Remanence, a Paris startup founded by three former leaders of French AI company H, has raised a $30 million seed round to help businesses train their own specialised AI models on top of open-weight models. EQT Ventures and FirstMark co-led the round, announced on 17 November 2026.
Quick Take
- Round: $30 million seed, co-led by EQT Ventures and FirstMark.
- Other investors: Zero Prime, Emblem, Drysdale, Global Founders Capital and Yellow; angels include Datadog CEO Olivier Pomel and Synthesia co-founders Victor Riparbelli and Steffen Tjerrild.
- Product: turns a company’s AI agent execution data into training environments and post-trains open-weight models with reinforcement learning.
- Ownership: the customer owns the model weights, and its data does not have to leave its infrastructure, according to EQT.
- Use of funds: a dedicated GPU cluster in Europe, hiring in Paris, London and San Francisco, and design partnerships in Europe and the US.
In plain English
Many companies now run AI agents on rented models from big US labs and pay for every token. Each time an agent works, it leaves a trace of what it tried and where it failed, and most of that is thrown away. Remanence wants to use those traces to train a company’s own model, based on a free open-weight model, so it gets better at that company’s specific work and costs less to run.
What happened
Remanence announced the round with backing from EQT Ventures and FirstMark, as reported by Tech.eu and Sifted on 17 November. CEO Pierre-Louis Cedoz said: “We have seen first-hand how powerful post-training can be, as well as how difficult it remains for most companies to access. We started Remanence to close that gap and help businesses turn their own agent data into specialised models they control.” EQT Ventures partner Tom Mendoza said: “Every company running agents is generating its most valuable training data and throwing it away.”
Background
Open-weight models have narrowed the gap with closed frontier models, and companies such as Cursor and Ramp already train their own models with reinforcement learning on their usage, EQT notes. The hard part is building realistic test environments where agents can practise safely and be scored. Remanence’s founders led post-training of H Company’s open-weight Holo models: Cedoz was H’s founding researcher and later CTO after working on Gemini post-training at Google DeepMind, Hamza Benchekroun is CTO and Axel Moyal is CPO.
How it works
- Learn from production: the platform analyses where agents succeed and fail in real runs.
- Generate environments: it automatically builds training sandboxes and evaluation checks, including scenarios agents have not yet met.
- Post-train: reinforcement learning rewards successful behaviour and updates the open-weight model’s parameters.
- Retrain over time: as traces accumulate or stronger open models ship, the model can be retrained on the same environments.
Why it matters
For companies running agents at scale, falling token prices have not stopped total AI bills from rising, because agents use so many tokens per task. Owning a smaller, tuned model is a way out.
For Europe’s AI scene, the round is another bet on reducing dependence on proprietary US models, a theme also visible in sovereign-model launches.
For developers, automated environment building could make reinforcement learning available to teams without post-training specialists, of which EQT says there are only a few hundred worldwide.
What’s still open
- Named customers and results from the design partnerships.
- How tuned models compare on cost and quality with frontier APIs over time.
- Pricing, and how much of the training stack runs on the customer’s own infrastructure.
Sources
- Tech.eu, “Synthesia and Datadog CEOs back AI startup Remanence in $30M funding round,” 17 November 2026
- EQT Ventures, “Own, Don’t Rent AI: Why We Co-Led Remanence’s $30M Seed Round,” 17 November 2026
- Sifted, report on Remanence’s seed round, 17 November 2026
i10x Perspective
The value in enterprise AI is shifting from the model to the loop around it. Remanence is betting that a company’s own agent traces are worth more than any rented model, which fits the rise of open weights. More rounds are tracked in the AI Funding Index and our Funding & M&A section.



