EU eTranslation Passes One Billion Requests: Scaling Language AI Takes More Than a Model

On October 9, the European Commission's Directorate-General for Translation announced that eTranslation had received more than one billion translation requests in 2026—more than the annual total in any previous year since the service launched in 2017. eTranslation covers all 24 official EU languages plus selected high-interest languages. Users can process text and documents through a web interface or integrate the service into public digital systems through APIs.

The announcement counts requests, while related historical materials sometimes describe output in pages; those units should not be compared directly. Whichever measure is used, however, the growth points to the same operational shift: machine translation has moved from occasional assistance into routine workflows spanning public administration, justice, procurement, health, migration, and online services.

The first asset behind one billion requests is not the model

The Commission attributes eTranslation's foundation to multilingual data accumulated over decades and curated by professional translators. Euramis, its language-resource repository established in 1995, now contains more than 100 billion tokens—roughly 270 million pages. Large volumes of translations produced across EU institutions are checked, filtered, quality-controlled, and curated by language professionals before supporting machine translation and other language-AI services.

The value of this data lies not only in volume, but in known provenance, aligned languages, institutional context, and ongoing maintenance. Users can also upload their own glossaries and select modes suited to general text, formal EU language, finance, or case law. Stable output therefore comes from the interaction of a model, translation memories, terminology, style constraints, and feedback—not from a single prompt.

Scale comes from integration—and trust

eTranslation is available through both a web interface and APIs for other digital services. Translation can therefore become infrastructure inside websites, internal communications, and administrative systems instead of a separate copy-and-paste task. For high-volume content, automated file movement, format preservation, terminology retrieval, and processing records often determine adoption more than an impressive one-off translation.

Security and eligibility are equally important. The Commission says the service runs under strict EU data-protection rules, does not use submitted data to train commercial AI models, and is available to eligible EU institutions, public administrations, small businesses, academia, NGOs, and related users. For government and regulated organizations, data flows, access controls, confidentiality, and auditability are prerequisites for production use alongside linguistic quality.

One billion requests do not mean one billion publish-ready outputs

The Commission also warns that machine-translation quality varies by text and language pair, does not guarantee accuracy, and must not replace the authentic language versions of EU legislation. That boundary matters. Usage proves that the tool addresses substantial real demand; it does not prove that every output meets the publication standard for legal, medical, brand, or safety-critical content.

Language AI at scale therefore needs risk tiers. Low-risk, repetitive content used for basic understanding may prioritize automated coverage. Content involving rights, product safety, regulatory submissions, or brand commitments should retain qualified linguists, domain experts, and formal approval. A mature system does not remove people everywhere. It concentrates human judgment where errors have the greatest consequences and models are least certain.

Four questions enterprises should answer before scaling

• Where does the data come from? Are training data, translation memories, and terminology assets traceable, correctly aligned, and continuously maintained by qualified people?
• How does it enter the workflow? Can APIs, files, and content platforms connect to real operations while preserving formatting, terminology, versions, and audit records?
• Which content must not publish automatically? Are risk tiers based on intended use and error consequences, with explicit rules for linguistic review, specialist sign-off, and authoritative versions?
• How does feedback improve the system? Are terminology errors, human edits, weak language pairs, and user reports captured and converted into reusable rules and data?

For companies expanding internationally, one billion eTranslation requests are not a product blueprint to copy directly; eligibility and public-service requirements reflect a specific European institutional context. The transferable lesson is the order of operations: build trusted language assets, integrate them into working systems, establish trust through security and risk tiers, and expand automation through continuous feedback. Models set the capability ceiling. Data, workflows, and accountability determine whether people will use them a billion times.

Sources: European Commission Directorate-General for Translation, “eTranslation reaches 1 billion requests in less than a year” (October 9, 2026); European Commission overview of multilingual AI services and language data; European Commission machine-translation use and liability notice. Requests, pages, and text volume are different units; this article does not convert between them.