Nyckel and Scale selling ML as API

The important point is that both Nyckel and Scale are trying to turn machine learning from a custom engineering project into a service bought like an API. In practice, that means a product team hands over examples, defines the output they want, and gets back a working classifier or labeling workflow without hiring an ML team. The overlap is strongest around making training data, model setup, and deployment feel packaged and outsourced rather than bespoke.

• Scale started with data labeling for autonomous vehicles and later expanded into Nucleus, fine tuning, deployment, and monitoring. That move shows the same basic destination Nyckel described, an ML stack where the customer buys outcomes and workflows, not separate infrastructure components.
  2 sacra 3 sacra 4 sacra
• The real difference is customer shape and depth, not the core promise. Scale has sold large enterprise and government programs, including defense and RLHF work, while Nyckel is a much smaller API first product for custom text and image models that can be trained from labeled examples.
  1 sacra 2 sacra 4 sacra
• This is similar to Webflow in one specific sense. The buyer shifts from a specialist building everything by hand to an operator defining the result in software. Dataiku shows the same pattern at the enterprise end, bundling data prep, automated ML, and visualization into a GUI so non technical teams can build AI workflows on top of existing data systems.
  1 sacra 4 sacra 5 sacra

The market is heading toward fewer standalone MLOps tools and more packaged AI products that hide the plumbing. The winners will be the companies that can make model building feel fast, reliable, and safe for ordinary product teams, while still serving the high precision enterprise workflows where Scale has already built distribution.