- Gradium, the Paris-based voice AI startup spun out of research lab Kyutai, has extended its seed round to $100 million with NVIDIA joining as a new investor, seven months after the company launched from stealth.
- The four co-founders collectively built EnCodec, SoundStream, and Moshi — the foundational research papers that underpin most modern voice AI systems, including those built by OpenAI.
- The global AI voice generator market is valued at $7.7 billion in 2026 and is projected to reach $20.7 billion by 2031, at a CAGR of 30.7%, according to MarketsandMarkets.
The people who invented the plumbing that powers your AI voice assistant just raised $100 million to compete with the companies that profited from it. That tension, researchers who built the field are now trying to commercialise it, is what makes Gradium different from every other voice AI startup chasing the same enterprise dollar.
Earlier this week, the Paris-based company announced an approximately $30 million extension to its seed round, bringing total funding to more than $100 million. NVIDIA joined as the headline new investor alongside undisclosed others.
The extension followed a $70 million first tranche raised in December 2025, led by FirstMark Capital and Eurazeo, with participation from DST Global Partners, French billionaires Xavier Niel and Rodolphe Saadé, and former Google CEO Eric Schmidt. Gradium has not disclosed its current valuation.
Who built this and why it matters
Neil Zeghidour, chief executive, previously worked on audio AI research at Meta and Google DeepMind. Alexandre Défossez, chief science officer, co-created EnCodec at Meta’s FAIR lab, a neural audio codec that became a compression standard for modern voice AI systems. Laurent Mazaré, chief coding officer, spent years at DeepMind and Jane Street. Olivier Teboul, chief technology officer, came from Google Brain.
Together, they co-authored Moshi in 2024 at Kyutai, the first full-duplex spoken dialogue model that processes speech directly rather than converting it to text and back, with 200 milliseconds of latency.
Gradium was spun out of Kyutai in September 2025 as its commercial arm, with Kyutai retaining a shareholding. The company is headquartered in Paris and is opening a San Francisco Bay Area office; it has not disclosed headcount.
What the product actually does
Gradium builds real-time voice infrastructure for developers and enterprises: streaming speech-to-text, expressive text-to-speech, semantic turn detection that determines when a user has genuinely finished speaking, on-device models that run without cloud connectivity, and Gradbot, an open-source framework for building production voice agents.
Its newest product, Gradium Translate, converts speech to speech across five languages in real time. For example, Renault uses the platform for customer-facing voice applications. Gradium says it generated revenue within weeks of launch.
“Voice AI is reaching an inflexion point. Surpassing $100 million in funding and expanding our investor base marks an important milestone for Gradium. It enables us to accelerate our roadmap, expand our Bay Area presence, and bring years of breakthrough research into products used by developers and enterprises around the world,” says Zeghidour.
The competitive field
ElevenLabs, currently in talks for a secondary sale at a $22 billion valuation after raising $500 million at an $11 billion valuation in February, dominates brand recognition and enterprise relationships in voice synthesis, with a focus on the full audio stack, including cloning, dubbing and agents.
LiveKit reached unicorn status on $100 million from Index Ventures, building real-time voice infrastructure for OpenAI and Tesla, the closest structural comparison to Gradium’s positioning. Deepgram closed a $130 million Series C at a $1.3 billion valuation, focused on speech recognition infrastructure for enterprise.
Where Gradium differs is in research depth: its founders did not study these problems from the outside; they defined the benchmarks competitors still measure themselves against.
NVIDIA’s involvement is the sharpest signal in the round. The chip maker has now backed both Gradium and ElevenLabs’ Series D, a strategy of owning compute relationships across the voice AI stack rather than picking one winner. FirstMark Capital, Gradium’s lead investor, has previously backed Pinterest, Shopify and Discord.
The Bay Area question
The $30 million is not the most revealing detail in the announcement. The San Francisco office is. For a lab that built Kyutai as an explicit open-source European alternative to Silicon Valley AI, opening a Bay Area outpost is a concession, one Gradium’s own press release frames as “strengthening its position at the heart of the world’s leading AI ecosystem.”
European voice AI startups raised €536 million in the first half of 2026 alone, nearly 50% above the same period in 2025. The infrastructure exists in Paris. The enterprise relationships still concentrate elsewhere.
The AI voice generator market is projected to grow from $4.16 billion in 2025 to $20.71 billion by 2031, a compound annual growth rate of 30.7%, according to MarketsandMarkets.
The researchers who wrote the papers now have the capital to compete with the companies that read them. Whether the research advantage holds once the sales motion becomes the deciding variable is the question the San Francisco office is designed to answer.
