Tesla targeted 5,000 Optimus units for 2025. Actual output was a few hundred — below ten percent of the target — and as of mid-2026 its Fremont production line for the robot had not even started. Across the whole humanoid-robotics sector, verified deployments remain small pilot programs at named factories, not the mass labour replacement the funding rounds are pricing in.
Humanoid robotics has produced some of the most dramatic funding and valuation numbers in all of technology this year, built on a narrative of imminent, mass-scale deployment into warehouses and factories. The actual, verified unit-shipment numbers tell a much smaller and much more interesting story — one about a handful of real pilot programs generating real operating data, running years behind the timelines the capital markets have already priced in.
Start with the company that made the boldest public promise. Tesla targeted roughly 5,000 Optimus units for 2025; actual output came in at a few hundred units, below ten percent of that target, and as of mid-July 2026 reporting indicated Fremont's dedicated Optimus production line had still not started. Elon Musk himself has said the robots are "not in usage in our factories in a material way" — a striking admission for the company that has done the most to popularize the humanoid-robot investment thesis in the first place.
The picture elsewhere is smaller in scale but more concrete in substance. Figure AI has not disclosed a total unit count, but its Figure 02 robot completed an eleven-month pilot at BMW's Spartanburg plant, logging more than 1,250 operating hours, with a newer Figure 03 model now handling logistics tasks at the same site — a real, sustained, named-customer deployment, just a small one. Agility Robotics' Digit units have together logged more than 65,000 operating hours across nine customer facilities, including GXO, Schaeffler, Toyota's Canadian manufacturing operation, and Mercado Libre — the most extensively documented set of named commercial deployments in the sector, though the company has not disclosed a precise unit count either. Unitree, the Chinese robotics maker, reported shipping roughly 5,500 humanoid units across its full product line in 2025 and is targeting 10,000 to 20,000 in 2026 — the largest publicly stated shipment figure of any company in this letter, though a meaningful share of that volume is understood to be smaller, cheaper developer and research units rather than full industrial-labor humanoids.
| Company | 2025/2026 Reality | Stated Target |
|---|---|---|
| Tesla Optimus | A few hundred units; Fremont line not yet started (mid-2026) | 5,000 (2025) |
| Unitree | ~5,500 units shipped (2025, full product line) | 10,000–20,000 (2026) |
| Figure AI | Pilot deployments at named sites (BMW); no total disclosed | Not publicly quantified |
| Agility Robotics (Digit) | 65,000+ operating hours, 9 named customer sites | Up to 10,000/yr at RoboFab full output (not yet reached) |
None of this means humanoid robotics is a dead end — the BMW and Toyota pilots are real, sustained, and generating operating data that genuinely did not exist two years ago, and that data is exactly what eventually justifies (or kills) mass deployment. But the gap between "5,000 targeted, a few hundred shipped" and the valuations being placed on this sector is the single most important fact a prospective investor needs to hold in mind. Verified, named, industrial-scale humanoid deployment industry-wide is still measured in the hundreds to low thousands of units, concentrated in pilot programs, not the tens of thousands of working robots the "humanoid labor force" narrative implies is already underway.
The most useful question to ask about any humanoid robotics company right now is not "how large is the addressable market" — everyone agrees it is enormous — but "how many units, at which named customer, doing which specific task, for how many hours." That number, quarter over quarter, is the only real signal in a sector still mostly running on announced targets.
Direct pure-play exposure remains scarce and skewed toward companies that are either not independently investable or newly and thinly public. Figure AI, Agility Robotics, and 1X remain privately held as of this letter, meaning retail investors cannot buy them directly. Tesla offers indirect optionality on Optimus, but the robot is a small, unproven piece of a much larger auto and energy business, and the 2025 shortfall is a real data point against near-term enthusiasm. Unitree has reportedly been pursuing a listing in China, which would be the most direct pure-play exposure available if and when it completes. The more durable, less speculative way to hold this theme in the meantime is through the actuator, sensor, and precision-gearbox supply chain that every humanoid maker depends on regardless of which brand wins — companies like Harmonic Drive in the motion-control space, alongside the broader industrial-automation and Chinese robotics-component supply chain.
Humanoid robotics remains a capital-intensive, execution-heavy category where even well-funded players have repeatedly missed their own production targets, as Tesla's own numbers show. Safety, reliability, and cost-per-unit all need further improvement before mass industrial deployment becomes economically obvious rather than experimental, and a string of additional missed targets across the sector could trigger a sharp repricing of the venture and public valuations built on faster timelines. Geopolitical risk is also relevant specifically to the Chinese supply chain names, given the same export-control dynamics affecting other advanced-manufacturing sectors this year.
Humanoid robotics is a genuine, multi-year technology story with real pilot deployments generating real data — but the actual shipped-unit numbers are running years behind the capital that has already been committed to the theme. Treat any individual humanoid-robotics position as a venture-stage bet sized for a long timeline, track the quarterly gap between targets and shipped units as the primary signal, and prefer supply-chain and component exposure over any single brand until that gap closes meaningfully.