On July 9, 2026, Paris-based Gradium said it had reopened its seed round to a total of about $100 million, with Nvidia leading a new tranche. It is an expansion, not a fresh round: the total builds on the $70 million Gradium raised when it emerged from stealth in December 2025, making Nvidia's arrival the headline.
Who is behind it
Gradium spun out of the French research lab Kyutai, and its founding team is a roster of people who wrote the underlying science of modern voice AI. CEO Neil Zeghidour (ex-Google and Meta), chief science officer Alexandre Défossez (a co-creator of Meta's EnCodec), Laurent Mazaré and CTO Olivier Teboul co-authored work including SoundStream, EnCodec and the full-duplex Moshi model — components other companies' voice stacks are built on.
What it builds
Gradium develops foundational audio models aimed at ultra-low-latency, real-time voice agents that respond almost instantly. Its early products include a real-time speech-to-speech translation system, an on-device text-to-speech engine that runs offline, and an open framework for building voice agents; French automaker Renault is a named customer. "There are less than a dozen players capable of training these models at scale," Zeghidour has argued — the scarcity that underpins the raise.
Why Nvidia
Nvidia rarely leads seed-stage financings, so anchoring Gradium's extension — atop existing backers FirstMark Capital, Eurazeo, DST Global Partners, Eric Schmidt and Xavier Niel — reads as a strategic signal as much as a check, pairing capital with GPU leverage. Gradium did not disclose a valuation.
The Bay Area pull
Zeghidour said the money lets Gradium "accelerate our roadmap, expand our Bay Area presence, and bring years of breakthrough research into products." That destination is telling: a French sovereign-AI champion, funded partly by French billionaires, is using the cash to open a San Francisco office and compete for talent near U.S. model labs — even as it positions itself against incumbents like ElevenLabs and OpenAI.
