Nebius Group's Toloka, which uses human experts to train and evaluate AI models and has worked with Amazon and Anthropic, raised $72M led by Bezos Expeditions
Alexander Marrow / Reuters :
Context & Ripple Effects
Toloka sits within Nebius Group while providing human expertise for model training and evaluation, with Amazon and Anthropic among the companies it has worked with. The $72M round brings Bezos Expeditions directly into that layer of the AI development stack.
The investment also precedes Nebius's later move to acquire agent-search provider Tavily, indicating that the group’s related coverage spans both the human-feedback and information-access components used by AI systems.
First-order effects
- Toloka gains $72M in new financing, led by Bezos Expeditions, strengthening its resources as an independent human-expert training and evaluation provider within Nebius.
- Bezos Expeditions obtains exposure to the model-quality pipeline serving customers that have included Amazon and Anthropic.
Second-order effects
- Other providers of human-led model training and evaluation face a better-capitalized Toloka when competing for AI-lab and platform work.
- For Nebius, the funding adds support around a specialized asset alongside its later Tavily acquisition, linking human evaluation with tools that help agents retrieve current information.
Third-order effects
- If investment continues to flow into both model builders and the services that improve their outputs, human feedback and evaluation could become a more strategically financed part of AI infrastructure rather than a peripheral vendor category.
- Bezos-backed activity across collaborative AI and physical-task models—including Humans&' funding round—suggests capital is reaching multiple layers of the AI stack, though these investments alone do not establish a coordinated strategy.
The trend: AI investment is broadening from frontier models into the data, evaluation, retrieval, and operational layers that determine whether those models work reliably in practice.