TL;DR
- Datadog accelerates its AI strategy with the acquisition of Adaptive ML, a startup specializing in Reinforcement Learning Operations (RLOps), in order to strengthen its internal artificial intelligence research laboratory.
- The acquisition is above all about talents and expertise in post-training of models, which has become the new battlefield of AI after the race for large language models.
- The value shifts from the model to its continuous improvement : companies are now seeking to specialize, control and optimize their AI agents using reinforcement learning rather than training new LLMs.
- Datadog has a major competitive advantage thanks to the billions of operational data collected on its platform, a strategic asset for training models specialized in observability and cybersecurity.
- The operation illustrates an underlying trend : infrastructure platforms seek to transform their proprietary data into differentiating artificial intelligence.
- Adaptive ML had raised $20 million in 2024 with Index Ventures, ICONIQ Capital, Motier Ventures, IRIS and Olivier Pomel. The amount of the buyout is not public, but a valuation of between $100 and $200 million seems plausible.
- The file also highlights Europe’s weaknesses : despite high-level research and available funding, large European groups remain too slow to adopt technologies developed by their own startups.
Datadog continues its transformation into an artificial intelligence laboratory, with the acquisition of Adaptive ML, specialist in Reinforcement Learning Operations, the American publisher, created by Olivier Pomel and Alexis Lê-Quôc, seeks to internalize expertise which could become decisive in the next generation of business software. Behind this acquisition lies a deeper evolution, infrastructure platforms now want to transform their operational data into proprietary intelligence.
The next battle now involves a much less visible layer, that of post-workout. Namely, the ability to constantly adapt, improve and evolve the models once deployed to customers. It is precisely in this area that Datadog has just positioned itself by announcing the acquisition of Adaptive ML, a Franco-Canadian startup specializing in Reinforcement Learning Operations (RLOps).
Datadog invests in research capacity more than in a product
Adaptive ML will integrate Datadog AI Research, an internal laboratory responsible for developing specialized models for observability and cybersecurity. Above all, Datadog is acquiring a team of researchers and rare expertise in one of the most complex areas of current artificial intelligence: the post-training of models using reinforcement learning.
Adaptive ML had never sought to develop a new major language model. Its ambition was to build the tools allowing large organizations to create, improve and deploy their own specialized agents based on their operational data.
Value migrates from the model to its permanent improvement
Since the arrival of ChatGPT, the industry has mainly focused its investments on model pre-training. The performance of LLMs was largely determined by the size of the datasets, the computing power mobilized and the number of parameters. This logic is now reaching its limits.
Generalist models are gradually converging in terms of performance, and companies are discovering that their true differentiation no longer depends solely on the model chosen, but on their ability to specialize, control and continually improve it.
This is precisely what Reinforcement Learning allows. Unlike classic fine-tuning, which consists of adapting a model from a static corpus, Reinforcement Learning allows the system to gradually learn from its interactions with its environment, correct its errors and optimize its decisions over time.
Adaptive ML specializes in this industrial layer, called RLOps.
Datadog data becomes a competitive advantage
This acquisition makes sense when we look at the assets Datadog has. Every day, its platform collects considerable volumes of data from its customers’ infrastructures: event logs, system metrics, distributed traces, alerts, incidents, human interventions and applied patches. This information is exactly the type of data that specialized models need to make progress.
Julien Launay emphasizes that “the most difficult thing has never been the algorithm, but the scale of production”. Datadog brings precisely this scale, the company has a constant stream of operational data from thousands of organizations around the world. Few players have such informational heritage.
A natural acquisition
During its $20 million fundraising in 2024, Adaptive ML brought together Index Ventures, ICONIQ Capital, Motier Ventures and IRIS, alongside Olivier Pomel, who participated as a business angel.
The financial terms of the acquisition have not been made public. Given the multiples observed on this type of acquisition in 2025/2026, a valuation of between $100 and $200 million is likely.
A French success… which highlights the limits of the European market
The story of Adaptive ML also tells of the limitations of the European market. Despite a Parisian office and a French ecosystem, the startup carried out most of its activity in North America, no major French client was then in its portfolio. An absence due to the slowness with which large European companies adopt the innovations developed by startups and their weak capacity to enter into contracts with them.
While Europe is effectively financing its technology startups, it is still struggling to create a sufficiently dynamic domestic market to enable them to become independent world leaders.
