Gradium Emerges from Stealth with $70M to Revolutionize Voice AI with Ultra-Low Latency Models

Gradium's Impressive Market Entry

Gradium, a Paris-based AI voice startup, has emerged from stealth mode with one of the most substantial seed funding rounds in recent memory. The company secured $70 million in seed funding led by FirstMark Capital and Eurazeo, with participation from prominent investors including French telecom billionaire Xavier Niel, DST Global Partners, and former Google CEO Eric Schmidt 1

1

. Remarkably, this funding was raised just three months after the company's founding in September 2025.

Revolutionary Audio Language Models

The startup has developed what it calls Audio Language Models (ALMs), specialized AI systems designed to process, understand, and generate natural language using audio-text data. According to CEO Neil Zeghidour, ALMs represent the "audio-native counterpart" to large language models and are engineered to support more natural and expressive voice interactions with dramatically lower latency 2

2

. The technology was initially developed during the founders' time at Kyutai, a nonprofit AI research lab backed by Xavier Niel.

Unlike traditional voice AI systems that rely on cascaded architectures, Gradium's ALMs are trained on datasets that pair audio with descriptive text, enabling them to learn complex relationships between sound and language. This approach uses natural language as a "supervision signal," allowing the models to perform tasks such as audio classification and speech synthesis more effectively than general-purpose language models 2

2

.

Addressing Current Voice AI Limitations

Zeghidour has identified significant shortcomings in existing voice AI systems, describing them as "brittle, costly and unable to deliver truly natural interactions" 2

2

. Current voice assistants suffer from issues including interrupting users mid-sentence, misjudging when someone has finished speaking, and responding with inappropriate emotional tones 3

3

.

The company's solution focuses on four key pillars: accuracy, latency, conversational flow, and expressive synthesis. Gradium's approach aims to eliminate the traditional tradeoff between quality and scalability by combining ultra-realistic expressivity, accurate transcription, and ultra-low-latency interactions at an accessible price point 2

2

.

Competitive Landscape and Market Strategy

Gradium enters a highly competitive market dominated by frontier LLM companies like OpenAI, Anthropic, Meta, and Mistral, all of which offer voice and multimodal capabilities. The startup also faces competition from well-funded voice AI companies like ElevenLabs and hundreds of voice models available on platforms like Hugging Face 1

1

.

Despite this competition, Gradium has demonstrated rapid commercial traction, generating revenue within six weeks of launch while their models were still in training 3

3

. The company's go-to-market strategy is unapologetically B2B, targeting customers building voice agents for customer support, medical appointments, coaching platforms, e-learning tools, and enterprise workflows.

Technical Innovation and Team Expertise

The startup boasts what it claims is one of the industry's highest concentrations of generative audio expertise, with a team comprised of researchers and engineers from Google DeepMind, Meta's FAIR research team, and Jane Street Capital 2

2

. Zeghidour himself previously worked with voice models as a researcher at Google DeepMind before becoming a founding member of Kyutai.

Gradium has launched its platform with multilingual support for English, French, German, Spanish, and Portuguese, with additional languages planned. The company offers flexible pricing plans designed to serve everyone from small developer teams to large enterprises, and maintains ongoing collaboration with Kyutai to ensure access to cutting-edge generative audio research 2

2

.