Why humanoid robots won't win factories—yet

The demos are convincing. Following a ten-month pilot at its Spartanburg plant, BMW says it will deploy Figure's newer humanoid on the production line — the footage reads like the future has arrived. It hasn't, and the reason has nothing to do with whether the robots are smart enough. It has to do with a spreadsheet.

The industry frames humanoid adoption as a capability problem: get the hands dexterous enough, the models general enough, and factories will buy. That's the wrong constraint. As The Robot Report puts it, the real issue "has less to do with intelligence and more to do with robotic unit economics and financial viability." A humanoid can be brilliant and still lose the purchase-order decision to a fixed arm that costs a fraction as much and moves twice as fast.

The barrier is unit economics, not intelligence

Start with the sticker. One leading humanoid developer put the cost at several hundred thousand dollars per robot, on its own disclosure. For most manufacturers, that number ends the conversation before a single integration meeting.

Then look at throughput, where the general-purpose design pays a tax. In parcel sorting — a task humanoid vendors like to showcase — the best demonstrations hit around 1,000 picks per hour in controlled conditions. Specialized parcel-sorting robots run between 1,300 and nearly 3,000 picks per hour. The humanoid costs more and does less, at the one job it was brought in to prove.

This is the uncomfortable part for the engineering-led roadmap: the very features being optimised — dexterity, retrainability, the ability to do many jobs adequately — are what keep the machine from doing any single job as cheaply as a purpose-built cell. Generality extends the timeline to cost viability rather than shortening it.

Total cost of ownership is the number that decides

Prices will fall. IDTechEx projects the average selling price of humanoids dropping from roughly $114,700 in 2024 to around $37,000 by 2030. Take that figure at face value and the objection looks like it disappears — but a manufacturer never compares purchase price to purchase price. It compares lifetime return.

Even at a low unit price, the math only closes under heavy utilisation: you need a multi-shift operation running the machine around the clock before the numbers begin to work. And here the logic turns on itself. A plant already running that kind of operation can put the same capital into a specialized automation solution that delivers more throughput and pays off faster. The humanoid's best-case financial scenario is precisely the scenario where its dedicated competitor wins more decisively.

That framing lines up with how at least one analyst reads the current state of play. Robotics & Automation News, summarising an IDTechEx assessment, argues that humanoids now show "clearer ROI" but that commercial success "depends on effective output" — the machine has to actually earn its keep in throughput, not just in a headline price that has finally come down. Cheaper hardware is a necessary condition, not a sufficient one.

Layer switching costs on top — retraining, line redesign, integration risk — and capability parity isn't enough to move a buyer. A new technology has to be visibly, durably better on total cost of ownership to justify tearing up a line that already works. What's left to prove, in the words of The Robot Report's analysis, is "that humanoid robots can beat the economics of the automation manufacturers are already running." That is a much higher bar than passing a dexterity benchmark.

Scale is the unlock, and it's a chicken-and-egg problem

BMW's Spartanburg pilot is the most cited proof point, and it's worth reading precisely. Over ten months in 2025, Figure's 02 robot logged about 1,250 operating hours and moved more than 90,000 components, contributing to 30,000 vehicles. BMW has since committed to deploying the newer Figure 03. That demonstrates viability. It does not demonstrate viability at scale — 1,250 hours is a proof of concept, not a fleet.

The route to competitive economics runs through manufacturing volume. Building complex hardware at massive scale drives down component costs, tightens supplier relationships and spreads engineering investment across far more units — every lever that improves robotics unit economics. Humanoid developers don't yet have that manufacturing base, and here's the bind: reaching it requires the volume orders, and the volume orders require proof of economic advantage that only scale can produce.

That is the loop the sector has to break, and it explains why the honest answer to "when do humanoids win factories" is not yet rather than never. Designs that trade some generality for cost and stability may get there first — a shift worth watching in Walden's wheeled humanoids.

What would change the verdict is narrow and measurable: a humanoid deployment that beats a specialized cell on cost-per-unit-of-output at a real utilisation rate, disclosed rather than demonstrated. Until a manufacturer publishes that comparison and keeps the robots on the line after the pilot budget ends, the winning machine on most factory floors is still the boring one bolted to the ground.