Tesla builds hundreds of Optimus robots a week that still take days to learn a basic task

The pitch for a humanoid robot is that an operator hands it a new job whenever the line changes. On Tesla's own floors, Optimus units run specific programmed tasks inside tightly controlled, supervised areas ^[6]. Teaching one even a basic new task reportedly takes several days ^[13].

Most units in the weekly count stay inside Tesla for testing, training and data collection, according to The Information as reported by Electrek ^[5]. Tesla is counting on that data to close the gap. It has more than 500,000 hours of training data and wants to double that by year-end ^[15]. Doubling would put it near 1 million hours ^[1]. It has moved much of its self-driving annotation team onto Optimus, hired data collectors in camera helmets and motion-capture suits, and set up training hubs in Colorado, Arizona and Florida ^[16]. The software plan is a library of basic movements the robot can recombine for new jobs ^[14].

The hardware has to hold up while that learning happens. The hand and forearm contain more than 100 screws and small components that workers still assemble by hand ^[8]. Some of the hand's touch sensors have had reliability problems. Tesla plans to add a replaceable "sensing glove" next year so a sensor fault does not mean swapping the whole hand ^[10]. Motors and the precision gears that drive the joints come from outside suppliers, many in China, and some that make good parts in prototype quantities struggle to keep quality consistent at volume ^[11]. The eventual goal of about 20,000 robots a week is twenty times the year-end target ^[2].

The first customers will be chosen to suit the robot. Tesla reportedly plans to lease Optimus to a short list of companies whose factories and warehouses look like its own, so the robots adapt more easily, and to use data from those sites to improve the AI ^[17]. Electrek described this as "the FSD playbook" ^[22]. Lease prices were not reported.

Tesla's dates have slipped before. In January 2025, Musk said Tesla would build about 10,000 Optimus robots that year and that "several thousand" would be doing useful work by year-end ^[18]. A year later he admitted that zero Optimus robots were doing useful work at Tesla ^[19]. He also promised a V3 reveal by mid-2026, and as of late September that demo had not happened ^[20].

For an operator weighing a humanoid pilot, two measures sort the pitch, and weekly output is neither. One is time to a new task, in hours or in days. The other is whether units stay in service without rework or part swaps. Put them on a grid: quick and reliable is a product you can deploy; quick but fragile needs a service contract priced in; slow but reliable competes with conventional automation built for one job; slow and fragile is still a research program. On the reporting so far, Optimus sits in the slow-and-fragile box ^[13][9]. In my view a lease today is worth signing only if it is priced as participation in Tesla's data collection ^[17]. The tradeoff is early access to whatever the robot learns next, paid for with floor space and supervision for a machine that runs one programmed task ^[6].