A new player has entered the booming AI voice industry and it’s already making headlines. Gradium, a Paris-based startup founded in September 2025, has officially emerged from stealth mode, backed by an unprecedented $70 million seed round. With this funding, Gradium positions itself as one of the most ambitious challengers in the fast-growing AI audio intelligence sector.
A New Category: Audio Language Models (ALMs)
Gradium’s core innovation is its development of Audio Language Models (ALMs) a new class of AI models built specifically for real-time voice interaction. Unlike existing LLM-based voice systems, which often suffer from latency and lack natural conversational flow, Gradium’s models are designed to deliver near-instant responses with human-like emotion and expression.
According to the company, traditional AI assistants struggle with the trade-off between quality and speed. Gradium’s ALMs aim to eliminate this barrier by being audio-native, trained directly on large-scale audio-text datasets. This approach enhances tasks such as speech recognition, voice synthesis, audio classification, and multimodal dialogue outperforming repurposed text-based models.
Vision and Leadership
The startup is led by Neil Zeghidour, a well-known figure in AI audio research and a founding member of Kyutai. Before launching Gradium, Zeghidour researched speech and voice systems at Google DeepMind.
Gradium spun out of the French nonprofit research lab Kyutai, backed by telecom billionaire Xavier Niel, who also participated in the seed round.
Funding and Investors
The $70 million seed investment unusually large for such an early stage underscores the confidence surrounding the company. The round was led by FirstMark Capital and Eurazeo, with participation from:
- Xavier Niel
- DST Global Partners
- Eric Schmidt (Former Google CEO)
This financial backing gives Gradium a substantial runway to grow rapidly and compete globally.
Market Strategy and Use Cases
Gradium launched with multilingual support including English, French, Spanish, German, and Portuguese and is already generating revenue from early enterprise clients. Its first phase focuses on B2B API integrations, enabling companies to power gaming, customer service, healthcare, agents, and language learning platforms with ultra-realistic voice capabilities.
Competitive Landscape
Gradium will face competition from OpenAI, Anthropic, ElevenLabs, and open-source acceleration. However, the company’s narrow focus on low-latency high-fidelity voice may offer a strategic advantage in real-time voice computing one of the fastest-expanding segments of AI technology.
