EU Data Act and cloud switching: the end of exit fees and what to do about it

What Chapter VI actually says

Regulation (EU) 2023/2854, known as the Data Act, has applied since 12 September 2025 and is directly applicable in all member states. Most of the text deals with data from connected devices. For infrastructure operations, what matters is Chapter VI (Articles 23–31) on switching between providers of “data processing services” and Article 34 on the parallel use of multiple providers.

The notion of a data processing service is broad. Article 2 defines it as a digital service that enables ubiquitous, on-demand network access to a shared pool of configurable, scalable and elastic computing resources. That covers IaaS, PaaS and SaaS – so not only the hyperscalers, but also the vendor of your CRM or observability platform.

The mandatory contractual minimum

Article 25 requires the customer's switching rights to be clearly set out in a written contract. Among other things, it must include:

• a notice period for initiating the switch of at most two months (para. 2(d));
• a transition period of at most 30 calendar days, during which the provider supports the switch (para. 2(a));
• an obligation to support the customer's exit strategy (para. 2(b));
• an exhaustive specification of the categories of data and digital assets that can be ported (para. 2(e));
• a data retrieval period of at least 30 calendar days after the end of the transition period (para. 2(g));
• a guarantee of full erasure of the data once that period has expired (para. 2(h)).

If the provider cannot technically manage 30 days, it must notify the customer within 14 working days, give its reasons and propose an alternative period, which may not exceed seven months (para. 4). The customer, in turn, has the right to extend the transition period once, for a period it considers more appropriate (para. 5). For a larger migration, that right matters more than the 30 days themselves.

Fees: what ends on 12 January 2027 and what does not

Article 29 introduces a two-phase rule. From 11 January 2024 to 12 January 2027, providers may charge only reduced switching charges that do not exceed the costs directly linked to the switch in question. From 12 January 2027, they may not charge any.

The definition is decisive. Under Article 2(36), “switching charges” are charges other than standard service fees or early termination penalties that a provider imposes for the actions the regulation mandates during a switch. Data egress charges are defined in point 35 as charges for transferring data over the network from the provider's infrastructure to another provider's system or to on-premises infrastructure.

The provider also has an information duty: under Article 29(4) and (5), it must clearly inform prospective customers before the contract is concluded about standard fees, early termination penalties and reduced switching charges, and flag services for which switching is highly complex or costly. This information is to be publicly available on its website (para. 6). It is worth finding and reading it before the price list.

How the hyperscalers have responded

The three largest providers have taken different approaches, and none of them is automatic – you always have to take action.

AWS

AWS announced free data transfer out when leaving as early as March 2024. According to the update of 30 September 2025, requests go through AWS Support, are assessed at account level, compensation comes as credits and there are 90 days to complete the migration. You do not have to close the account, and AWS scrutinizes repeated requests from the same account more closely. For parallel use of another cloud, AWS stated, according to The Register, that it will not charge more than cost; here too, you request it through support.

Microsoft Azure

For parallel operation, Microsoft offers data transfer at cost. According to the documentation (updated 5 May 2026), the customer must have a billing address in the EEA, EFTA or the United Kingdom, the data must not be stored outside these territories and the destination endpoint must belong to the same organization. You apply through Azure Support, stating the subscription ID, the autonomous system number (ASN) of the destination endpoint and an estimate of the share of data transferred. The difference is returned as a refund which, according to the documentation, is applied automatically in subsequent billing periods. It does not cover CDN or transfers to third-party customers.

Google Cloud

On 10 September 2025, Google launched Data Transfer Essentials for customers in the EU and the United Kingdom. Qualifying multicloud traffic between workloads of the same organization is metered separately and appears on the invoice at zero cost. Google itself notes that the regulation would allow it to pass costs on. The customer has to opt in and tag the traffic through configuration; all other traffic is billed at standard rates.

Digital Omnibus: the deadlines still stand unchanged

In November 2025 the Commission proposed the Digital Omnibus package, which amends the Data Act. In Chapter VI, according to an analysis by White & Case, the rules are adjusted for custom-built services and for small and medium-sized providers – these are limited exemptions for contracts concluded before 12 September 2025 and proportionate early termination penalties.

According to an analysis by Greenberg Traurig of 15 July 2026, the Council's compromise text provides for two exemptions for contracts predating 12 September 2025: for services where the provider has customized most features for a specific customer, and for small and medium-sized providers of services other than infrastructure services. The text is also meant to confirm explicitly that proportionate early termination penalties for fixed-term contracts are permissible.

