A translation company that launched from a Roman apartment in 1999 now counts SpaceX, Airbnb, and Uber among its clients while powering real-time Mass translations at the Vatican in 60 languages. That trajectory reveals how far AI-driven language technology has come, and where it heads next.
Translated operates across 200 languages, combining its proprietary AI systems with a roster of more than 400,000 professional translators. Co-founder Marco Trombetti said the company serves over half a million customers. The goal, he explained, remains unchanged since he and his wife Isabelle founded the business: letting people understand and be understood in their own language.
**A Research-First Approach**
Most translation firms license their AI from Google, Microsoft, or OpenAI. Translated builds its own. Lara, the company’s flagship translation model, beat Google Translate and DeepL in blind evaluations by professional translators, according to Trombetti. ModernMT, its adaptive neural engine, learns in real time from translator corrections. T-Rank matches each project to the best translator using more than 30 factors, including domain expertise and past performance.
The company also coordinates DVPS, a €29 million European Union project on multimodal AI involving 20 institutions, including Oxford, ETH Zurich, and the Alan Turing Institute. The five-year effort aims to build systems that learn from visual input, spatial audio, and sensor data rather than text alone.
**The Vatican Partnership**
Pilgrims at St. Peter’s Basilica can now scan a QR code and follow Mass on their smartphones in 60 languages. The partnership, announced in early 2026, represents one of the first large-scale AI interpretation deployments in a sacred space. Trombetti called the technology “invisible to the user, which is what technology should be.”
**Language Singularity**
Trombetti predicts the company will reach what he calls “language singularity” by 2027 or 2028. At that point, checking a machine translation takes a professional translator no longer than reviewing a colleague’s work. Lara V3, released in July 2026, already translates 23 times as many characters per second as Claude Fable 5 while scoring highest on quality among systems tested.
The company’s trajectory suggests translation may soon stop being a bottleneck for global collaboration, scientific research, and human connection.
