reach end users.
Take that limb seriously, and the DMA has a subject matter of its own: bottlenecks that let certain firms control how business users meet their customers. Hollow it out, and the regulation becomes a checklist of obligations for companies that have been big for three years running.
This gateway model fits marketplaces, social networks, and app stores. It fits raw compute and storage much less comfortably. A business that hosts its application on AWS does not reach customers “through” AWS in any way those customers see or experience. Cloud computing is, instead, an input into the business’s own offering.
Article 3(1)(b) contains two distinct requirements. The service must be a gateway, and the gateway must be important. The presumption that ordinarily carries this limb—Article 3(2)(b)’s thresholds of 45 million monthly active end users and 10,000 yearly active business users—does double duty. Both numbers matter because the legal concern is intermediation between two groups.
The size of the numbers speaks to importance. A service reaching users at that scale plainly matters to the internal market. But the structure of the threshold—the requirement of large numbers on both sides, business users and end users—speaks to the gateway element. A service connecting 10,000 businesses to 45 million consumers is, almost by definition, intermediating between the two groups.
For cloud, both halves of that proxy are missing, and for the same reason. The end users are indirect. People who use an app usually have no relationship with the infrastructure on which it runs. So the numbers are not there. And because cloud has no second side in the relevant sense, the Commission cannot infer “gateway” from “many users of both types.”
That is what forced the Commission into Article 3(8). But Article 3(8) comes with a price: the Commission must prove both “important” and “gateway” separately, in substance, without leaning on the presumption. Evidence that cloud is big—revenues, capacity, investment, or ubiquity as an input—may establish importance. It says much less about gateway status. A decision that stacks up evidence of size and criticality and calls the pile an “important gateway” has proved only half the statutory phrase. It has skipped the half that makes a gatekeeper a gatekeeper, rather than merely big, significant, or useful.
This should not be controversial. The Commission declined to designate X despite X clearing the DMA’s turnover thresholds precisely because it was not an important gateway. It accepted that Gmail and Outlook were not gatekeepers despite each having many millions of business and end users. iMessage cleared the quantitative thresholds and was nonetheless found not important enough.
Those decisions concede that the words in Article 3(1)(b) matter. Designating cloud—where the gateway-to-end-users fit is at its weakest—therefore creates a dilemma. Either the Commission is applying a more lenient gateway standard to infrastructure than it applied to X and email services, raising an equal-treatment problem the firms can and likely will plead; or it is conceding that, for infrastructure, “gateway” now means “big.”
There is, admittedly, a reading of the gateway limb that keeps it meaningful while leaving cloud eligible in principle. It is also the reading to which the Commission should be held. On this view, “gateway” captures the market structure that motivated the DMA in the first place: platform intermediation between distinct user groups, where network effects, data feedback loops, and extreme returns to scale can tip markets and lock them in. That was the diagnosis of the Crémer report and the Furman review, and it is the theory on which the regulation was built.
If that is what “gateway” means, a cloud designation is not impossible. But it requires the Commission to show that something about this market inherently tends toward tipping and entrenchment. That leads directly to the second contested limb, where the evidence can actually be inspected.
Entrenched? Tell That to the Price Cuts.
Here, the economic reality is observable—and it points the wrong way for the Commission.
Start with the structure of the DMA. Article 3(2)(c) presumes an “entrenched and durable position” when the user thresholds in Article 3(2)(b) have been met in each of the last three financial years. In the DMA’s own scheme, entrenchment means gateway numbers sustained over time. When those numbers are unavailable—and their absence is what pushed the Commission into Article 3(8) in the first place—something else must fill the gap. The only plausible substitute is evidence about contestability itself: whether rivals can challenge the firms’ positions.
The case law points the same way. The DMA sets aside the dominance test under Article 102 of the Treaty on the Functioning of the European Union (TFEU), which governs abuse of dominance. But the two inquiries are not unrelated. Dominance is a legal test aimed at an economic question: whether rivals can discipline a firm if it raises prices, restricts output, or lets quality and innovation slide. The General Court defines an “entrenched and durable” position in similar terms, referring to situations where “the contestability of that position is limited” and looking to “the stability of that position over time” (ByteDance, §§296-297).
That makes entrenchment an economic question. The Commission must show actual obstacles to contestability. Strip out that economic reality, and “entrenched and durable” becomes another word for “important,” which the DMA treats as a separate inquiry, or simply for “big.” To designate AWS or Azure on this limb, the Commission must show that their positions are, to some meaningful degree, insulated from competition.
