Remanence raises $30M to build custom AI models for enterprises
The Paris-based startup secured seed funding to help companies move away from expensive proprietary models toward cheaper, specialized open-weight alternatives.
The short version
- Remanence raised $30M in seed funding co-led by EQT Ventures and FirstMark.
- The company uses open-weight models to help businesses create specialized AI systems for specific tasks.
- Founders include former executives from Google DeepMind, Meta, and the French AI startup H.
- The startup uses simulated enterprise environments and reinforcement learning to train AI agents.
- Angel investors include the CEOs of Datadog and Synthesia.
Why it matters
This reflects a growing trend where companies seek to reduce costs and increase control by moving away from closed models like OpenAI's. It signals a shift toward 'proprietary intelligence' that companies can own and refine with their own data.
Closing the accessibility gap
Pierre-Louis Cedoz, the co-founder and CEO of Remanence, noted that post-training is incredibly powerful but difficult for most companies to access. The startup aims to bridge this gap by providing the tools necessary to turn internal data into specialized models. This allows companies to maintain control over their own intelligence rather than relying on third-party providers.
By focusing on customization, Remanence allows businesses to create models that are specifically tuned to their unique workflows. This approach is intended to be more cost-effective than using general-purpose models for every single task. It provides a path for companies to build systems that are deeply integrated with their specific business logic.
The ultimate ambition of the startup is to give every company the ability to build their own proprietary intelligence. This intelligence is designed to improve over time as it is used. By owning the model and the training process, companies can create a lasting advantage that is unique to their organization.
The future of enterprise AI
The rise of Remanence reflects a broader trend in the AI industry toward decentralization and specialization. As the costs of using massive, general-purpose models become more apparent, the demand for efficient, niche models is expected to grow. This shift favors companies that can provide the infrastructure for fine-tuning and post-training.
For businesses, this means a move toward more reliable performance on complex, multistep tasks. Instead of a model that knows a little bit about everything, companies can deploy a model that knows a lot about their specific industry. This transition is driven by the need for both cost efficiency and data privacy.
As more companies explore open-weight technology, the role of specialized startups like Remanence will likely expand. They provide the bridge between raw, open-source models and the polished, reliable tools required for enterprise-grade operations. This evolution marks a significant step in how organizations plan to integrate AI into their long-term strategies.
Strategic investment and backing
The seed funding round for Remanence was co-led by EQT Ventures and FirstMark. The investment also saw participation from several other firms, including Zero Prime, Emblem, Drysdale, Global Founders Capital, and Yellow.
High-profile angel investors also joined the round to support the startup. These include the CEO of Datadog, Olivier Pomel, and the co-founder of Pigment, Romain Niccoli. Additionally, the co-founders of Synthesia, Victor Riparbelli and Steffen Tjerrild, provided backing.
Training in simulated environments
Remanence uses a specific technical approach to train AI agents within simulated enterprise environments. In these digital settings, the agents can practice various tasks, make real-time decisions, and learn from their mistakes as they navigate complex scenarios.
This method allows the models to gain experience in a controlled space before being deployed in the real world. By practicing in these simulations, the AI can refine its behavior and improve its ability to handle the specific nuances of a business's internal operations.
Reinforcement learning techniques
The startup employs reinforcement learning to enhance the performance of the underlying models. This technique involves rewarding successful behaviors, which helps the AI understand which actions lead to the desired outcomes for a specific business task.
By adjusting the models' parameters based on these rewards, Remanence can create systems that are more reliable for multistep processes. This process turns general open-weight models into specialized tools that can be tuned to perform complex work more effectively than general alternatives.
Moving away from US labs
Remanence is betting that companies will eventually prefer cheaper, customized AI models over the proprietary offerings from major US AI labs. This shift is happening as many organizations look for ways to reduce their heavy reliance on models from companies like Anthropic and OpenAI. The primary driver for this change is the high cost associated with using these large, closed systems for everyday operations.
By using open-weight models, the startup provides a way for businesses to build specialized tools that they can actually control. This allows companies to move away from a one-size-fits-all approach and toward a model that is tailored to their specific needs. It represents a strategic move toward independence from the dominant tech giants in the AI space.
What to watch
- How many other startups will adopt the 'simulated environment' training approach.
- Whether large US labs will release cheaper, more specialized versions of their own models to compete with this trend.
Try it
Research whether your company currently uses a general-purpose AI model for a specific business task and consider if a specialized model could reduce costs.
Sources and links
- Reportingtech.eu: the article we sourced
First reported by tech.eu. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.