Launched 1999 by Institut Mines-Telecom. Incubator for tech startups from students and researchers with programs like START@IMT and SCALE'Up; focuses on AI, big data, and VR with early-stage funding options.
Scope: Regional (Ile-de-France).
Paris-Saclay, France
Incubator
/france/incubator/imt-starter
PhysicsX applies machine learning and generative AI to advanced engineering simulation across aerospace, automotive, and energy systems. The platform accelerates design loops by reducing simulation time for fluid, thermal, and structural modeling, helping industrial teams move from concept to validation faster.
Overview: Respeecher is a cutting-edge AI voice synthesis startup that enables one person's voice to be transformed into the voice of another specific person, with uncanny realism. In other words, it's known for its voice cloning technology. Respeecher's system takes in an actor or user's speech and outputs audio that sounds like a target voice - be it a famous actor, a historical figure, or anyone for whom it has trained a model. This technology has huge applications in entertainment: filmmakers can revive voices of actors from years past, game developers can have characters speak in iconic voices, and creators can produce content where, say, a celebrity appears to narrate (with permission). What makes Respeecher stand out is the high fidelity of the cloned voices - they are nearly indistinguishable from the real person, capturing emotion and intonation accurately. Respeecher provides its tech via a software interface; typically, a voice model is trained on recordings of the target voice, and then a voice actor provides performance which is converted. Importantly, Respeecher emphasizes ethical use: it requires consent from the voice owners and has gained a reputation for an "ethics-first" approach in the emerging synthetic media field. Founding Story (2018): Respeecher was founded in February 2018 in Kyiv, Ukraine by Alex Serdiuk, Dmytro Bielievtsov, and Grant Reaber. Serdiuk and Bielievtsov are Ukrainian, and Reaber is an American speech technologist - the trio met through mutual research interests in speech processing. They realized that recent advances in deep learning could make high-quality voice conversion possible, and they set out to create a system far beyond the rudimentary voice changers of the past. In 2018-2019, they went through the Comcast NBCUniversal LIFT Labs Accelerator (powered by Techstars) in the US, which gave them industry connections. Early on, Respeecher got a lucky break: it was approached to work on a high-profile but secret project - which turned out to be Lucasfilm's The Mandalorian (Season 2 finale in 2020). The task was to recreate the voice of a young Luke Skywalker as he would have sounded in the 1980s, using Mark Hamill's old recordings. Respeecher delivered successfully, astonishing Hollywood sound engineers. This successful debut put Respeecher on the map in the film industry. In subsequent months, they also worked on a Super Bowl commercial voicing Vince Lombardi (AI-cloned from archival audio). The Ukrainian team, working under NDA, built these groundbreaking demos which then led to word-of-mouth referrals in Hollywood. Product and Use Cases: Respeecher's core product is delivered as a service to content creators. A client provides audio of the target voice (e.g., recordings of a person) and Respeecher trains a custom voice model. Then the client (often a voice actor or the person themselves) provides the new dialogue, which Respeecher's system converts into the target's voice. One notable use case was the documentary short film "In Event of Moon Disaster" - Respeecher recreated President Nixon's voice to simulate him reading a speech that was written in case Apollo 11 failed, and this project won an Emmy Award in 2021 for Outstanding Interactive Media (Respeecher's team was credited for the voice work). Another big use case: Darth Vader's voice in 2022's Obi-Wan Kenobi series - James Earl Jones, aged 91, could no longer perform with the same power, so Lucasfilm used Respeecher to generate Vader's lines using Jones' archival voice data. This was done with Jones's consent (he effectively authorized the AI recreation of his iconic voice). The result was so good that audiences thought Jones had recorded the lines himself. Outside of film/TV, Respeecher has been used in video games (e.g., to recreate voices of deceased voice actors so characters can return), and in music - in 2022, it enabled artist Aloe Blacc to perform a song where his voice was converted to sound like the late Avicii's voice, as a tribute. Researchers and archivists have also shown interest in using it to restore voices for people who lost theirs (like throat cancer patients), although that's still experimental. Traction and Achievements: Though a B2B company, Respeecher gained public fame because of the high-profile nature of its projects. By 2022, virtually every Star Wars fan and many industry professionals had heard of Respeecher due to media coverage of its involvement in resurrecting Luke Skywalker's youthful voice and Darth Vader's classic tones. The company has a relatively small number of clients (dozens of studios, game companies, etc.), but each project is high value. Its technology earned it an Engineering Emmy Award in 2021, and it won TechCrunch's Startup Battlefield (audience choice) at CES in 2020. In terms of funding, Respeecher raised about $1.5M in early 2020 (around the Techstars time), and as of 2021 it had raised over $3M in total from investors including Techstars, ff Venture Capital, and Acrobator Ventures. The team remained fairly lean (under 50 people) but comprised specialized