Gradium raises $100M for voice AI
Nvidia-backed Gradium, spun from Kyutai in Paris, raised $100M seed on July 9 to build ultra-low latency audio for AI agents. Renault is an early customer.
Gradium, a voice AI startup based in Paris, closed a $100 million seed round on July 9. Nvidia led alongside FirstMark Capital, Eurazeo, DST Global Partners, Eric Schmidt, and Xavier Niel. The round extended an initial $70 million seed from December 2024 by an additional $30 million.
Gradium was spun out of Kyutai, the French AI research lab. Co-founder Neil Zeghidour previously worked at Google Brain, DeepMind, and Facebook. The company’s product is voice AI models designed to eliminate perceptible delays in AI agent conversations through ultra-low audio latency. Renault is listed as an early customer since the December launch. Gradium is opening a Bay Area office to compete for talent near the major AI companies.
The company competes with ElevenLabs (valued at $11 billion as of its last round) and against Google Gemini in the live-voice space.
What it means for you
The latency problem in AI voice is well documented. Users will tolerate latency in chat that they will not tolerate in a voice call. Gradium is specifically building on that gap, and the Nvidia backing suggests the compute supply chain for low-latency inference is part of the solution, not just the model architecture.
The xAI Voice Agent Builder launched earlier at $2 per million input tokens and $10 per million output tokens. How Gradium’s audio infrastructure fits alongside or underneath those tools is not yet published, but the Renault deployment suggests it is already in production use at scale.