Companies
The catalog of physical AI companies tracked in DEPLOY, grouped by role. A company can appear in more than one role: a company that operates robots in its facilities AND ships robots to other operators shows up in both sections. Roles are derived from existing relations.
Focused role surfaces: manufacturers (914) · operators (2) · AI developers (55).
Filter applies to the visible list; the full page source contains every company.
Manufacturers (16)
Companies with at least one robot model on file. Sorted by funding raised, then model count, then alphabetical.
- Coco Robotics2 models· private$160M
- Serve Robotics2 models· public$80M
- Neubility1 model· active$42M
- Cartken1 model· private (~$22.5M raised; Magna/Mitsubishi Electric-backed)$20M
- Refraction AI1 model· private (still operating; ~21 employees)$4M
- Ottonomy1 model· private (~$4.9M-$7.8M raised; sources conflict)$3M
- Daxbot2 models· private (Salem, Oregon; StartEngine crowdfunded)$1M
- Amazon4 models· NASDAQ: AMZNno funding on file
- Avride2 models· Nebius subsidiaryno funding on file
- Kiwibot2 models· privateno funding on file
- Starship Technologies2 models· privateno funding on file
- Delivers AI1 model· activeno funding on file
- Neolix Technologies1 modelno funding on file
- Stub Test Co1 modelno funding on file
- TeleRetail1 model· activeno funding on file
- Yape1 model· activeno funding on file
Operators (0)
Companies running at least one verified deployment of someone's robots in their facilities. Sorted by operated-deployment count, then alphabetical.
None yet.
AI developers (1)
Companies credited as developers of at least one robot brain (foundation model, VLA, world model, framework, or OS-layer) in the registry. Sorted by brain count, then funding, then alphabetical. See also /brains.
- Avride· Nebius subsidiary1 brain
Machine-readable: this page as markdown.
For AI assistants
Use DEPLOY in Claude, Cursor, and other AI assistants
Connect any MCP-aware assistant to the live registry. Ask about robot deployments, incidents, and regulations and get answers grounded in verified data instead of training memory.