Tesla plans to sell Optimus in 2027 and I

Tesla announced they’ll sell Optimus to consumers in 2027. $25,000.

I half believe them.

The half I believe

The hardware is plausible. I estimated the BOM at $6-7K at scale last July. Tesla’s manufacturing capability is real. They make half a million cars a month. They know how to scale production of complex electromechanical systems.

$25,000 (up from the $20,000 Musk initially suggested) is more realistic and still aggressive. A 3-4x markup on the BOM covers assembly, software, warranty, and a reasonable margin. The price is in the range of a car, which is exactly the comparison Tesla wants people to make.

The timeline is plausible too. The Optimus prototypes from late 2025 show capable hardware: walking, object manipulation, basic navigation. 18 months of refinement for a consumer product is tight but not impossible for a company that iterated from Model S to Model 3 in similar timeframes.

The half I don’t believe

The software for home autonomy is a different problem than the software for warehouse navigation. A warehouse has structure. A home has chaos.

Can Optimus navigate a living room with toys scattered on the floor? Can it handle a toddler running unpredictably? Can it understand a verbal request in a noisy kitchen with the TV on and water running? Can it decide, with taste, how to fold a sweater that’s already been worn but isn’t dirty?

That last question is the one that sticks with me. Not the navigation. Not the manipulation. The judgment. The “how should this sweater be handled” question that a human answers unconsciously based on context, culture, personal preference, and something we call taste.

Hardware is engineering. Home autonomy is engineering plus taste. And taste is what AI doesn’t have yet.

My prediction

Tesla will ship something called Optimus in 2027 or 2028. It’ll do 5-10 useful things in a home. Carry items between rooms. Load a dishwasher with standard-size plates. Vacuum and mop. Simple cleaning tasks.

It won’t do the things that make a home feel maintained. The judgment calls. The contextual decisions. The “put this in the right place” problem that requires knowing what the right place is.

That will come later. Maybe 2030. Maybe 2032. But the first version will be limited enough that early adopters love it and critics dismiss it. Like every first version of every Tesla product.

I’m watching. With one eye on the hardware specs and one eye on the software demos. The hardware will be ready. The question is whether the software is ready enough.

“Ready enough” is doing a lot of work in that sentence. It always is, with first-generation products.

Related thinking: