Toloka, la nouvelle startup sur laquelle mise en direct Jeff Bezos

Bezos Expeditions, the personal investment fund of Jeff Bezos, has invested $72 million (approximately €64 million) in Toloka , a company based in Amsterdam specializing in data solutions for artificial intelligence. Announced in early May, this investment represents a significant step for Toloka, which already serves clients such as Microsoft, Amazon, Shopify, and Anthropic .

Toloka, founded by Olga Megorskaya , aims to provide high-quality human data to tech companies for training, validating, and improving AI models. Initially created as a micro-task platform, Toloka has evolved to specialize in expert data annotation, open source benchmark creation, and the development of virtual testing environments for generative AI. The company operates a network of over 200,000 human collaborators across more than 100 countries and 40 languages.

With Jeff Bezos joining as an investor, there is also a change in governance. Mikhail Parakhin, Chief Technology Officer at Shopify, is participating in the funding and will join the board as Executive Chairman. He emphasizes the increasing demand for world-class AI data expertise and commends Toloka's strong engineering, academic recognition, and its ability to combine scientific rigor with large-scale human coordination.

Previously part of Nebius, a spin-off from the international assets of Yandex, Toloka now enjoys greater autonomy as Nebius retains significant economic interest but no longer holds majority voting power. Arkady Volozh, CEO of Nebius Group, regards the involvement of Bezos and Parakhin as "a validation of the quality of the project led by Olga and her team," marking a transition to a new development phase.

The investment is intended to enhance Toloka's ability to design hybrid systems that integrate AI with human intervention, providing reliable, secure, and transparent solutions in a climate where algorithmic governance is increasingly significant for businesses and regulators. Olga Megorskaya notes that the funding will reinforce their approach based on human-assisted automation, aligning with responsible, user-focused AI.

This development comes amid growing discussions on data quality for training language models, especially in Europe with the introduction of the AI Act. Toloka's promise gains particular relevance as producing traceable, expert-validated, and regulation-compliant data becomes a strategic lever. The company is already recognized in the scientific community, with publications at conferences like NeurIPS and ICML, and regular contributions to MLCommons consortium work.

D’après FrenchWeb.