A Paris startup called Gradium just raised $100 million in seed funding, with Nvidia joining as a new backer. That number is worth pausing on: $100 million is a massive amount for a company at the earliest stage of its life, founded only in September 2025.
The company builds AI voice technology. Specifically, it builds the systems that let software pick up a phone call, listen to a person, and respond in near real time, with no awkward pause. That pause is the current weak point of most AI phone systems. Gradium's whole reason for existing is to eliminate it.
The founding team comes with serious research credentials. Gradium was spun out of Kyutai, a Paris-based non-profit AI research lab launched in 2023 with 300 million euros from Xavier Niel, Rodolphe Saadé, and Eric Schmidt. Kyutai published its voice research openly, and Gradium's job is to turn that research into production-ready products that businesses can actually buy and use. The company's co-founder previously led voice research at Google DeepMind and worked at Google Brain and Facebook.
Gradium's product line covers real-time speech-to-text, text-to-speech, voice-based translation, and models light enough to run on laptops and phones. It also released an open-source toolkit for developers who want to build their own voice applications on top of its technology. The company says it already has enterprise customers across healthcare and customer service, with Renault as its most publicly named client.
The market Gradium is entering is moving very fast. European voice AI startups raised 536 million euros in the first half of 2026 alone, nearly 50 percent more than the same period the year before. The market leader, ElevenLabs, closed 2025 at $330 million in annual revenue and was valued at $11 billion in February 2026. It is reportedly now in talks for a secondary share sale that would value it at $22 billion. A rival, Deepgram, raised $130 million at a $1.3 billion valuation in January.
None of this is abstract. Gartner forecasts that AI voice systems will reduce contact center labor costs by $80 billion globally in 2026. An automated AI phone interaction costs roughly $0.40 per call, compared to $7 to $12 for a human agent. Companies deploying these systems report 35 percent reductions in call handling time and wait time drops of up to 50 percent. Production deployments of voice AI agents grew 340 percent year over year across more than 500 organizations surveyed.
For any organization that runs a customer-facing phone operation, this is no longer a future consideration. More than two-thirds of Fortune 500 companies are already running live voice AI systems. The question has shifted from whether to try it to how quickly to move.
For Gradium specifically, the Nvidia investment carries a practical signal beyond the money. Nvidia backs the startups that run on its hardware, and its involvement tells you something about where Gradium sits in the technical stack. The company's decision to open a San Francisco office alongside its Paris headquarters is also telling: even with a world-class European research pedigree, it is going where the talent competition is.
Gradium's real advantage, if it holds, is speed. Its co-founder says the company publishes new models almost every month, while competitors typically need six to twelve months per release cycle. In a market where the quality gap between providers is closing fast, shipping faster and staying close to enterprise customers may matter more than any single technical edge.