Gradium Secures $70 Million Seed Investment for Ultra-Low Latency Voice AI

Paris-based AI voice technology developer Gradium has emerged from stealth mode announcing a $70 million seed funding round led by FirstMark Capital and Eurazeo.

This substantial initial investment underscores strong investor belief in the necessity for rapid, high-fidelity voice models within the evolving generative AI ecosystem. The company, which was spun out of the French AI research lab Kyutai, focuses specifically on developing audio language models engineered for near-instantaneous response times at scale. Gradium’s core proposition is reducing latency to near zero, a critical factor as conversational AI moves toward real-time agent interactions across various enterprise applications.

Neil Zeghidour, a founding member of Kyutai and former researcher at Google DeepMind, established Gradium in September 2025 to address developer needs for faster and more accurate voice synthesis capabilities.

As a European entity, Gradium launched with immediate multilingual support encompassing English, French, German, Spanish, and Portuguese, with plans to expand its language offerings.

The investment syndicate included participation from Xavier Niel, DST Global Partners, and Eric Schmidt, demonstrating broad institutional and individual backing for the European AI sector.

This financial backing positions Gradium to aggressively pursue technological development against established global players in the synthetic voice arena. Gradium enters a crowded field where frontier large language model developers such as OpenAI, Anthropic, and Mistral already integrate voice and speech recognition features into their broader multimodal offerings.

Furthermore, specialized startups, including the highly funded ElevenLabs, compete directly for developer mindshare in high-quality voice generation.

The market demand for nuanced, rapid voice interaction is projected to accelerate as AI systems transition from simple text interfaces to autonomous digital agents across entertainment and professional workflows.

Consequently, technological differentiation in speed and expressive accuracy will become a key determinant of platform adoption. The company intends to utilize the new capital to enhance the speed and accuracy benchmarks of its proprietary audio models, pushing the boundaries of what developers expect from real-time conversational AI tools. Success in this niche could allow Gradium to capture substantial market share in high-stakes, low-tolerance latency environments.