AI Training Data Landscape Revealed: 50+ Firms Valued at Nearly $100B

On July 12, investor @deedydas published a comprehensive landscape of the AI training data sector, sparking widespread community sharing and additions. He pointed out that as of July 2026, over 50 companies are selling training data and reinforcement learning (RL) environments to major AI labs, serving as an overlooked driver of model capability improvements. Collectively, these companies generate about $8.5 billion in revenue and hold a combined valuation of $100 billion, with over 75% of the revenue concentrated among a few top players. This review quantitatively unveiled the hidden "picks and shovels" business of the AI industry for the first time.

Sector Profile and Classification

@deedydas categorized the startups by sector, identifying areas such as data marketplaces (e.g., Kled, Luel) and computer/browser usage (Chakra Labs, Habitat, Refresh, Cua, Originator, Dojo), specifically naming Scale AI among the leaders. This classification reflects that as agent and computer-use capabilities become hot topics, RL environments and real-world operational data are emerging as new growth points for demand.

Community Additions and Data Corrections

Following the publication, @himanshustwts amplified the reach by reposting the same data set; @jonstephens85 added LightwheelAI, noting it does both egocentric and simulation data with a valuation exceeding $1 billion. @deedydas himself has been continuously revising the list, pointing out that some companies' revenues were underestimated or omitted (e.g., Arena's revenue has reached $100 million), and stated he would prioritize public sources, explicitly marking uncertainties before including rumors.

2026-07-12 ~ 2026-07-13 · 7 related posts

1 near-duplicate retellings: deedydas