The most consequential humanoid robotics chart of the past 90 days is a four-bar bar graph posted to Threads by Figure CEO Brett Adcock on April 23 and amplified all week. It is not pretty, and it is not subtle:
- All of 2025: ~150 robots total, across the entire calendar year.
- February 2026: ~60 robots in a single month.
- March 2026: ~120 robots in a single month.
- April 2026: ~240 robots in a single month.
April 2026 alone shipped more humanoid robots than Figure shipped in the entire prior calendar year. Three consecutive months of clean doubling. Adcock paired the chart with a separate claim that the Sunnyvale plant is now producing one Figure 03 every 90 minutes.
Whether that doubling continues a fourth time in May matters less than the fact that the doubling already happened three times in a row and the production line ran the whole time. That is the moment a hardware company stops being a hardware company and starts being a manufacturing company.
The data, exactly as Adcock disclosed it
Per Forbes, Tech360.tv, and the Humanoids Daily writeup:
- Production cadence: one robot off the line approximately every 90 minutes at the Sunnyvale facility. At 24-hour utilization, that’s ~16 units per day, ~480 per month — meaning April’s 240 reflects something closer to a 12-hour single-shift operating cadence, with ample slack for May to clear 480 if Adcock chooses.
- Cumulative installed fleet at end of April 2026: ~570 units. (150 from 2025 + 60 + 120 + 240.) That is the single largest deployed humanoid fleet of any Western OEM and roughly comparable to the combined Q1 2026 output of all named Chinese humanoid OEMs except Unitree.
- Hardware spec on the units leaving the line: Figure 03 generation. Upgraded camera system in the head plus a camera in each hand. Tactile sensors at the fingertips. Designed for the BMW Plant Spartanburg deployment and the Helix 02 end-to-end policy released in March.
- Customer mix: undisclosed at the unit level, but the revealed customer set includes BMW (Spartanburg), an unnamed large logistics/warehouse operator, and Adcock’s repeated framing of “100,000 humanoids over the next few years” as the stated four-year target.
Doubling three months in a row is not a press-release chart. It is a manufacturing-execution chart, and it is empirically real because somebody had to actually solder the boards, mate the harnesses, calibrate the tactile fingertips, and pass QA on each of those 420 robots in a 60-day window.
Where the doubling probably stops
Doubling forever is not a thing. The realistic upper bound on the current line is roughly 480 units per month at full 2-shift utilization (one robot per 90 minutes × 24 hours × 30 days, with realistic downtime). If May ships ~480, that’s still doubling. If June asks for ~960, the line capacity breaks before the demand does, and Figure has to either spin up a second line or admit the doubling cadence is over.
The interesting tell is the ramp-up in supplier orders Figure has placed, which industry watchers read as preparation for a second production line at scale. The most likely curve from here is:
- May 2026: doubles or holds (~400–480 units), single line saturated.
- Jun–Sep 2026: flattens to ~480/month while the second line spins up.
- Q4 2026: second line online, monthly volume steps to ~800–1000 if customer-side install capacity holds.
The iPhone moment for humanoids framing — popularized by Adcock himself — is partially right and partially wrong. It is right that this is the inflection from prototype to volume. It is wrong if it implies smooth exponential ramp through 2027. The next 12 months are where humanoid deployment economics get tested against the actual last-mile that Levie wrote about last week — except this time the last mile is the customer’s deployment integrator, not the model.
Why the bottleneck has already moved
Until February, the binding constraint on the humanoid industry was can the robot reliably do useful work for an entire shift in a real factory. The Helix 02 release in March, the NVIDIA GR00T N1.7 EgoScale paper yesterday, and Figure’s own deployment data at BMW have largely answered that one. The Apr-2026 binding constraint is can the customer’s facility integration team install, train, and certify a humanoid into the existing line fast enough to absorb the units the OEM is producing.
Per the Humanoids Daily analysis:
The real test isn’t whether they can keep scaling production, but whether customers can actually deploy these units fast enough to justify the ramp.
