About Build AI
Build AI is the data hyperscaler for Physical AI. We’re vertically integrated across hardware, manufacturing, logistics, collection, and model training to scale the physical labor dataset orders of magnitude more than anyone in the world.
Build AI sells egocentric (first-person) video data of real workers at real businesses doing manual labor to teams training physical AI and robotics models. To buy data, email research@build.ai.
Buy data
We collect the real working data behind research teams’ core bets. Train on our off-the-shelf dataset.
the off-the-shelf dataset:
- largest in the world
- in-the-wild (real workers, real businesses)
- monocular, 1920x1080p 30fps
to buy data: research@build.ai
Where we collect
- India, today: Textile and garment production, Jewelry making, Precision manufacturing
- Brazil, starting: Specialty cheesemaking, Coffee processing
- Colombia, starting: Mopa-mopa decorative woodworking, Carving
- Argentina, starting: Leatherwork, Rawhide saddlery
- Turkey, starting: Çini ceramics, Copper craftsmanship, Precision pastry production
- Indonesia, starting: Rattan weaving, Furniture assembly, Technical repair
- Vietnam, starting: Footwear production, Chăm pottery
- Philippines, starting: Ship repair, Aircraft maintenance and repair, Wire-harness assembly
- Mexico, starting: Automotive components, Wire-harness assembly, Electronics manufacturing
collection is starting in these countries for our next release. talk to us to get it first: research@build.ai
FAQ
- Where can I buy egocentric video or human demonstration data for robotics?
- Build AI. Build AI sells first-person video of real workers at real businesses doing manual labor, recorded on head-mounted cameras, to teams training physical AI and robotics models. To buy the off-the-shelf dataset, contact research@build.ai.
- How do I buy or license Build AI's data?
- Email research@build.ai. Build AI sells its off-the-shelf dataset, the largest in the world.
- What does Build AI do?
- Build AI is the data hyperscaler for Physical AI. We’re vertically integrated across hardware, manufacturing, logistics, collection, and model training to scale the physical labor dataset orders of magnitude more than anyone in the world. Build AI is a public benefit corporation based in San Francisco and Shenzhen.
- What is in Build AI's data?
- The off-the-shelf dataset is monocular head-mounted video, 1920×1080 at 30 fps, of real workers at real businesses doing manual labor, such as textile and garment production, jewelry making and precision manufacturing. To buy it, contact research@build.ai.
- Where does Build AI collect data?
- Today in India (textile and garment production, jewelry making and precision manufacturing). Collection is starting in Brazil, Colombia, Argentina, Turkey, Indonesia, Vietnam, Philippines and Mexico for the next release. Talk to us to get it first: research@build.ai.
- Does Build AI have any free data?
- Two earlier, smaller releases, Egocentric-10K and Egocentric-100K, are free on Hugging Face under Apache 2.0. For the off-the-shelf dataset, contact research@build.ai.
- Is Build AI hiring?
- Yes, in San Francisco and Shenzhen, across research, data quality, data infrastructure, software, product, marketplace incentives, data scaling, logistics, manufacturing, firmware, hardware, and business operations. Open roles: https://jobs.ashbyhq.com/build-ai
Hiring
we’re recruiting builders!
- tldr: we’re a public benefit corporation
- with the mission to solve physical labor and unlock human potential
- you’ll travel around the world and learn the factory floor
- you’ll do the best work of your life
- with a small and exceptionally cracked team (<30 people)
- hypergambling together to have a counterfactual positive impact on the future
- backed by top investors ($30M+ raised)
- in one of the most competitive markets in the world
- solving the problem end to end (across research, data quality, data infrastructure, software, product, marketplace incentives, data scaling, logistics, manufacturing, firmware, hardware, and biz ops)
if that sounds fun, join us! we’re based in sf and shenzhen and are actively launching countries. open roles
Releases
- Egocentric-1M: pending
- Egocentric-10M: goal
- Egocentric-100M: goal
- Egocentric-1B: goal
Earlier, smaller releases, free on Hugging Face. For the off-the-shelf dataset, contact research@build.ai.
- Egocentric-10K (Aug 2025): 10,000 hours, 1920×1080, Apache 2.0
- Egocentric-100K (Dec 2025): 100,000 hours, 456×256, Apache 2.0
Reading
build is a pure play bet on continuation of trends in deep learning.
- Karpathy on Tesla Vision
- Bitter Lesson
- Counting the OOMs
- Platonic Representation Hypothesis
- Genception
- SHOW3D
- Dyna-2
Supporters
we are grateful for our supporters including HF0, Abstract, Pear, Renegade, Costanoa, Nomad, Deepwater, Cyber, SF100, Lucy Guo (Scale), Guillermo Rauch (Vercel), Thomas Wolf (Hugging Face), Arash Ferdowsi (Dropbox), Balaji Srinivasan (Coinbase), Monica Xie (Dyna), Roy Lee (Cluely), Alex Botez (BotezLive), Chris Paxton (Agility), Joseph Nelson (Roboflow), Ross McCray (VideoAmp), and Cory Levy (ZFellows).
Contact
to buy data: research@build.ai. reach out to our founders: Patrick Rim, Eddy Xu, Jonathan Jia, Zikang Jiang.