Owning and operating the latest large-scale GPU hardware is a non-negotiable, essential component for maintaining a competitive edge in AI translation.
Specialized translation models are inherently superior to general-purpose generative AI for translation tasks, exhibiting lower rates of hallucination and enabling better performance through custom architectures.
The business of AI translation for enterprises is won on incremental quality improvements, as each gain provides a significant ROI by reducing human review time.
Training unique, large-scale AI models for each individual customer is an unscalable and generally poor business strategy, justifiable only in rare cases with a clear and substantial return on investment.
The future of translation services lies in moving beyond simple text conversion to automating entire enterprise workflows, including complex processes like review cycles and versioning.
▶Vertical Integration as a Competitive MoatApr 2026
Kotelowski argues that owning and operating the latest large-scale GPU hardware, building proprietary data centers since 2017, and developing custom software for model scheduling are essential to DeepL's competitive edge. This control over the full technology stack, from hardware to model architecture, is presented as a core strategic advantage.
This high-capital, vertically-integrated strategy creates a significant barrier to entry for potential competitors, but also introduces risks related to hardware dependency, particularly on NVIDIA, and high fixed costs.
▶The Superiority of Specialized AI ModelsApr 2026
A core tenet of Kotelowski's philosophy is that specialized models, trained on curated data, outperform general-purpose generative AI for translation. He claims these models are less prone to hallucination and that DeepL has found custom neural network architectures that are better suited for translation than the standard Transformer model.
This focus on specialization positions DeepL as a best-in-class solution against larger, generalist AI providers, arguing that domain-specific optimization still trumps raw scale for high-stakes enterprise use cases.
▶The Evolving Enterprise Translation MarketApr 2026
Kotelowski observes a shift in how enterprises procure translation services, moving from a centralized function to a decentralized, self-serve model where individual departments like legal and marketing buy directly. This trend is driven by the clear ROI that even small improvements in translation quality can offer by saving time on human review.
This market dynamic suggests that DeepL's go-to-market strategy must focus on product-led growth and targeted marketing to specific departmental personas, rather than solely relying on large, top-down enterprise sales.
▶Human-AI Symbiosis in Translation's FutureApr 2026
Kotelowski predicts that while AI will severely reduce the volume of work done exclusively by humans, they will remain essential for quality guarantees in high-compliance sectors. At DeepL, thousands of human translators are already integral to the AI development process, providing feedback, quality assurance, and training data.
DeepL's strategy is not to fully automate and replace human translators but to create a scalable, hybrid model where human expertise is a critical component for training, validation, and handling high-value tasks.