AI researchers and sound engineers. Financially, Respeecher's revenue grew with each Hollywood contract; while not public, one can infer that by 2022 it was profitable or close, given the repeat work from Disney/Lucasfilm and others. One of Respeecher's achievements is also on the ethical front: in 2023 it was one of ten companies (alongside OpenAI and TikTok) to sign the Partnership on AI's framework for responsible media, committing to practices like obtaining consent, watermarking AI content, etc. This proactive stance has made them a trusted partner in an area often seen with suspicion (deepfakes). The war in Ukraine did pose a challenge - much of Respeecher's team was Kyiv-based, and during the Russian invasion in Feb 2022, the team amazingly continued working (the Obi-Wan Kenobi Vader voice project was completed in the spring of 2022 while Kyiv was under threat). This dedication further earned them respect in the industry. Key Partnerships: Respeecher's notable partnerships are with major studios like Disney/Lucasfilm, and with game studios such as CD Projekt Red (who used Respeecher in 2023 to voice a character in Cyberpunk 2077 in Polish, whose original actor had passed away). It also collaborated with the creative team of the documentary "Val" to recreate actor Val Kilmer's voice, which he lost due to cancer (Kilmer and his son provided the training data). Each successful partnership validates the tech and leads to more clients. On the academic side, Respeecher worked with medical research to explore giving people who lost their voice (e.g., ALS patients) a chance to speak in their own reconstructed voice - a heartwarming application of the tech. Future Outlook: Respeecher sits at the forefront of AI in media. The demand for voice cloning is likely to increase - whether to dubbing movies in a star's own voice across languages, bringing historical figures' voices to life for museums, or personal uses (with permission, someone might license their voice). The company's focus will be to maintain quality leadership as big players (like Microsoft or Google) also invest in voice AI. Respeecher's Ukrainian origin is a source of pride: it shows that despite adversity, Ukrainian tech can lead globally in innovation. The founders have expressed interest in eventually developing real-time voice conversion (currently it's not instant, it takes some processing time) and making the tech more accessible perhaps in creator tools. But they tread carefully to avoid misuse. In summary, Respeecher is a Ukrainian deep-tech startup that achieved what was once sci-fi - lending voices across time and space - and has done so responsibly. By powering some of Hollywood's most iconic moments with AI, it has cemented itself as a trailblazer in synthetic audio. As long as movies and games want to push creative boundaries, Respeecher's technology will likely be in the credits, helping the impossible become possible in voice. Sources: en.wikipedia.org, failory.com, techcrunch.com, youtube.com.
Bioptimus is the self-proclaimed "Mistral of Biology," spun out in early 2024 by a team of alumni from Google DeepMind and Owkin, including CEO Jean-Philippe Vert. The company is building what it calls the first universal foundation model for biology. Where large language models were trained on text to learn human language, Bioptimus trains on biological data—DNA sequences, protein structures, cellular imagery, and clinical phenotypes—to infer the underlying rules of life. The ambition is not just better predictions but a unified biological reasoning layer that can simulate outcomes before expensive lab work happens. By 2026, Bioptimus has moved from a research thesis to early commercial traction. After a $41M Series A in 2025, the company released its first commercial model that can predict how a specific molecule will interact with a human cell with unusually high accuracy. This enables in-silico screening that reduces wet-lab costs and improves hit rates for drug discovery. French pharma leaders such as Sanofi have begun using these models to prioritize compounds and compress early-stage discovery timelines. The product value is immediate: fewer failed experiments, faster candidate selection, and deeper mechanistic insight. Bioptimus' roadmap focuses on "multi-scale biology," connecting the micro level (genomics and proteomics) to the macro level (patient outcomes and clinical data). Rather than being limited to protein folding, the models aim to bridge across data modalities so a genetic mutation can be linked to disease pathways, tissue behavior, and potential therapeutic interventions. This requires data breadth and regulatory trust, and Bioptimus is building a defensive moat through European data sovereignty. It leverages partnerships with European research hospitals and biobanks to access high-quality patient data that is difficult for US competitors to acquire under GDPR constraints. That compliance burden becomes a competitive advantage: safer access, better provenance, and stronger alignment with European health data governance. The company remains deeply embedded in the Paris ecosystem. It was incubated inside Owkin before spinning out, maintains a presence around Station F, and benefits from the cross-pollination between French AI and biotech communities. Its investor base reflects that positioning: Sofinnova Partners, Bpifrance, Cathay Innovation, Xavier Niel, and Frst provide a blend of life-science expertise, sovereign capital, and deep-tech conviction. In 2026, Bioptimus is the clearest European bet that foundation models can unlock biology at scale—and a contender to become the default AI layer for drug discovery in Europe.
