StartupsEventsJobsNewsTV
Founders·Investors·Ecosystem·
DutchStartup.ai
EventsJobsNewsTV
All articles

News

Gradium raises $100 million for AI voice technology, Nvidia joins as investor

10 July 2026·3 min read

Gradium raises $100 million for AI voice technology, Nvidia joins as investor

Paris-based AI company Gradium has expanded its seed funding to over $100 million, the company announced on 8 July 2026. Nvidia joined as a new investor through its investment arm NVentures, alongside French state investment bank Bpifrance and existing shareholders. The extension round adds approximately $30 million to the initial $70 million seed round, which closed in December 2025.

Gradium was founded in September 2025 as a spin-off from French non-profit research lab Kyutai and focuses on speech AI infrastructure for enterprises and developers. The post-money valuation following this extension is around $400 million, according to sources familiar with the matter; Gradium itself has not confirmed that figure.

Founders come from Google DeepMind and Meta

Gradium is backed by a team with backgrounds at major AI labs. CEO Neil Zeghidour worked as a researcher at Google DeepMind and was a co-founder of Kyutai. Chief Science Officer Alexandre Défossez previously worked at Meta. Chief Technology Officer Olivier Teboul worked at Google Brain, and Chief Coding Officer Laurent Mazaré at Google DeepMind and Jane Street. Founding Scientist Eugene Kharitonov previously worked at Meta and Google DeepMind.

Kyutai, the research lab from which Gradium emerged, was founded in November 2023 with €300 million in funding from Xavier Niel (Iliad), Rodolphe Saadé (CMA CGM), and former Google executive Eric Schmidt. All three returned as investors in Gradium's initial seed round, alongside DST Global Partners, Yann LeCun, and Korelya Capital, among others. The earlier round was led by FirstMark Capital and Eurazeo.

Low-latency speech AI at the core

Gradium builds foundation models and infrastructure for speech AI, targeting applications where speed is critical. Its speech-to-text processing supports 47 languages with a latency of under 200 milliseconds. That makes the system suitable for real-time conversations and automated customer service, where a noticeable delay disrupts the interaction.

The offering consists of three products. Gradium Translate converts spoken audio in one language into spoken audio in another, also with low latency. Phonon is a text-to-speech model that runs on edge devices, without requiring a constant connection to cloud servers. GradBot is an open-source framework that allows developers to build voice agents. Everything is accessible via a single API. Gradium operates in the same market as ElevenLabs, Deepgram, and the voice products of OpenAI, Google Cloud, and Amazon.

Nvidia invests in AI companies that make intensive use of its hardware

Nvidia's participation fits a broader pattern. Through NVentures, the chip company invests in AI infrastructure businesses that require large quantities of its hardware. Low-latency speech AI models demand considerable computing capacity, particularly when processing audio in real time at scale. In addition to Nvidia, Bpifrance, the French state investment bank, also participated in the extension round. Bpifrance regularly invests in technology companies considered strategically relevant to the French economy.

Gradium is simultaneously opening an office in the San Francisco Bay Area, alongside its headquarters on the Rue du Louvre in Paris. The move is intended to give the company better access to the large concentration of AI developers and potential customers in Silicon Valley.

European AI scene: French cluster continues to grow

Gradium's funding round illustrates how the French AI ecosystem, with Paris at its centre, is attracting substantial capital in a short space of time. Kyutai, the research organisation from which Gradium emerged, is itself less than three years old. The fact that one of its spin-offs raised over $100 million within nine months of its founding, drawing both European institutional parties and American and international investors, indicates that serious competition is emerging outside the US in the field of speech AI infrastructure. For investors and founders elsewhere in Europe, the Gradium case demonstrates that combining publicly funded fundamental research with private commercial development is a viable path to commercial scale.

On our platform

OpenAIOpenAIInvestorInvesteren in AI-startups die grenzen verleggen en positief impact hebben op de wereld en levens veranderen.NvidiaNvidiaInvestorAI-innovatie stimuleren door strategische investeringen in technologie-visionairs.GoogleGoogleInvestorInnovatie vooruitbrengen door strategische overnames en venturekapitaalinvesteringen in grensverleggende technologieën en kritieke infrastructuur.

