The Egress Fee War Is Over. Data Gravity Has Already Won.

On June 21, 2026, the European Commission took a procedural step that cloud-platform finance teams had been bracing for since the Digital Markets Act was first drafted. The Commission's preliminary view, reported by Tech Times and confirmed days later by Reuters, is that Amazon Web Services and Microsoft Azure should be designated as gatekeeper platforms under the DMA, even though neither meets the law's usual revenue and user-count thresholds. The designation, if finalized, would mandate data portability, open APIs, and, most consequentially, an end to the egress fees that have for more than a decade made it expensive to move data off a hyperscaler's infrastructure.

The June findings represent the sharpest regulatory escalation yet in a multi-front campaign against cloud lock-in. The European Union's Data Act, which took effect in September 2025, already requires providers of data-processing services to phase out switching charges, including egress fees, and to offer standard contractual clauses for cloud migration. The United Kingdom's Competition and Markets Authority opened its own cloud-services market investigation, turning what was initially billed as an Ofcom referral into a broader inquiry that now encompasses Microsoft's software-licensing practices. And in the United States, antitrust enforcers have been collecting complaints about egress pricing, spending minimums, and bundled licensing, as the law firm JD Supra documented in a December 2025 roundup of the transatlantic enforcement landscape. The combined effect is a regulatory consensus that the tollbooth must come down.

Yet the tollbooth was already being dismantled from within, and the sequence of voluntary moves tells its own story about competitive pressure. In January 2024, Google Cloud announced it would no longer charge data-transfer fees for customers who wanted to leave the platform. The move was widely read as an asymmetric attack: Google Cloud held roughly 11 percent of global cloud infrastructure revenue at the time, and making migration costless was a way to lower the perceived risk of choosing a distant-third provider while simultaneously raising the political cost for AWS and Azure to keep their own fees in place. AWS responded in March 2024 with a program allowing departing customers to request credits for data-transfer-out charges, though the credit-application process introduced friction that Google's blanket policy did not. Microsoft, which had historically waived Azure egress fees for large enough commitments, broadened its free-egress language but tied it to specific contract tiers.

This is the moment when the vocabulary of the debate shifted. For most of the 2010s, cloud lock-in was discussed as a pricing problem: the hyperscalers made it cheap to upload data and expensive to retrieve it, creating a cost asymmetry that discouraged multicloud architectures. The 2024 wave of egress-fee eliminations, voluntary and regulatory alike, was supposed to solve that problem. But what became visible once the fees began to fall was a deeper structural force that predated the price lists and would outlast them. That force is data gravity.

The term, coined by software engineer Dave McCrory in 2010, describes a simple dynamic: as data accumulates in one location, it attracts applications, services, and additional data. The metaphor holds because the attractive force grows with mass. A terabyte of customer-transaction logs sitting in an S3 bucket is not expensive to move in raw dollar terms if egress fees are zero. But if that terabyte feeds a SageMaker training pipeline, which writes model artifacts to another S3 prefix, which a Lambda function reads to serve inferences through an API Gateway endpoint that writes telemetry back to CloudWatch, the cost of moving the data is no longer primarily about the transfer charge. It is about the engineering effort required to rewire a constellation of deeply integrated, platform-native services.

Where your data is, is becoming very much a gravitational center for where your computing is happening.Dr. Hillery Hunter, CTO and general manager of innovation, IBM Infrastructure, as quoted by The Motley Fool, June 2026

Hunter's remark, made during an IBM event covered by The Motley Fool in June 2026, captures the strategic reality that the egress-fee war was always a skirmish over the wrong territory. IBM, which competes with the hyperscalers largely from the hybrid-cloud flank, has a commercial interest in arguing that data is stubborn. But the gravitational pattern shows up in the numbers. The Flexera 2026 State of the Cloud report, cited by CRN in March 2026, found that enterprises running on a single primary cloud platform continue to expand their spending on that platform's adjacent services faster than they diversify to competitors. The same report noted that while multicloud adoption has become nearly universal at the organizational level, most individual workloads remain anchored to a single provider, a pattern that is consistent with data exercising a gravitational pull on compute.

The egress-fee story thus splits into two narratives that only partially overlap. The first is about regulation catching up to pricing practices that were, by any reasonable measure, designed to penalize exit. The second is about whether exit is even a meaningful concept when an enterprise's data estate has been shaped around a specific provider's service mesh. The DMA gatekeeper designation, if it comes, will have the most to say about the first narrative and far less about the second, because no regulator can mandate that an S3 bucket be functionally interchangeable with an Azure Blob Storage container in a way that preserves the performance, latency, and security characteristics an application was built to expect.