H (formerly Holistic) represents France’s second-wave AI builders: less focused on text generation and more focused on action. Founded in 2024 by a team that included former DeepMind researchers such as Charles Kantor and Laurent Sifre, the company drew attention with a seed round that eclipsed prior European records, signaling investor appetite for agentic AI. H’s core thesis is that the next leap is not bigger language models but AI systems that can plan, execute, and verify multi-step workflows across real software interfaces. The company frames this as “RPA 2.0,” using vision-language models to interact with screens and applications rather than brittle scripts and APIs. The firm’s early months were turbulent, including leadership turnover in late 2024, but the product roadmap accelerated. In November 2025 H launched Runner H, a vision-language agent that can see a desktop and operate mouse and keyboard actions to complete tasks such as booking travel, compiling invoices, or processing customer tickets. The release of Surfer-H-CLI soon after gave developers a programmable surface for building web agents, which helped H win mindshare with the European developer community. An acquisition of Mithril Security in 2025 signaled a focus on secure, private agent deployment — a crucial requirement for enterprise adoption, where the biggest fear is an autonomous agent mishandling data or credentials. H is deeply embedded in the Paris AI ecosystem, with early incubation support from Aglae Ventures and strong ties to the Paris AI Hub around Sorbonne and Ecole Polytechnique. Its backers read like a who’s-who of strategic supporters: Accel led financing, Eric Schmidt invested personally, and strategic relationships with Amazon and UiPath position H as a bridge between cloud infrastructure and automation workflows. French tech patrons Xavier Niel and Bernard Arnault provide both capital and political signal value. By early 2026, H is widely watched as a potential category leader in action-oriented AI, competing to become the default “application layer” for autonomous digital workers across Europe’s enterprises.
Founded in April 2023 by former Google DeepMind and Meta AI researchers, Mistral AI has become one of Europe's most valuable AI startups. The company builds frontier large language models such as Mistral Large and Mixtral and follows an open-weight plus enterprise deployment strategy. Its products are used via API, cloud partners such as Microsoft Azure, and local deployments for regulated organizations requiring data control and model transparency. Le Chat is the company's conversational interface built on its model stack.
Poolside is arguably the most important "import" to the European AI ecosystem in the last decade. Founded by Jason Warner (former CTO of GitHub) and Eiso Kant, the company started in the United States but relocated its headquarters to Paris in 2024 to tap France's deep talent pool in mathematics, systems engineering, and AI research. Its mission is unusually ambitious: not just to build code autocomplete, but to create the world's most capable AI for software engineering, with a long-term goal of systems that can build and evolve software autonomously. This framing has made Poolside a defining player in the AI-for-developers category. By 2026, Poolside has established itself as a core node in the "Paris AI Triangle" alongside Mistral and H. After a massive $500M Series B in late 2024 led by Bain Capital Ventures, the company launched its flagship foundation model trained on a distinct corpus of code, software architecture, and repository-level structure. The emphasis on engineering artifacts rather than general internet text allows Poolside's models to reason about large codebases, infer intent from architectural patterns, and propose higher-level refactors instead of one-off snippets. This technical posture differentiates Poolside from generalist models and makes it uniquely useful for enterprises grappling with decades of technical debt. The company's 2026 platform looks less like a plugin and more like a "shadow engineering team." Poolside can analyze legacy banking systems, plan migrations across stacks, and coordinate changes across modules while preserving compliance and auditability. That capability has made it a darling of large financial institutions such as HSBC and Citi, and it explains why strategic investors are embedded alongside venture capital. Poolside is also scaling Project Horizon, a 2-gigawatt AI data center campus in Texas in partnership with CoreWeave, while keeping its research and model development anchored in Paris. This split setup reflects a broader pattern: US-scale compute, European R&D depth. Poolside's ecosystem ties are deep. It is a central node at Station F and an active participant in the Paris AI Hub, and its relocation is often cited by the French government as a proof point for attracting foreign tech leadership. The investor roster underscores its strategic importance: Bain Capital Ventures led the growth round, DST Global joined as a late-stage backer, Xavier Niel provided early strategic support to facilitate the Paris move, and Nvidia is a hardware partner. In 2026, Poolside stands out as Europe's most credible attempt to build an AI-native software engineering stack at frontier scale.
Checking login status...
Loading comments...