Tesla can't build its robot's hands reliably, so it will lease the robots

One hundred screws, placed by hand

The hardest thing Tesla builds is now five fingers. The Information reported on 25 September↗ that Optimus V3's hand and forearm hold more than 100 screws and small parts, and workers assemble them by hand on a Fremont line that stopped building the Model S in May ¹²³. Fixtures meant for car parts cannot align components far smaller and tighter-toleranced, so finished robots need immediate rework, and the hands' touch sensors fail often enough that Tesla's fix is a replaceable "sensing glove" due next year ¹²³.

A tenfold ramp against a 1,000-a-week promise

Output has multiplied roughly tenfold since spring, from a few dozen robots a week in the second quarter to several hundred by August ¹⁴. Managers want a line above 1,000 a week by year-end, on the way to an eventual 20,000 ¹³⁴. The stock fell about 2% on the news, to near $369 ⁴⁵.

Lease now, finish the robot later

The catch is that V3 is not the commercial product. It runs tightly programmed tasks in supervised zones, still needs days to learn basic jobs, and must clear durability thresholds before anyone buys one ¹³⁴. So early commercial units will be leased, not sold, to companies whose warehouses look like Tesla's own, which lets Tesla pull hardware back for upgrades and keep the training data ¹³⁴. A lease is what a company offers when the product is not finished. What no report carries is the term sheet: who absorbs the cost of a machine that stops mid-shift is not public.

Shareholders are paying for the churn

The bill for the ramp is already in the filings. Second-quarter revenue rose 26% to $28.2B, but operating income fell 57% to $0.4B on AI and R&D spending, free cash flow went negative at $1.1B, and capex jumped $3.3B in a single quarter ⁶. Capex landed at $5.8B, up 142% year over year ⁷. Musk's answer on the July call↗: Optimus will be the "biggest product ever," and the hardest Tesla has ever scaled ²⁶. The promise record is rougher: 10,000 robots pledged for 2025, zero doing useful work a year later, and a V3 reveal promised by mid-2026 that still has not happened ¹. He has pointed to 2027 for first commercial shipments since Davos in January, with supply-chain reports putting parts orders for about 15,000 units this year↗ ⁸.

The bulls' case, and the competition

The ramp itself is real, tenfold in months. StoneX kept its Buy and $475 target↗ into third-quarter deliveries, where Goldman sees just 435,000 vehicles, and TSLA is down more than 15% this year, the worst of the Mag 7 ⁹. XPeng started an automated humanoid line in Guangzhou this month, targeting 2027 sales ¹. Tesla, meanwhile, has gone back to Chinese suppliers in Ningbo for Optimus parts and is billing a dedicated Texas plant at 10 million robots a year, eventual capacity ⁵.

What December answers

Three markers decide it: weekly output crossing 1,000 by 31 December, the first named lease customer, and the sensing glove shipping on time. Musk's 2027 sales window survives only if the hands do.