TL;DR
Tesla has ramped Optimus production to several hundred robots per week, a 10x jump from Q2, but the hardware is fragile, the AI cannot generalize, and the commercial product does not yet exist.
What happened
- Several hundred Optimus units per week are now rolling off a converted Fremont line (formerly Model S/X) as of August, up from a few dozen per week in Q2.
- Tesla shut down Model S and Model X production in early May to free the Fremont floor and redirect engineers to Optimus.
- Internal target is 1,000+ robots per week by end of 2025, with an eventual goal of 20,000 per week.
- Nearly all current units go to internal testing and tightly supervised factory tasks, not commercial customers. The V3 being built now is not the version Tesla plans to sell.
- Elon Musk promised 10,000 units doing useful work in 2025. As of early 2026 he admitted zero were doing useful work, and the V3 reveal promised for mid-2026 has still not happened.
Why it matters
- Generalization is unsolved: Optimus still takes several days to learn basic tasks and behaves unpredictably outside trained scenarios, making it an expensive specialized robot, not a general-purpose one.
- Hardware reliability is a live problem: hands contain 100-plus hand-assembled screws and small parts, touch sensors have had failures, and Tesla's near-term fix is a replaceable "sensing glove" arriving next year.
- Supply chain is a bottleneck: key motors and precision joint gears come from Chinese suppliers who can hit prototype quality but struggle to hold tolerances at volume.
- Tesla is leasing, not selling, Optimus to a short list of companies whose facilities resemble Tesla factories, mirroring the FSD data-flywheel strategy: ship hardware, collect deployment data, improve the AI later.
- XPeng launched an automated IRON humanoid production line in Guangzhou this month, targeting commercial sales in 2027, signaling that competitors are not waiting for Tesla to solve these problems first.
What to watch next
- Whether Tesla hits 1,000 robots per week by year-end and whether the V3 reveal finally materializes. Both are overdue milestones that would reset credibility.
- Progress on AI generalization: Tesla wants to double its 500,000-hour training dataset by year-end and has redeployed its FSD annotation team to Optimus. Watch for any demonstration of multi-task performance without days of retraining.
- First external commercial deployments: the lease-not-sell model means early customer data will be the real signal. If no outside companies are running Optimus by mid-2027, the FSD parallel becomes harder to dismiss.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.