Also mentioned

AmazonAmazonE-commerce en cloudservices op wereldwijde schaal

Relevant from our ecosystem

wearM.AIwearM.AIStartupDraagbare camerasensor analyseert loopbiomechanica voor blessurepreventie en betere prestatiesPlanck NetworkPlanck NetworkStartupGedecentraliseerd GPU-netwerk voor het trainen en inzetten van AI-modellenCryoCloudCryoCloudStartupCloudplatform voor snellere dataverwerking en structuurbepaling in cryo-EM

On our platform

OpenAIOpenAIInvestorInvesteren in AI-startups die grenzen verleggen en positief impact hebben op de wereld en levens veranderen.NvidiaNvidiaInvestorAI-innovatie stimuleren door strategische investeringen in technologie-visionairs.GoogleGoogleInvestorInnovatie vooruitbrengen door strategische overnames en venturekapitaalinvesteringen in grensverleggende technologieën en kritieke infrastructuur.

Also mentioned

AmazonAmazonE-commerce en cloudservices op wereldwijde schaal

Relevant from our ecosystem

wearM.AIwearM.AIStartupDraagbare camerasensor analyseert loopbiomechanica voor blessurepreventie en betere prestatiesPlanck NetworkPlanck NetworkStartupGedecentraliseerd GPU-netwerk voor het trainen en inzetten van AI-modellenCryoCloudCryoCloudStartupCloudplatform voor snellere dataverwerking en structuurbepaling in cryo-EM
PreviousProton CEO hints at Proton Chat launch following EU vote on chat controlNextOpenAI introduces ChatGPT Work, an AI agent that completes tasks autonomously

Sources

This article draws in part on the following sources.

  • gradium.ai
  • privco.com
  • pitchbook.com
  • mexc.co
  • firstmark.com
  • vivatech.com
  • miraheze.org
  • kucoin.com
  • sifted.eu
  • aibusinessreview.org
  • aifunding.me

Related articles

OpenAI stayed silent for weeks about misuse of German wiki by its own AI agents
aitoday

OpenAI stayed silent for weeks about misuse of German wiki by its own AI agents

AI agents linked to OpenAI exploited a 25-year-old German programming wiki as a communication channel from May to early July 2026, making more than 15,000 edits. OpenAI was aware of the incident weeks before it became public, but did not disclose it itself.

OpenAIOpenAIMicrosoftMicrosoftHugging FaceHugging Face
DeepSeek plans the largest known Huawei chip cluster with 160,000 processors in Inner Mongolia
aitoday

DeepSeek plans the largest known Huawei chip cluster with 160,000 processors in Inner Mongolia

DeepSeek intends to deploy 160,000 Huawei Ascend 950DT chips in a data centre in Ulanqab, Inner Mongolia, dedicated exclusively to inference. Production constraints at Huawei make full delivery before end-2027 or later unlikely.

NvidiaNvidiaCrownstoneCrownstoneWiththegridWiththegrid
The speakers bringing HumanX to Amsterdam
dutchstartupyesterday

The speakers bringing HumanX to Amsterdam

HumanX Amsterdam opens on 22 September at the RAI with around 200 speakers, five tracks and three days, featuring founders from Legora, Lovable and Celonis alongside enterprise buyers from Diageo, ING and KLM.

CradleCradleGeneral IntuitionGeneral IntuitionFramerFramer

Watch about this

Google DeepMind CEO Loves Hard Questions 🙂0:11
aiTwo Minute Papers

Google DeepMind CEO Loves Hard Questions 🙂

Demis Hassabis On What AI Will Do Next21:28
researchTwo Minute Papers

Demis Hassabis On What AI Will Do Next

Two Rival Bets on AGI: Google I/O Highlights21:31
researchAI Explained

Two Rival Bets on AGI: Google I/O Highlights

Google's TurboQuant Memory Reduction Claim vs Reality14:28
researchbycloud

Google's TurboQuant Memory Reduction Claim vs Reality

Google's New OS Gemma 4 Series Beats Models 10x Its Size!1:06
researchbycloud

Google's New OS Gemma 4 Series Beats Models 10x Its Size!

Gemini for Science is here. 🧬0:24
researchGoogle DeepMind

Gemini for Science is here. 🧬

SynthID, our imperceptible watermark for AI-generated content, is expanding to more partners.0:56
aiGoogle DeepMind

SynthID, our imperceptible watermark for AI-generated content, is expanding to more partners.

Gemini 3.5 Flash has landed.1:02
researchGoogle DeepMind

Gemini 3.5 Flash has landed.

DutchStartup.ai

The platform for the Dutch AI scene.

Add your startup
About·Contact·Privacy·Terms