Nvidia has backed French voice AI startup Gradium in a $30 million seed extension, pushing the company’s total seed round past the $100 million mark just seven months after its launch.
The extension included the US chip giant as a new investor, though Gradium declined to name other participants or share its latest valuation. It comes on top of a $70 million first tranche the company secured at the end of 2025, led by US venture firm Firstmark and French investor Eurazeo.
Gradium’s investor lineup already reads like a who’s who of tech royalty: French billionaires Xavier Niel and Rodolphe Saadé, former Google CEO Eric Schmidt, and now Nvidia — a signal that the chipmaker sees voice AI as a strategic growth layer rather than just a niche.
The startup spun out of Kyutai, a Paris-based non-profit AI research lab launched in 2023 with €300 million in backing from Niel, Saadé, and Schmidt. Kyutai focuses on open-source AI research and has centered much of its work on voice AI.
Gradium builds tools for developers to create voice applications — real-time text-to-speech, speech-to-text, voice translation, and models optimized for edge devices like laptops and phones. It also offers an open-source framework for building voice agents. Customers range from healthcare (AI medical secretaries) to gaming (voice-powered video game characters).
The fresh capital will fund a new San Francisco office and accelerate its research and product development. Gradium cofounder Neil Zeghidour says the company is publishing new models “almost every month,” compared to development cycles of six to twelve months for competitors.
The voice AI sector is heating up fast. London-based ElevenLabs is reportedly targeting a $22 billion valuation on the secondaries market, while OpenAI continues building its own voice capabilities. According to Sifted data, European AI-native voice startups raised €536 million in the first half of 2026 — nearly 50% more than the €360 million they raised in the same period in 2025.
Zeghidour notes that while thousands of startups and enterprises use voice models, fewer than a dozen players can train these models at scale — meaning the competitive pressure is intense but concentrated.
Source: Sifted