That raises a broader point. As Chad Syverson put it, concentration “is an outcome, not an immutable core determinant of how competitive an industry or market is”—so much so that “we cannot even generally know which way the barometer is oriented.”
The same logic applies to the DMA’s qualitative test. Outside the DMA’s presumptions, “entrenched and durable” is not a structural fact one can read off a market’s shape. It is the result of limited contestability. The question is not how concentrated the market looks in a snapshot, but whether competitive pressure can still discipline the incumbents.
That is why entrenchment cannot rest on market shares alone. It also cannot credibly describe a market where rivals are actively winning and losing business. Yet that is exactly what the cloud industry shows.
AWS’ market share has, by all accounts, declined while Microsoft and Google have made large gains. Meta is reportedly exploring ways to turn unused capacity from its massive Meta Compute buildout into a cloud business, potentially by hosting AI models or selling bare-metal capacity—dedicated server capacity without a cloud provider’s software layer—to so-called neocloud providers, which are newer cloud firms focused largely on AI workloads.
Prices have also fallen dramatically. Amazon Simple Storage Service (S3), AWS’ basic data-storage service, dropped more than 80% in price in the early 2010s and has largely stagnated since, despite rising inflation. Tellingly, AWS accelerated its price cuts once Microsoft entered at sufficient scale to post competitive prices. That is not the behavior of a firm floating above the market. It is rivalry doing its work.
The very practices the Commission treats as lock-in mechanisms have also loosened under competitive and regulatory pressure. AWS, Azure, and Google have scrapped or slashed egress fees—charges customers pay to move data out of a cloud provider’s system—for customers leaving their platforms. A firm that keeps lowering the switching costs that supposedly trap its customers is demonstrating, in real time, that its position is contestable. Around 70% of cloud users also multi-home, meaning they use more than one cloud provider rather than putting all their eggs in one server rack.
Article 3(8) does not let the Commission paper over any of this. Removing the quantitative thresholds means the Commission must prove limited contestability in substance. And the substance here is a market where prices fall, shares move, customers multi-home, and incumbents cut their own switching costs to keep business.
There is another problem the Commission’s reasoning must survive: aggregation. The figure most often cited for cloud—a combined share north of 50%—is the share of two firms together. But the DMA’s test applies at the undertaking level, and Article 3(9) requires each designated service to be a gateway “individually.”
There is no equivalent of Article 102 TFEU’s “collective dominance” under the DMA. A combined number tells us little about whether AWS or Azure individually occupies an unassailable position. Instead, it lumps together two companies that compete intensely with each other and with other providers.
The Commission’s decision to designate AWS and Azure while leaving Google Cloud outside the gate makes the problem sharper. Without the thresholds, what principled line puts Amazon and Microsoft in, but Google out? If the honest answer is “Google has fewer users,” then the unstated basis for the whole exercise is user numbers after all—which is to say, bigness. Microsoft and Amazon are right to press this point: there is no obvious, legitimate basis for excluding Google while including them that does not collapse into the very head-counting Article 3(8) was supposed to move beyond.
This is not to say Google should be designated. It is to say that the apparent basis for designating AWS and Azure is that they have acquired an arbitrarily “high” number of users. That number is not high enough to trigger the presumptions in Article 3(2), but it is apparently high enough to support an inference of “significant impact.” If the whole designation turns on user numbers anyway, what exactly are the quantitative presumptions for?
One caveat remains. The Commission may answer that the problem is forward-looking. The DMA allows it to designate a firm when it is “foreseeable that it will enjoy” an entrenched position “in the near future.” The Commission could concede that AWS and Azure face real competition today, but predict that AI workloads, “data gravity”—the tendency of applications and services to cluster around large pools of data—and sovereign-cloud demand will cement their positions tomorrow.
But “foreseeable entrenchment in the near future” is still a claim about contestability. It requires concrete evidence that competition will stop constraining the firms. The General Court’s Meta ruling enforces that standard.
In that case, the Court annulled the Marketplace designation, not after a full substantive reassessment, but because the Commission’s reasoning about one factual change was “vague and hypothetical,” offered no “specific analysis,” and because the legality of a designation must be assessed “on the basis of the facts and the law as they stood at the time when the measure was adopted.”
A near-future-entrenchment finding built on “cloud is large and AI will lock things in” is conclusory in exactly that way. The forward-looking limb does not license regulatory fan fiction. It requires evidence.