Neocloud providers have spotted this gap and are moving into it. In November 2025, CoreWeave launched its Zero Egress Migration program, which SDxCentral reported was designed to save customers more than a million dollars on typical data migrations for AI workloads. The program's pitch is not merely that egress is free but that CoreWeave's purpose-built infrastructure for AI training and inference offers a different gravitational center, one organized around GPU availability and InfiniBand interconnects rather than the broad service catalog of a hyperscaler. The question CoreWeave is asking enterprises is not "can you afford to leave?" but "do you want to stay somewhere that was never optimized for what you are building now?"

That question gains force from the changing composition of cloud spend. AI workloads, particularly large-scale training runs, are less coupled to a specific provider's platform services than traditional enterprise applications. A training job on a GPU cluster reads data from object storage and writes checkpoints back, but it does not typically depend on a provider's identity-management system, its queue service, or its proprietary database engine. The data gravity of an AI training pipeline is fungible in ways that a decade-old Oracle-on-EC2 deployment is not. If the hyperscalers' moat was built from a hundred small service dependencies, AI workloads may be the first major category where that moat looks crossable.

Still, the hyperscalers have not been passive. The third quarter in a row of AWS earnings calls has featured analyst questions about AI workload portability; in each case, AWS executives have pointed to the Bedrock managed-service layer and the native integration between S3 and SageMaker as reasons customers stay. Azure's posture is similar, with the OpenAI API serving as a gravitational anchor that pulls data into the Microsoft ecosystem. Google Cloud has leaned hardest into the open-ecosystem argument, but its own BigQuery and Vertex AI services are designed to be sticky in precisely the ways that make migration expensive even when egress is free.

The FinOps View From the Ground

Among the people who actually track cloud costs inside large enterprises, the elimination of egress fees has been received with a mix of relief and realism. The relief is straightforward: data-transfer-out charges, which could run to six or seven figures annually for data-intensive organizations, were a line item that FinOps teams could point to as pure waste. The realism is that eliminating that line item does not make a multicloud strategy any easier to execute. The Flexera 2026 report found that managing cloud spend remains the top challenge for enterprises, cited by 84 percent of respondents, and that optimizing existing cloud usage now consumes more effort than negotiating new commitments. The egress fee was a visible cost of movement; the invisible cost, in engineering hours and architectural complexity, is larger and harder to budget for.

The partner ecosystem offers a useful cross-confirmation. Datadog's quarterly customer surveys have shown a steady increase in the number of organizations monitoring multiple cloud environments, but the same surveys show that the median customer still derives more than 80 percent of its cloud resources from a single provider. Snowflake, which positions itself as a cross-cloud data platform, has been one of the clearest beneficiaries of the egress-fee unwind: its architecture allows customers to store data in one cloud and query it from another without paying transfer penalties. But Snowflake's own revenue growth, which decelerated in 2025 before stabilizing, suggests that cross-cloud analytics is a niche use case, not a mass migration.

What to Watch for in the Second Half of 2026

The DMA gatekeeper designation for AWS and Azure is not yet final; the Commission's preliminary findings trigger a consultation period and a negotiation over remedies. What makes this process different from the EU Data Act is that the DMA carries the power to impose structural remedies, not merely behavioral ones. An outright ban on egress fees, enforceable with fines of up to 10 percent of global annual turnover, would remove the last legal ambiguity about whether the voluntary programs of 2024 went far enough. The more consequential question is whether the Commission will also require open APIs and interoperability standards that address the data-gravity problem at the architectural level. If it does, the hyperscalers will face a compliance burden that their voluntary egress-fee rollbacks were partly designed to preempt.

The UK CMA's Strategic Market Status investigation into Microsoft's software ecosystem, noted by Computer Weekly in March 2026, adds a second front. Where the EU is focused on infrastructure-platform lock-in, the CMA is examining whether Microsoft's bundling of Windows Server, Office 365, and Azure creates a licensing penalty for enterprises that run Microsoft software on competing clouds. The two investigations are complementary and, taken together, suggest that 2026 will be the year when cloud competition policy moves from information-gathering to intervention.

The egress-fee war, measured by headline pricing, is effectively over. Google Cloud fired the first shot in January 2024; AWS and Microsoft adjusted their policies within months; the EU Data Act codified the principle in law by September 2025; and the DMA designation will strip away whatever friction remains. But the data-gravity question remains open, and it will be answered not in a Commission press release but in the architecture decisions made by the next thousand enterprises that face a genuine choice about where to place their next petabyte. The cheapest signal that gravity is weakening will be a sustained shift in the Flexera data showing that the share of workloads anchored to a single hyperscaler is declining. That number has barely moved in three years. Watch for it in Q4.

There is a final, uncomfortable possibility that the regulators and the neoclouds are both right, but for different reasons. The regulators are right that egress fees were an artificial barrier to competition, and removing them is a necessary condition for a functioning market. The neoclouds are right that the real barrier was never the fee; it was the architecture. If both are correct, then the end of the egress-fee war will not produce a sudden flowering of multicloud workloads. It will simply reveal that the data-gravity problem was always the deeper one, and that solving it will require a different set of tools than a price cap.