Cuban gives humanoids five years. The one with real hours does logistics

Mark Cuban gives humanoid robots five years before they fail miserably, maybe ten. He listed the reasons in July and repeated the argument today. The interesting part is that his prediction already has a scoreboard, and the scoreboard agrees with the half of it nobody quotes.

“Humanoids have all kinds of issues. Latency. Processing power. Dexterity.”

— Mark Cuban, X, 23 July 2026

Mark Cuban, on X, 23 July 2026

What the money looks like on the other side of that call:

• Figure AI carries a valuation near $39bn after a round of about $675m led by OpenAI, up roughly 15 times from $2.6bn in early 2024.
• Humanoid startups raised about $4.3bn this year, out of more than $8.5bn into robotics as a whole.
• Apptronik is valued at $5.5bn. Unitree has shipped more than 5,500 robots.
• Tesla has never published an Optimus unit count.
• Agility's Digit has logged more than 65,000 operating hours across nine customer facilities.

Figure's valuation alone is about nine times what every humanoid startup raised this year put together, which tells you the market is pricing an outcome rather than a business. Cuban's objection is narrower than a mood. Humanoids, he wrote, are not designed to be optimal at much of anything and have to be upgraded continuously, and they will be great novelties for a while. Get me a beer. Fold my clothes.

The line that gets left out

Now the line that gets left out of the headlines: logistics have already set this path. Cuban is not predicting that robots fail, he is predicting that general-purpose bodies lose to application-specific ones, and he expects buildings to be designed around those machines within 25 years. That is a different claim, and it is testable today.

Test it and the exemption is the only part with hours on the clock. Agility's Digit is working at Schaeffler, GXO, Toyota Motor Manufacturing Canada and Mercado Libre, with more than 65,000 operating hours across nine facilities and, per its SPAC filing, $300m of multi-year orders covering 1,000 units on a three-year service contract. Divide it out and that is about $100,000 per robot per year, which is a real price for a real job. Tesla, the loudest humanoid story in the world, has published no unit count at all, and on its Q4 2025 call Elon Musk said Optimus was not in usage in its factories in a material way and that the units were primarily for learning.

Where the hours actually are

The pattern is familiar from the narrow end of the market. We counted it this month in policing, where 1,300 machines patrol Chinese streets doing one job each while a single showpiece stands in a New York subway station. Volume follows the narrow task. The humanoid is the demo.

Cuban could still be wrong, and the reason is the layer under the body rather than the body itself. The people building robot brains only recently left their GPT-2 moment, and latency, processing and dexterity are exactly the things a better foundation model fixes rather than a better chassis. If that layer improves fast enough, the general-purpose machine gets its reprieve. If it does not, the five-year clock started in July and the beer stays in the fridge.

Informational only, not investment advice. Valuations and order figures come from company statements and filings, and Tesla has never published an Optimus unit count.