DeepL, the Cologne-based language translation unicorn, has announced a strategic infrastructure upgrade, integrating VAST Data’s AI operating system into its backend storage environment. The move aims to optimize the company’s high-performance computing (HPC) footprint, which supports its expansive suite of neural machine translation services, including text translation, the DeepL Write assistant, and its real-time voice translation platform.
As DeepL’s research teams continue to scale their Large Language Models (LLMs) and develop autonomous AI agents, the demand for storage efficiency and data accessibility has reached a critical inflection point. By adopting VAST Data’s platform, DeepL anticipates a threefold increase in effective storage capacity from its existing hardware. This efficiency gain is expected to provide researchers with greater headroom for model development without necessitating immediate, massive expansions of physical hardware.
The company, which reached a $2 billion valuation following a $300 million funding round in 2024, has traditionally relied on a combination of bare-metal servers and high-speed NVMe storage. Currently, DeepL operates a sophisticated hybrid infrastructure, utilizing NVIDIA DGX SuperPOD systems—including the cutting-edge DGX GB200 NVL72 architecture housed at EcoDataCenter in Sweden—alongside cloud-based processing via Amazon Web Services (AWS). This hybrid model allows DeepL to leverage the raw performance of on-premises liquid-cooled clusters for training while utilizing cloud resources to handle global API distribution.
Stefan Andresen, DeepL’s Chief Scientist, emphasized that the transition is driven by the need for a more unified data environment. Beyond raw capacity, the VAST integration is designed to streamline data accessibility across DeepL’s private data centers and, eventually, its cloud-based environments. By reducing manual data movement, the company hopes to accelerate the research lifecycle for data-intensive projects, such as its expanding DeepL Voice capabilities.
The migration process is being facilitated by VAST’s SyncEngine technology, which will move existing datasets onto the new platform. The architecture will support both file and object storage, providing a unified foundation for the company’s AI workloads. This shift is particularly significant given DeepL’s reliance on containerized translation code, which allows for seamless deployment across its hybrid infrastructure.
Jeff Denworth, co-founder of VAST Data, noted that the partnership reflects the evolving nature of AI production environments. As models move from experimental phases into high-stakes production, the focus shifts from pure performance to long-term reliability and infrastructure stability. By consolidating its storage under a single AI-optimized OS, DeepL is positioning its backend to handle the increasing complexity of real-time, multi-language AI experiences while maintaining the high-quality, context-aware output that defines its market position.