That is the whole game now. A Figure 03 in a shipping crate in Sunnyvale is depreciating capital. A Figure 03 working a station at BMW Spartanburg is generating measurable labor-cost displacement. The gap between those two states for an enterprise customer is currently 6–14 weeks of safety review, line integration, retraining for the specific task envelope, and union and OSHA conversations.
If that gap doesn’t compress, Figure’s exponential production curve runs into a deployment cliff sometime in Q3, and the unsold inventory headlines write themselves. If it does compress — driven by the same enterprise pressure that’s putting a third of Customers Bank’s headcount on the OpenAI partnership timeline — Adcock’s “100,000 over four years” target stops being aspirational.
What the doubling does to the rest of the field
A western humanoid OEM hitting 240/month with three months of doubling resets the competitive frame for everyone else:
- Tesla Optimus: Musk’s Q4 2025 admission that no Optimus units are doing useful work was already six months old. Figure’s volume now lapped Tesla’s first Optimus deliveries, even though Optimus has been on the runway for a year longer.
- Apptronik: Just appointed Daniel Chu CPO from Waymo and has a new humanoid imminent. Total deployed fleet still in low double digits via the Mercedes / GXO / Jabil partnerships. Figure’s Sunnyvale line is currently outshipping the cumulative Apollo fleet by ~10:1 on a quarterly basis.
- 1X: Consumer NEO preorders are open at $20K-equivalent or $499/mo. Different category — home rather than industrial — but the production cadence comparison is unflattering. 1X is shipping in tens, not hundreds, monthly.
- Boston Dynamics / Atlas: All 2026 units already committed to Hyundai RMAC and Google DeepMind. The keynote at the Robotics Summit & Expo today (April 30, Boston) by CTO Aaron Saunders is the direct response to Figure’s volume disclosure — Boston Dynamics needs to communicate its own ramp before “fully committed for 2026” reads as code for we’re still in the low hundreds.
- Unitree: 5,500+ humanoids shipped in 2025 at the bottom-of-market G1 / R1 price points. The China-vs-West frame here is real: Unitree wins the volume battle by an order of magnitude, Figure wins the work-actually-being-done-in-Western-factories battle. Both can be true.
What LostJobs is watching
- Whether Adcock posts a May number. If the May print is ~480, the doubling story holds and the second-line announcement follows in Q3. If May is flat at 240, the line is at saturation and the deployment-cliff scenario is the story.
- Whether BMW Spartanburg discloses a humanoid station count. The BMW partnership is the most-cited deployment but the actual unit count at the plant is not public. A BMW disclosure of “we now have N humanoids working line tasks at Spartanburg” is the deployment-side data point that confirms or denies the cliff.
- Whether a second large industrial customer is named. Figure has hinted at an undisclosed warehouse / logistics customer. If that’s named in May or at the Robotics Summit panel today, the unit-deployment math goes from “one big customer + small ones” to “two-customer-platform-business” — meaningfully different valuation.
- Whether Tesla’s Q1 2026 Optimus update on the next earnings call makes any quantitative production claim. Musk has been forced into specifics by Figure’s chart. Either he matches the numbers or he changes the subject.
The dry coda
The math underneath the chart is simpler than the chart looks. Figure’s plant builds a Figure 03 every 90 minutes today. There are 525,600 minutes in a year. A single, fully-utilized 90-minute-cadence line produces about 5,840 robots per year. Adcock’s stated four-year target is 100,000 humanoids, which requires roughly the equivalent of three additional Sunnyvale-equivalent lines running flat-out, plus a second-shift on each.
That is a hardware-company-becoming-a-Tier-1-manufacturer level of capex. Adcock’s competing narrative — that humanoids are the “AI’s body” and the next platform — is the bullish version of that capex story. The bearish version is that a hardware company that ships 5,840 of anything per year per line is a company whose unit economics, defect rate, supplier dependency, and scrap line all matter a lot more than the Helix 02 demo video does.
What April just told us is that, for the first time, a humanoid company is being graded on the boring questions. That is the better grading rubric. Figure earned it.