Gradium, a new AI voice startup spun out of Kyutai and backed by telecom billionaire Xavier Niel, today launched publicly after raising a $70 million seed round led by FirstMark Capital and Eurazeo, with additional investment from Niel, Eric Schmidt, DST Global Partners, and others.
Founded in September 2025 by former researchers from Google DeepMind, Gradium builds audio‑language AI models designed for ultra‑low latency speech synthesis and recognition. The company aims to provide voice that “responds almost instantly,” targeting developers needing scalable, high-fidelity voice output.
At launch, Gradium supports five languages — English, French, German, Spanish and Portuguese — with plans to expand to more. The startup believes demand will grow as AI shifts from text‑based chat toward voice-driven agents in gaming, customer support, education, and more.
Gradium enters a competitive space where frontier AI firms (like OpenAI, Meta, and others) and voice‑specialist startups such as ElevenLabs already operate. But investors appear confident: the size of the seed round and the pedigree of the founding team suggest trust in Gradium’s ability to deliver on latency and quality.
This launch echoes a broader trend of rapid investment in AI infrastructure: just as startups are racing to deliver efficient, customizable image models or agentic tools, voice is becoming the next major interface. The rise of companies like Gradium underscores how voice‑first AI could soon join text and image as core modalities in the AI ecosystem.