None of this has been adopted, however. According to a status overview of 24 September 2026, the data part of the package is still at first reading, the Council vote on a negotiating mandate was canceled because member states could not agree, and trilogue has not started. Adoption is not expected before the end of 2026.

The practical conclusion: plan according to the law as it stands. The cited analyses do not list a shift of the 12 January 2027 deadline among the proposed changes, and the proposed exemptions target providers of custom-built services and smaller SaaS vendors, not the standard catalog services of the large clouds. If you buy a heavily customized service, however, expect your position to be weaker.

What the Data Act does not solve: technical lock-in

Egress was always the most visible item of an exit, but rarely the largest. What is expensive is rewriting an application built on one provider's proprietary queue, database, identity and serverless runtime. The regulation answers this only partially.

• Functional equivalence applies only to IaaS. Under Article 30(1), an infrastructure provider must take all reasonable measures to enable the customer to achieve functional equivalence after switching to a service of the same type. Moreover, it is defined as a “minimum level of functionality”, not as identical behavior.
• For PaaS and SaaS, it is about interfaces and export. The provider must make open interfaces available free of charge and, as long as no common specifications exist, export data on request in a structured, commonly used and machine-readable format (paras. 2 and 5). But exporting data is not migrating an application.
• Nobody will build anything for you. Under para. 6, the provider is not required to develop new technologies or services, nor to hand over assets protected by intellectual property rights.
• Exemptions for custom-built services. Article 31 exempts from some obligations (including the abolition of fees under Article 29) services where most of the main features were built for a specific customer and which the provider does not offer at broad commercial scale in its catalog. Chapter VI does not apply at all to time-limited non-production versions intended for testing. The provider must point this out in advance.

What matters, then, is how many proprietary dependencies an application has and how well isolated they are. Infrastructure defined as code and a standard deployment pipeline shorten the time to exit far more than any contractual clause; the wider context is covered in our articles on getting infrastructure in order and on Terraform module design.

Multi-cloud: an argument in favor, or just a cheaper tax?

Article 34 makes parallel operation in multiple clouds cheaper: a provider may charge egress between your own environments only up to the costs it actually incurs. That changes the calculation for a few specific scenarios – analytics or AI workloads in a different cloud from the primary data, disaster recovery at a second provider, data replication to meet regulatory requirements for an exit strategy.

The counterargument is just as concrete. Egress was only one of the cost items of multi-cloud. Twice the team skills, twice the security model, twice the networking and twice the tooling remain. For a company without a specific reason – regulation, negotiating leverage or a service available only elsewhere – one well-run cloud with a prepared exit strategy is usually cheaper and safer than two run halfway.

A sensible compromise is portability without parallel operation: Kubernetes as a common layer for applications, open data formats, infrastructure as code and a regularly tested export. Where a proprietary service delivers clear value, use it – just do so deliberately and with a written estimate of what replacing it would cost.

What to do now

Only months remain until 12 January 2027. The contractual rules already apply today, though, so there is no point in postponing most of the steps.

Negotiating contracts

Under Article 41, the Commission has prepared non-binding standard contractual clauses for cloud. The draft recommendation published in November 2025 covers six areas: switching and exit, termination, security and business continuity, a ban on fragmenting contractual documentation, a ban on unilateral changes, and liability. They are voluntary but useful as a reference point – especially with smaller SaaS vendors, where there is room to negotiate.

With hyperscalers, standard terms are hard to negotiate. The real leverage lies elsewhere: the length and size of the commitment, the discount in exchange for it, and the terms for terminating it early.

The Czech context

The regulation applies directly, regardless of Czech implementing legislation. Supervision and penalties, however, are set by the member states. The Czech Telecommunication Office (ČTÚ) stated in September 2025 that, under the implementing act being prepared, it should take on the role of data coordinator for the Czech Republic. Check the current status of this legislation and the level of penalties; we were unable to confirm them reliably for this article.

Key takeaways

• The deadline is 12 January 2027. From that date, switching charges, including egress on exit, are prohibited; until then they may cover only direct costs.
• The contractual rules already apply. A notice period of at most two months, a 30-day transition period and at least 30 days to retrieve data should be in every contract.
• Commitments remain. Standard fees and early termination penalties are not switching charges.
• Hyperscaler relief is not automatic. AWS and Azure require a request through support, Google requires opt-in and configuration.
• The Digital Omnibus changes nothing yet. As of October 2026 the data part of the package has not been adopted, and the proposed exemptions mainly concern custom-built services and smaller providers.
• Technical lock-in remains your problem. Functional equivalence applies only to IaaS; portability of PaaS workloads is a matter of architecture.