Paris, France – Gradium, an artificial intelligence (AI) voice startup, has emerged from stealth operations, announcing a $70 million seed funding round. The company, a spin-out from the French AI research lab Kyutai, specializes in developing audio language AI models engineered for delivering voice at scale with ultra-low latency, enabling near-instantaneous AI voice responses.
Founded in September 2025 by Neil Zeghidour, a founding member of Kyutai and former researcher at Google DeepMind with expertise in voice models, Gradium aims to provide developers with faster and more accurate voice models. The startup launched with native multilingual support, initially offering English, French, German, Spanish, and Portuguese, with additional languages planned for future integration.
The seed round was co-led by FirstMark Capital and Eurazeo. Additional participation came from notable investors including French telecom billionaire Xavier Niel, DST Global Partners, and billionaire Eric Schmidt, among others. The significant capital infusion is expected to support Gradium's continued development and market expansion efforts.
Gradium enters a competitive landscape populated by established frontier large language model (LLM) companies, including OpenAI, Anthropic, Meta Llma, and Mistral, all of which offer voice, speech recognition, and multimodal AI capabilities. The market also includes well-funded startups such as ElevenLabs, alongside a broad spectrum of voice and speech models available on platforms like Hugging Face.
Despite the existing competition, industry projections indicate a growing demand for advanced AI voice capabilities, particularly those offering ultra-realistic expression and accuracy. As AI applications transition from text-based interfaces to sophisticated AI agents and expand into diverse use cases across industries, the need for high-performance, low-latency voice interaction models is anticipated to rise.
