Tesla Is building 1,000 Optimus robots a week, but the hands are a major problem

1,000 robots a week?

Tesla is building Optimus robots at a rate that would have seemed like science fiction three years ago. According to multiple reports, the company is approaching -- and by some internal targets will soon exceed -- 1,000 units per week. That's an extraordinary manufacturing achievement for a product with roughly 10,000 unique parts built on a line that didn't exist eighteen months ago, if it's anywhere near accurate.

There's just one problem. The hands aren't really working yet.

And, honestly, a humanoid or other form of general purpose robot is generally a delivery system for hands, because hands do the work.

So this is a problem.

They're not entirely non-functional, but they're bad enough that the majority of units rolling off the Fremont line are being kept for internal testing rather than delivered to customers. And the root cause is a sobering reminder that building humanoid robots at scale isn't just a software challenge. 

Sometimes it's about nuts and bolts ... or screws.

What's actually breaking

According to reporting from The Information, drawing on internal Tesla sources, the hand and forearm assembly is the primary production bottleneck. Each unit contains more than 100 screws and tiny components that workers still assemble by hand, using factory fixtures that can't consistently align parts operating at tolerances far tighter than anything on a car line.

(So yes: mass automotive manufacturing is a major benefit when you need to mass produce robots. BUT ... it's not a silver bullet. Some things on robots are just harder because they're at a much smaller scale.)

The touch sensors built into the hands have proven unreliable at scale. They work in prototypes, they work in small batches, but they don't work consistently when you're trying to build hundreds per week from suppliers who haven't done this before. Tesla's reported workaround is a hot-swappable "sensing glove" -- an external sensor layer that can be replaced rather than repaired -- planned for the next model year.

The underlying mechanical design is ambitious to the point of being nearly unreasonable to manufacture. 

The Gen 3 hands have 50 actuators and 22 degrees of freedom per hand, using a tendon-driven system designed to replicate human mechanical advantage. That's 50 precision actuators per robot, each requiring rare-earth magnets and sub-millimeter tolerances.

This is ... not easy.

Tesla's supplier problem

Making this harder: Tesla doesn't manufacture these components itself. 

The motors and precision gears driving the hand joints come from outside suppliers, many based in China. These suppliers can hit quality targets in prototype volumes: the kind of careful, hand-checked production that goes into engineering samples. 

At production scale, their consistency seems to be breaking down.

This isn't unusual in manufacturing. 

The entire automotive industry spent decades learning how to qualify suppliers for high-volume precision parts. Tesla is trying to compress that learning curve for components far more complex than anything in a car, on a timeline that doesn't allow for the usual years of supplier development. 

The result is exactly what you'd expect: inconsistent quality at exactly when they're trying to scale.

The Optimus humanoid robot production reality

Tesla scaled output roughly tenfold from Q2 2026 to late September, from a few dozen units per week to several hundred. 

The internal target is to exceed 1,000 units per week by end of 2026. The longer-term stated ambition, articulated repeatedly by Elon Musk, is 20,000 robots per week, with an eventual Gigafactory Texas line targeting 10 million units per year.

(No-one has ever accused Elon Musk of aiming low.)

The hand bottleneck is threatening the 1,000/week target. Most of what's being produced right now is staying inside Tesla and used for testing, training-data collection, and validation. It's not for commercial customers.

Musk was characteristically candid on the Q1 2026 earnings call: 

"Optimus production will be extremely slow at first, as everything is new. This is not like making a car. It is literally impossible to predict the production rate this year, given that Optimus has roughly 10,000 unique parts running on an entirely new production line."

That was Q1. We're now in Q4, and it doesn't seem solved yet.

Software complicates everything

Even if Tesla solved the hand problems tomorrow, it wouldn't immediately unlock commercial scale. 

Analysts who've tracked the program closely note that the hardware constraints are actually secondary to the software constraints for deployment purposes: Optimus still requires per-task programming and days of training to handle basic jobs, and cannot generalize across novel tasks despite hundreds of thousands of hours of training data.

"The hardware is not a barrier right now -- the barrier is software and training," one analyst told Electrek. 

Which is true for deployment, but somewhat misses the point: if the hardware doesn't work reliably, you can't collect the training data you need to close the software gap.

It's a vicious cycle: the two problems are coupled. Broken hands produce corrupted training signals. Corrupted training signals slow the generalization work. Slower generalization means longer deployment timelines.

The only solution is taking the time to do it right, and Musk is not known for being patient.

What Tesla's bet looks like now

Tesla converted the Fremont Model S/X line -- torn down in 46 days -- to Optimus production. Engineers were pulled from multiple vehicle programs. The scale of the commitment is genuine and irreversible in any near-term sense. 

If Optimus fails as a product, it won't be because Tesla didn't try hard enough or bet big enough.

The early commercial strategy is to lease rather than sell: placing units with enterprise customers while maintaining control over repair and servicing, and collecting real-world operational data that feeds back into training. 

Broader commercial sales are now penciled in for 2027 at the earliest.

None of this means Optimus will fail. A company that goes from zero to several hundred units per week in eighteen months, in a product category with no precedent at commercial scale, is doing something impressive. The hand problems are real, but they are the kind of problems that get solved through iteration ... if the company gives the engineering teams and suppliers the time that they need. 

1,000 units per week is a milestone worth watching. 

Whether Tesla hits it by year-end -- and with hands that actually work -- will say a lot about whether the production ramp is on track or just running hot.

Sources

• Electrek: Tesla ramps Optimus to hundreds a week, but the robots can't generalize (Sept 25, 2026)
• Futurism: Tesla Reportedly Can't Get the Hands to Work on Its Optimus Robot
• 247 Wall St: Elon Musk Wants To Build 20,000 Optimus Robots A Week. There's Just One Problem (Sept 29, 2026)
• Interesting Engineering: Tesla targets 1,000 Optimus weekly, but robots struggle with tasks
• Neoteo: Tesla Optimus production problems -- reported snags