_{A humanoid is mostly motion. Roughly forty actuators — the motors, gearboxes, and drives at every joint — dominate the bill of materials long before the AI does. Photo: DARPA (public domain).}

There is a clean way to think about what a humanoid robot costs, and it is not the way the demos invite you to think. Strip off the friendly face and the choreography, and a humanoid is, financially, a box of roughly thirty to forty actuators, the powered joints that let it move. Those actuators are not a line item. They are the product. They are somewhere between 40 and 56 percent of the entire bill of materials, and every other part, the cameras, the battery, the computer, the frame, is a rounding error next to them.

That one fact explains the whole industry. Tesla's Optimus is estimated to cost somewhere around $50,000 to $60,000 to build today, and Elon Musk wants that down to $20,000 to $30,000 at scale. The entire path from the first number to the second runs through the actuator, and specifically through the precision parts inside it. Which is exactly where China's vertically integrated robot makers, Unitree above all, already build a working humanoid for a bill of materials near $9,000. This is the free read on why the cost curve bends the way everyone says it will. The full teardown, every subsystem in dollars and the labor-payback math, is below the line for paid subscribers.

Tesla has never published an official bill of materials for Optimus, so every figure here is an analyst estimate, and I will flag them as such throughout. The most-cited teardown comes from Morgan Stanley, which pins the Gen-2 robot near $55,000 and, strikingly, puts about $21,000 of that, nearly two-fifths, into the legs alone. Bank of America gives a similar $50,000 to $60,000 range. The reason the legs cost so much is the same reason the whole robot is expensive: each leg is a stack of high-force actuators, and high-force actuators are hard to build.

SpaceX just filed for an IPO valued at up to $1.75 trillion. When that much capital becomes liquid all at once, where it goes next is the big question.

Meanwhile, spring art auctions in NY cleared $2.5 billion, with 15+ new artist records.

Prized, physical assets with fixed and scarce supply. When the ultra-wealthy get liquid, it’s one of the markets they reach for to diversify.

Masterworks lets you into that art market without needing the nine figures. Its members invest in shares of blue-chip artwork by artists like Banksy, Basquiat and Warhol.

The track record to-date?

Skip the waitlist to join Masterworks.

^{*Investing involves risk. Past performance is not indicative of future returns. See important disclosures at masterworks.com/cd.}

There are two kinds of joint in a humanoid, and they fail the cost test in different ways. A rotary joint, used at most of the body's hinges, is a frameless motor turning through a harmonic reducer, a compact, near-zero-backlash gearbox precision-machined in roughly thirteen steps and priced accordingly. A linear joint, used where the robot must push hard, replaces that gearbox with a planetary roller screw, and the roller screw is the villain of this entire story. Investment-bank estimates put a single one at $1,350 to $2,700, and an Optimus uses about fourteen of them. At the high end, that is more than thirty thousand dollars of screws in one robot.

Now hold that next to Unitree. SemiAnalysis estimates the BOM of Unitree's G1 humanoid at about $8,976, against a $16,000 base retail price and a fat gross margin. The G1 is a smaller, simpler robot with fewer degrees of freedom and no roller screws, so it is not an apples-to-apples Optimus, it is the floor-setter. Unitree's trick is quasi-direct-drive: a beefier motor with a low-reduction gear that is roughly eighty percent cheaper to machine than a harmonic-and-screw stack. It trades some precision for a far cheaper joint, and it builds its own motors and gearboxes off its quadruped line at thirty to forty percent of Western motor cost. Whoever builds the cheapest actuator sets the floor that everyone else's fifty-thousand-dollar teardown has to fall toward.

BOM (bill of materials): the full list and cost of every physical part in the robot. It excludes assembly labor, factory overhead, and margin, so a robot's price is always well above its BOM.

Actuator: the powered joint. It converts electrical input into motion and is, in cost terms, a motor plus a reducer (plus a screw if it is linear) plus sensors and a driver. The single biggest cost bucket in the machine.

Harmonic reducer (strain-wave gear): a compact gearbox that gives a high reduction ratio in one stage with almost no backlash. Precision-machined and therefore costly, it sits inside most rotary joints.

Planetary roller screw: converts a motor's spin into a hard linear push or pull using threaded rollers. It handles high force in a small package and, at $1,350 to $2,700 a unit, is the most expensive single drivetrain part in the robot.

Quasi-direct-drive (QDD): Unitree's approach, a low-reduction actuator with a beefier motor. Much simpler and cheaper to machine, roughly eighty percent cheaper, but bulkier; it trades precision for cost.

Learning curve (Wright's law): the empirical rule that unit cost falls a fixed percentage, here about fifteen to thirty percent, each time cumulative production doubles. It is the engine under every cost-down forecast in this space.

1. SemiAnalysis, how China's Unitree wins on the actuator, the best BOM and QDD teardown [18 min]

2. Fast Company, the tiny roller screw powering the humanoid revolution [7 min]

3. CryptoBriefing on Morgan Stanley, Optimus costs ~$55k to build, $21k just for legs [5 min]

4. 36kr, Musk's million-robot plan and the China motor, reducer, and lead-screw supply chain [9 min]

So the story for free readers is simple: a humanoid is a box of actuators, the actuators are most of the cost, and the cost-down is really the actuator getting cheap, which is a game China's quasi-direct-drive makers are already winning. Below the line, the part that took the most work, is the full model: the BOM line by line, the inside of a single joint, the cost-down curve, and the question that decides everything, does the robot actually beat a human worker.