There is also some irony in invoking AI to predict cloud entrenchment. Cloud is precisely what has democratized AI for smaller firms, putting frontier models within reach of startups through services such as Amazon Bedrock and Azure’s OpenAI offering. The dynamic the Commission points to as tomorrow’s bottleneck is, so far, helping lower the barriers today.
The DMA Without Its Training Wheels
None of this guarantees that potential cloud designations will fall. Article 3’s relationship between quantitative presumptions and qualitative criteria allows two readings. On the restrictive reading, the qualitative criteria serve as a real filter. On the permissive reading, they are broad enough that the Commission can “always find something” once a firm reaches sufficient scale.
Because the DMA is written at a high level of generality, the honest prediction is that the Commission will choose the permissive route—and the courts may let it. The regulation does not clearly foreclose that reading. Brussels left itself a wide lane, and it should surprise no one if it drives down the middle of it.
The European Court of Justice’s early rulings suggest any judicial check will not come from a court-imposed market-power standard. Recital 5 cuts against that. Instead, it will come—if it comes at all—from careful review of the Commission’s reasoning.
ByteDance and Meta were both presumption cases. The Commission won on substance because the thresholds, not its own independent analysis, carried the qualitative criteria. The one time the General Court struck down a designation—Marketplace—it did so because the Commission’s reasoning failed. The Court held the Commission to the facts as they stood when the designation was adopted and demanded a concrete analysis the Commission had not supplied.
The standard is the ordinary one for any EU act:
[T]he statement of reasons required by … Article 296 TFEU … must disclose in a clear and unequivocal fashion the reasoning followed by the institution which adopted the measure … to enable the persons concerned to ascertain the reasons for it and to enable the court having jurisdiction to exercise its power of review. (Meta, §53)
In other words, the European Court of Justice is unlikely to say “market power.” But it can vindicate the same substance by requiring the Commission to explain, with evidence rather than adjectives, why a sub-threshold service is both a gateway and entrenched. Whether the qualitative criteria mean anything will depend on whether the courts insist that the Commission do more than admire the size of the target.
That is why the AWS and Azure designations matter so much. In every prior designation, the presumption let the hardest questions go largely untested. The courts could avoid asking what “important gateway” and “entrenched and durable” mean when the numbers do not carry the answer. They cannot duck those questions here.
If the Commission designates AWS and Azure, appeals are almost certain. The resulting litigation may be the most consequential the DMA has yet produced. Article 3(8) offers no numerical crutch. The Commission must prove gateway power and entrenchment on a full record, under the kind of reasoning review the General Court applied to Marketplace.
As the General Court has confirmed, Article 3(8) is the route for examining “whether an undertaking should be designated pursuant to Article 3(8)… despite the fact that the undertaking or the [core platform services] in question do not satisfy the thresholds.” Put differently, these two designations would be the first time the DMA’s substantive criteria must stand on their own.
They will also answer the question this post began with. If a sub-threshold cloud designation survives on “large and sticky”—in a market with falling prices, shifting shares, widespread multi-homing, and incumbents cutting their own switching costs under competitive pressure—then the answer is settled. The DMA is a test of bigness. There was nothing else available to carry the designation, and bigness carried it anyway.
The contrast with the parallel exercise across the Channel makes the stakes even clearer. The United Kingdom’s Competition and Markets Authority (CMA) examined the same market and the same two firms under a regime whose gatekeeper logic closely tracks the DMA’s. Its market investigation found that AWS and Microsoft hold significant market power, and the inquiry group recommended that the CMA Board prioritize strategic market status (SMS) investigations into both providers’ cloud businesses.
The Board looked at the same record and—so far, at least—declined. In March 2026, it opted instead for supervised commitments. The one concern the CMA deemed serious enough for possible designation—Microsoft’s software licensing—was routed into an SMS investigation of Microsoft’s business-software ecosystem, not cloud itself.
So the divergence is not about the facts. It is about what the facts warrant. Armed with designation powers of its own and a dedicated market-investigation record, the UK authority concluded that supervised commitments—not designation—were the proportionate response to a market that is still moving, and whose centrality to AI-driven growth counsels against regulatory overkill.
The Commission now proposes to go further on a thinner basis: a first-ever sub-threshold designation, carrying the full weight of the DMA’s obligations, in the very market its British counterpart just concluded could be steered with a lighter touch.
If that designation is made, and if it survives review, the divergence will tell us less about cloud than about the DMA. One regime asked whether these firms control a gate. The other may decide it is enough that they are very large and standing nearby. In Brussels, that may be all the gate you need.
