About
Hub builds distributed multimodal data infrastructure that captures, curates, and validates real-world human and egocentric data for physical AI. It sells production-ready datasets to frontier AI labs, robotics companies, and large enterprises, differentiating through source-captured, provenance-attested data, automated pipelines, and human-in-the-loop quality assurance rather than synthetic or public-web data.
Market
Hub competes in physical-AI data infrastructure and multimodal AI training-data markets, serving organizations that need diverse real-world, egocentric, and sensor-rich data for robotics and embodied intelligence. It differentiates through a distributed contributor and commercial-site network, synchronized multimodal capture, diversity controls, and rapid custom-dataset delivery, positioning it as a data-collection-and-dataset provider rather than only a conventional annotation platform.
Hub primarily serves frontier AI labs, leading robotics companies, research labs, and technology enterprises that need large-scale, real-world multimodal training data. Its likely buyers are AI/ML, robotics, and data-infrastructure leaders seeking custom datasets and continuously refreshed physical-world data.
At a Glance
Problem
AI labs and robotics companies face a data ceiling: public-web data is finite and increasingly contaminated by AI-generated output, while the next generation of models needs original, high-fidelity examples of human voice, language, nuance, environments, and physical motion. The core pain is therefore not simply labeling existing data, but collecting diverse real-world multimodal data in settings that synthetic data cannot reproduce, then validating and delivering it at production scale. Hub’s main use case is training physical-AI and robotics systems on egocentric demonstrations of hands, objects, tasks, and movement across real commercial and residential environments.
The economics are designed to make this supply available by turning work that businesses and employees already perform into paid training data. Hub advertises US$3–10 per approved recording hour for contributors and gives an example of a restaurant with eight employees earning approximately $5,280 per month, while AI companies get a more scalable alternative to building their own collection operations.
Product / Service
Hub is a B2B real-world data infrastructure company and multimodal data lab. Its HubApp coordinates a distributed network of contributors and verified commercial sites, matching collection tasks to people based on hardware, location, and language. Contributors use head-mounted cameras to record normal work, while Hub captures and packages modalities including egocentric RGB-D video, synchronized IMU and motion data, images, video, audio, scene descriptions, object and hand tracks, and annotations. Automated processing is combined with human-in-the-loop quality assurance so the output is production-ready for frontier AI labs and robotics companies.
Customers can request bespoke datasets and have Hub parse the specification, generate a quote, match suitable environments and contributors, provide initial samples, and deliver the approved collection in formats such as MCAP, LeRobot, or the customer’s own format. Hub says the process can begin delivering samples within hours and complete a collection in roughly 36 hours. On the supply side, setup is intended to require little or no workflow change: employees record voluntarily during existing work, footage uploads through the app, and the business receives payment for approved hours.
Market
Hub competes in AI training-data infrastructure, multimodal data collection, and the emerging physical-AI and robotics-data market. Its closest disclosed comparables include Scale AI, which offers real-world data from global robotics data factories for Physical AI, and Sensei, which helps robotics companies outsource human-demonstration data collection through a hardware platform. Hub’s positioning is differentiated by combining a broad distributed contributor network with verified commercial sites and an end-to-end pipeline for custom egocentric, image, video, and audio datasets.
The company is not pre-revenue according to its Y Combinator profile: it reported moving from five-figure to seven-figure data-delivery revenue agreements with major AI and robotics companies. Hub was founded in 2024, is an active Spring 2026 Y Combinator company with a listed team of 10, and reports more than 150,000 active contributors, 730-plus commercial sites across 150 countries, and more than 540,000 hours captured. It also reported $1.7 million in total fundraising through September 2025.
Founders & Leadership
Funding History
SwissBorg, Y Combinator
SwissBorg
Recent News
Hub.xyz announced the closing of its pre-seed round and the opening of its seed round, with SwissBorg leading the pre-seed and participating in the seed round. The rounds brought Hub.xyz's total fundraising to $1.7 million.
Active Roles
2Business Model
Hub sells production-ready real-world training datasets and data-delivery agreements to frontier AI labs, robotics companies, and enterprise customers. It sources the data through a distributed contributor and business network, paying contributors per submitted and approved recording hour at region-specific rates.