Late on a weekday night, a five-foot-eight machine walks a set loop across Tesla’s Fremont lot at about three miles an hour. Nobody is holding a controller. The figure, matte black limbs and a chest plate that looks borrowed from a concept car, is Optimus, the humanoid Elon Musk has spent four years describing as the most valuable product Tesla will ever build.
The gap between that pitch and the reality on the ground has narrowed. Optimus units now walk factory aisles in Texas and California, and Musk has told investors the company will produce thousands of them before the decade is out. Whether those robots can do work worth paying for is the only question that really matters. The answer sits somewhere between the cinematic demo reels and the skeptics who call the whole thing a puppeteered stunt.
What Tesla Optimus actually is
Take away the stagecraft and you have a general-purpose humanoid standing about 173 cm tall and weighing roughly 57 kg, after Tesla trimmed 10 kg off the second-generation chassis. It balances on two legs, walks about 30% faster than the first version, and carries a 2.3 kWh battery pack in its torso, enough for a few hours of light work between charges.
The interesting hardware sits in the extremities. Each hand has eleven degrees of freedom, with tactile sensors in the fingers that let it judge how hard it is gripping something. Tendon-driven digits, forearm-mounted actuators and a neck that pans and tilts like a camera gimbal round out the design. Tesla builds most of those actuators in-house because nothing off the shelf hit the torque-to-weight targets.
The job list Tesla keeps pointing to is deliberately unglamorous:
- Lifting battery cells and modules between stations on a production line
- Sorting parts into trays using camera-guided grasping
- Walking between workstations over uneven concrete and up stairs
- Repeating the same motion for hours without losing accuracy
- Eventually, folding laundry and unloading dishwashers in homes
The brain is borrowed from Tesla’s cars
Optimus does not map its surroundings with lidar. It sees through cameras and interprets the world with neural networks, the same philosophy Tesla applies to driving. That inheritance may be the company’s real advantage: years of training vision models on messy, unpredictable streets translates more cleanly to a warehouse aisle than most people expected.
Running those models on a mobile platform needs serious compute, and Tesla would rather build its own silicon than buy it. That ambition now extends to fabrication, with plans for a $16.8 billion Terafab chip factory in Texas that Tesla and SpaceX intend to build together.
Paying for all of this takes cash flow, and Tesla’s Q1 revenue growth from EV sales and FSD subscriptions gives the robotics program room to spend while rivals burn venture capital.
From stage prop to factory floor
The 2021 reveal was a man in a suit
At AI Day in August 2021, Tesla wheeled out a dancer in a spandex bodysuit doing the robot dance. The crowd laughed. The spec sheet promised a machine that did not exist yet, and plenty of engineers wrote the whole thing off as theater.
Gen 2 raised the bar
December 2023 brought footage of a redesigned Optimus doing squats, balancing on one leg and picking up an egg with two fingers without cracking the shell. It was a real robot this time, though still on a leash. The video was heavily edited, and the machine had never worked an actual shift.
Then the robots started clocking in
By 2024, units were walking the floor at Gigafactory Texas and later at Fremont, handling battery cells on a pilot line. At the October 2024 We, Robot event in Los Angeles, Optimus robots poured drinks, played rock-paper-scissors and chatted with guests. Attendees later noticed that some of the conversational sparkle came from remote human operators speaking through the machines, which took the shine off the moment.
Musk’s production forecasts have wandered. He floated roughly 1,000 units for 2025, then revised upward to 5,000, then settled on a few thousand. The V3 design, with a rebuilt hand and forearm, has slipped into 2026.
Why hands are the nightmare
A human hand packs 27 bones and dozens of muscles into something smaller than a paperback. Replicating that in metal and tendon means actuators tiny enough to fit in a forearm and tough enough not to strip under repeated load. Musk has admitted the hand is the hardest part of the program, and it is a big reason V3 got pushed back.
Grip strength is not really the issue. Durability is. A factory robot might close its hand tens of thousands of times a week, and every gear, cable and sensor has to survive that without drifting out of calibration.
The money problem no demo solves
Musk has said Optimus should eventually sell for $20,000 to $30,000, with costs dropping toward $10,000 at high volume. Run the arithmetic against a warehouse wage and it looks compelling: a robot working two shifts a day, seven days a week, could pay for itself inside a year.
Real deployments are messier. Robots need power, maintenance, spare parts, software updates and a human supervisor for anything unusual. Manufacturers will want proof of uptime before they sign a contract. Tesla’s likely route is to deploy internally first, prove the economics inside its own plants, then lease units to outside customers rather than sell them outright.
The wider context matters too. Tesla is retiring the Model S and Model X and putting its weight behind the Cybercab as its next flagship platform, a sign the company is comfortable reallocating capital when it believes the future has shifted.
The competition is not standing still
Optimus grabs the headlines, but it is one entry in a crowded field:
- Figure AI has run humanoids on logistics tasks at BMW’s Spartanburg plant
- Agility Robotics’ Digit moves totes in GXO warehouses under a commercial agreement
- 1X is aiming its Neo robot at household chores, with pre-orders already open
- Unitree sells a small humanoid in China for around $16,000
- Boston Dynamics’ electric Atlas is the most athletic machine in the category
Hardware is commoditizing fast. The durable advantage sits in manufacturing scale, actuator cost and the data flywheel that comes from real deployments. Tesla has two of those three. The third depends on robots actually doing useful work at volume.
What would actually prove Optimus works
Skip the highlight reels. The signals worth watching are dull and specific. A robot completing a full eight-hour shift with no human minder standing next to it. Published uptime figures. Cycle times measured against a human worker on the same task. A cost per operating hour that undercuts a wage. Third-party footage shot by someone with no stake in the outcome.
Tesla has cleared one bar already: the robots exist, they walk, and they perform limited work inside Tesla’s own buildings. Everything after that is an execution problem, involving actuators, supply chains and software that generalizes beyond a rehearsed task. Musk’s attention is split across cars, rockets, satellites and social media, with SpaceX’s business drifting steadily beyond rockets in recent years.
Humanoid robots will probably become a real industry. The open question is whether Tesla, Figure or a Chinese manufacturer running on thinner margins gets there first. Watch the factory floor, not the keynote stage.