DeepL, a global leading AI language platform, has announced the launch of Europe’s first NVIDIA DGX SuperPOD computing cluster equipped with DGX GB200 systems. The new computing infrastructure, located at the Swedish EcoDataCenter, will enable DeepL to translate the entire internet in just 18.5 days, instead of the previous 194 days. 

This implementation represents a breakthrough in the development of language technologies and generative AI. Thanks to the new cluster, DeepL will significantly improve the quality, speed, and interactivity of its translations and expand its offering with new features based on generative and multimodal models. In the long term, the company plans to personalize its services and further develop AI models that integrate different modalities (e.g., text, image). 

The new computing capabilities will also enable faster and more advanced training of the language models. For example, the time needed to translate the Oxford English Dictionary will be reduced from 39 to 2 seconds, and Marcel Proust’s monumental novel In Search of Lost Time from 0.95 to just 0.09 seconds. 

This is DeepL’s third implementation of the NVIDIA DGX SuperPOD cluster, with the new solution significantly outperforming DeepL’s existing Mercury supercomputer. As Charlie Boyle, vice president of DGX systems at NVIDIA, pointed out, such investments are crucial to maintaining Europe’s competitiveness in the field of AI. 

DeepL’s new infrastructure is not only a milestone in language translation, but also an inspiring example of the use of the latest technological advances in artificial intelligence research.