Issue #001 — The Humanoid Gap
Published 2026-07-09 · Updated 2026-08-18
Robotics has a humanoid problem — but probably not the one you've read about.
The headline version is that humanoids are overhyped. The actual version, once you stop reading press releases and start counting products, is weirder: the humanoid story is, in catalog terms, largely a Chinese-catalog story — though as of August 2026, that's changing faster than most databases admit.
I spent recent weeks pulling five public robot directories into one merged dataset. It's a cross-section of catalogs, not a census, and the counts need a reconciliation caveat up front: my merge produced 929 entries (deduped by canonical name, as of 2026-07-09), but the FinBrain catalog — our running internal tally — currently reports 627 robot entries plus 184 component/hand/gripper/arm records (811 total). The gap is methodology, not mystery: my merge counted some component and subsystem listings as robots, and FinBrain's tags are stricter. Treat the FinBrain figures as the cleaner denominator where the two disagree; I've flagged each affected number below.
Two of my five sources are Western (an open-source robot index, a startup list); two are Chinese (a community database, an industry portal); one is a Western media catalog. They barely overlap. Here's the part nobody reports — the humanoid share by source, as a proper exhibit:
Humanoid share by source (2026-07-09 snapshot)
| Source | Approx. humanoid share of entries | |---|---| | Chinese community database | High — the bulk of humanoid entries | | Chinese industry portal | High | | Western open-source index | Near zero then; visibly rising (see below) | | Western startup list | Near zero | | Western media catalog | Near zero |
Overall: humanoids were ~20% of the whole merged list, and ~59% of the 34% of entries I could categorize at all. The catalog attention concentrates here even if the install base doesn't.
The humanoid "boom" lives almost entirely inside Chinese directories — or did at collection time. The Western sources most analysts quote listed close to zero humanoids as of early July. Six weeks later, that framing needs an asterisk: the NYT ran a major piece on America's push to build its own humanoids (2026-08-16), Jiemian reports Tesla's Optimus entering a production ramp (2026-08-16), and Singapore's Menlo Research open-sourced Asimov-1 — a 1.2m, 35kg, 25-DoF biped with full design files on GitHub and a $15k DIY kit (2026-08-17). Western and open-source humanoid entries are visibly accumulating in near-real-time. The structural point survives: if your embodied-AI thesis comes from Crunchbase and arXiv, you were sampled from one side of the room. But "close to zero" is a July number, not a permanent condition.
Three things the data actually says
1. The industry is long-tail, not concentrated. Across ~513 distinct companies (my merge; unreconciled against FinBrain's stricter tagging — treat as directional), roughly 69% ship exactly one product. Median: one robot. Mean: ~1.8. For an industry everyone describes as "consolidating around winners," the product graph says the opposite: a field of single-product specialists, and nobody with catalog breadth. Whatever moat you think the incumbents have, it isn't catalog breadth.
2. The companies shipping the most models are not who Western coverage implies. Of the ~31 companies with five or more entries, the top six at collection time were all Chinese — REEMAN (18), Leju (14), Engineai (14), Chuansanjia (14), Elephant Robotics (14), Zhiyuan (12). Boston Dynamics sits at 12. Important caveat: those Chinese shippers are mostly service, cleaning, and education robots — tagged from product names, so with the tagging-layer humility noted below — not humanoids. The "China is flooding humanoids" narrative doesn't survive contact with the actual product list. What China is flooding is the boring, revenue-generating service category.
3. The funding-vs-catalog gap is real, but I can't size it honestly yet. Humanoids dominate funding attention while occupying ~20% of the merged catalog (on a denominator where 65% of entries were uncategorized — every percentage here carries that asterisk, and the 65% figure itself predates FinBrain's stricter tagging). I don't have a funding-share number I trust: PitchBook and Crunchbase both under-cover Chinese rounds. So no fake ratio. The honest claim is narrower: the part of the field that attracts capital and the part that shows up in cross-sourced catalogs are not the same part. Resolving that gap is on my roadmap, not in this issue.
The caveat I owe you
This dataset is a cross-section of public catalogs, not the industry, and it's now six weeks stale against a market that added an open-source biped and an Optimus ramp in the same fortnight. Industrial arms (FANUC, KUKA, ABB) are badly undercounted — they ship thousands of SKUs that never make these lists. 65% of entries didn't fit any category tag I had, which is a flaw in my taxonomy (no buckets yet for surgical, agricultural, underwater, or cobot-specific), not a flaw in the field. And yes — the first draft of this issue mislabeled REEMAN as a humanoid maker. It isn't; its 18 entries are delivery and cleaning robots. If my tagging can misfire there, treat every "mostly service, cleaning, education" claim above as name-inspection, not ground truth.
That's Issue #001. Every week: one claim I can defend with counts I actually collected, one caveat I owe you, and no fabricated ratios. If I can't size something, I'll say so.
Next week: the funding-vs-catalog gap — building a defensible number instead of a vibe.
→ Browse the methodology notes → mazeintelli.com → [Hit reply — especially with the funding data I'm missing.]