DeepL’s AI Breakthrough: How It Can Now Translate the Entire Internet in Just 18 Days

What if language barriers across the internet could disappear in less than three weeks?

DeepL, the AI-powered translation startup from Germany, says that’s no longer science fiction. With the help of NVIDIA’s cutting-edge DGX SuperPOD systems and a powerful eco-friendly data center in Sweden, DeepL claims it can now process and translate the entire Common Crawl dataset, an open-source archive of the internet, in only 18.5 days.

This is a sharp drop from the previous benchmark of 194 days. While the translation hasn’t yet been run from start to finish at that speed, the infrastructure now makes it feasible. And that changes the pace of global communication and AI development.

From 194 Days to 18: What Changed?

At the center of this leap is the NVIDIA DGX SuperPOD, an AI supercomputing solution built to handle the world’s most demanding language tasks. DeepL’s deployment at Sweden’s EcoDataCenter marks the first use of the DGX SuperPOD with Grace Blackwell GB200 systems in Europe.

The architecture combines 36 racks of Grace Blackwell Superchips, each loaded with multiple GPUs, Mellanox InfiniBand networking, and shared throughput reaching up to 1.4 TB/s. This infrastructure now delivers over 30 times the text output compared to DeepL’s previous setup.

With this capacity, DeepL projects massive speed-ups across translation tasks. Translating the Oxford English Dictionary, which once took 39 seconds, now takes just two. Proust’s famously long novel In Search of Lost Time now requires only 0.09 seconds—down from nearly one second.

Why Speed Is More Than a Metric

This performance boost is not just about bragging rights. Faster processing enables quicker model training, faster product testing, and real-time updates.

DeepL’s new system accelerates its development of tools like DeepL Voice, a feature that launched in late 2024 to support real-time text-based translations in meetings and mobile apps.

With more power behind the scenes, DeepL can now deliver sharper context, better grammar handling, and more adaptive translation in business, manufacturing, and healthcare. They are also preparing for future expansions into generative and multimodal AI models that go beyond static text.

Real Business Value

The benefits extend to the enterprise level. A Forrester Consulting study showed that DeepL delivers a 345% return on investment over three years for global companies, with a payback period under six months. That ROI stems from reducing translation workloads by 50% and saving over €2.7 million in process efficiency.

More speed means fewer bottlenecks and faster go-to-market for multilingual campaigns, legal documents, and customer support in over 30 languages. For the 50% of Fortune 500 companies already using DeepL, the speed upgrade is a direct business advantage.

A Glimpse Into the Future

While DeepL has not yet translated the full internet in 18 days, the hardware now makes that goal achievable. This is not just about speed. It’s about opening the door to real-time global communication at scale.

With plans to keep improving voice features, expand multimodal AI, and further reduce latency, DeepL is building more than a translator, it is building a language engine for the connected world.

And with NVIDIA’s AI dominance and Europe’s growing tech infrastructure, DeepL’s advances may soon become the new normal for how we cross borders with words.

Also read: Meta Buys 49% of Scale AI as 28‑Year‑Old Dropout CEO Joins Its New Superintelligence Team