MLOps toolchains merging into one
Oscar Beijbom, co-founder and CTO of Nyckel, on the opportunites in the AI/ML tooling market
MLOps is converging into suites because the hard part is no longer training one model, it is turning messy data, evaluation, deployment, and monitoring into one repeatable workflow. Nyckel is built around that simpler workflow for smaller teams, where a product lead uploads examples, labels about 100 samples, gets a model in seconds, and ships an API without hiring a full ML team. Scale is moving in the same direction, but from the opposite end of the market, starting with large scale labeling and enterprise ML infrastructure.
-
Nyckel’s product thesis is that the right abstraction is inputs and outputs, not models. Customers bring their own text or image data, label it in the UI, and the system handles model search, training, deployment, and monitoring behind the scenes. That is what a merged toolchain looks like in practice for non experts.
-
Scale’s path to convergence came from vertical integration. It started in data labeling, then expanded toward training and deployment APIs, while its revenue surged from the LLM and RLHF wave, reaching $760M ARR in 2023 and an estimated $1.5B by the end of 2024. That scale gives it enterprise reach, but also ties it to more complex, higher touch workflows than Nyckel’s self serve motion.
-
The broader market already shows this bundling pattern. DataRobot combines model building, registry, deployment, observability, and governance in one enterprise suite, while Dataiku bundles data prep, model building, LLM routing, and app creation for business users. The standalone point tools are increasingly getting wrapped inside larger operating systems for AI work.
The category is heading toward two merged stacks. One will serve product owners and developers who want an API and a fast answer, which is where Nyckel is positioned. The other will serve large enterprises that want one control plane for models, agents, compliance, and infrastructure, which is where Scale, DataRobot, Dataiku, and the clouds are pushing.