Paris AI Voice Startup Gradium Raises $70M Seed Round from Eric Schmidt and Tech Giants

A Paris-based AI voice startup has emerged from stealth mode with one of the largest seed funding rounds in recent AI history. Gradium raised 70 million dollars from prominent investors including former Google CEO Eric Schmidt and French telecom billionaire Xavier Niel, just three months after its September 2025 founding.

The round was led by FirstMark Capital and Eurazeo, with participation from DST Global Partners and other high-profile investors. The massive seed investment signals growing investor conviction that voice will become the primary interface for AI systems, replacing text-based interactions that currently dominate generative AI applications.

Gradium has developed audio language models designed to deliver voice at scale with ultra-low latency—essentially, AI voices that respond almost instantly. This technology addresses one of the most significant barriers in current voice AI systems: the delay between user input and AI response that disrupts natural conversation flow.

Audio language models are specialized AI systems designed to process, understand, and generate natural language using audio-text data, representing the audio-native counterpart to large language models. Unlike text-based systems that convert speech to text, process it, then convert back to speech, Gradium's ALMs work directly with audio, enabling faster and more expressive interactions.

The founding team—Neil Zeghidour, Olivier Teboul, Laurent Mazaré, and Alexandre Défossez—previously held roles at Meta, Google DeepMind, Jane Street, and Google Brain. Their combined expertise represents what the company claims is one of the highest concentrations of generative audio talent assembled in a single startup.

Neil Zeghidour, Gradium's founder and CEO, explained that existing voice AI systems are "brittle, costly and unable to deliver truly natural interactions," with the goal to "make voice the primary interface between humans and machines."

Current voice AI systems face multiple challenges. Most rely on complex pipelines that convert speech to text, process the text through language models, then synthesize speech from the text output. This multi-step process introduces latency that makes conversations feel unnatural. Quality issues persist, with AI voices often sounding robotic or struggling with expressiveness. Scalability remains expensive, limiting deployment in cost-sensitive applications.

According to Zeghidour, ALMs can outperform LLMs in voice AI tasks including speech recognition and audio generation, using natural language supervision that replaces traditional labeling methods. This approach enables the models to learn complex relationships between sound and language more effectively.

The startup launched with multilingual support from day one, offering English, French, German, Spanish, and Portuguese, with additional languages planned. This positions Gradium advantageously against primarily English-focused competitors in the global market.

Gradium faces competition from established players and well-funded startups. OpenAI's voice capabilities in ChatGPT, Anthropic's voice features in Claude, and specialized companies like ElevenLabs, which has raised substantial funding for voice synthesis, all compete for developer adoption.

However, the market opportunity remains massive. As AI systems evolve from simple chatbots to autonomous agents capable of booking appointments, making purchases, and handling complex workflows, voice interactions will become essential. Typing or reading text responses won't scale for many real-world applications where hands-free, eyes-free interaction is necessary.

Gradium reports it began generating revenue within weeks of formation, offering access models from developer usage to enterprise-scale deployment. This rapid path to revenue likely appealed to investors increasingly focused on AI startups demonstrating commercial traction quickly rather than pursuing years of research before monetization.

The company maintains an ongoing collaboration with Kyutai, the nonprofit French AI research lab where the technology originated, providing continued access to research outcomes in generative audio. This connection offers Gradium a pipeline of academic breakthroughs that can be rapidly commercialized.

Gradium's Paris location positions it within Europe's growing AI ecosystem, which has produced several significant AI companies including Mistral AI and Hugging Face. European AI startups often emphasize multilingual capabilities and data privacy considerations from inception, potentially offering advantages in global markets beyond English-speaking countries.

Xavier Niel, the French telecom billionaire backing both Kyutai and Gradium, has emerged as a major force in European AI development. His involvement provides not just capital but also strategic guidance and network effects across the French tech ecosystem.

The team's background at major AI labs like Google DeepMind and Meta provides technical credibility that attracted top-tier investors. Building cutting-edge voice models requires deep expertise in audio processing, machine learning, and large-scale systems engineering—skills the founding team developed through years at frontier AI research organizations.

Gradium sees 2026 as the horizon for tackling deeper technological limitations in voice AI. The company's immediate focus involves refining models for production use, expanding language support, building developer tools and APIs, and scaling infrastructure to handle growing demand.

The funding round's size reflects investor belief that voice will become the dominant AI interface. As autonomous AI agents become more capable, voice provides the most natural way for humans to interact with these systems. Whether booking travel, managing schedules, controlling smart homes, or accessing information, voice offers advantages over typing for many scenarios.

For developers, Gradium's technology promises to enable more sophisticated voice applications without the engineering complexity of current solutions. Instead of stitching together multiple services for speech recognition, language processing, and synthesis, developers could access end-to-end voice capabilities through unified APIs.

The $70 million seed round positions Gradium with substantial runway to compete against well-funded rivals while continuing research and development. In an AI market where compute costs and talent expenses run high, significant capital provides breathing room to perfect technology before needing additional fundraising.

Gradium's emergence with such substantial backing validates the strategic importance of voice as AI's next interface frontier. The technology has moved beyond simple voice assistants executing commands to sophisticated systems capable of natural conversation, real-time information synthesis, and complex task execution.

The competitive dynamics will likely intensify as established players recognize voice's importance. OpenAI, Anthropic, Google, and others will continue improving their voice capabilities, while specialized startups like Gradium push technological boundaries in specific areas like latency and expressiveness.

For consumers and businesses, this competition should accelerate improvements in voice AI quality, availability, and affordability. Applications that seemed futuristic even a year ago—truly conversational AI assistants, real-time translation, personalized voice interfaces—are becoming technically and economically feasible.

The next 12 to 18 months will be critical in determining which companies establish leadership in voice AI. Gradium's combination of technical talent, strategic funding, and focused mission positions it as a serious contender in this rapidly evolving market.