I, Robot | ✉️ #93
Hey! 👋
The robotics news of the past two weeks alone tells the story. On May 1st Meta quietly closed the acquisition of Assured Robot Intelligence, signaling a clear "Android-for-humanoids" play to control the software layer the way it never managed with mobile. Amazon had already bought Fauna in March. Tokyo's Haneda Airport, in partnership with Japan Airlines, starts deploying humanoid robots for ground handling this very month. And on April 19th, an autonomous humanoid named Lightning, fielded by Honor, won the Beijing E-Town half marathon in 50 minutes 26 seconds — roughly seven minutes faster than the human world record. Add the established players (Unitree shipped over 5,500 humanoids in 2025, Figure has supported the assembly of 30,000 BMWs in Spartanburg, 1X is taking NEO preorders for delivery this year, Tesla is converting its Fremont Model S/X lines for Optimus production starting this summer) and the picture is unambiguous: 2026 is the year humanoids stopped being a demo and started becoming an industry.
The parallel with the automotive market is no longer a metaphor — it's the explicit playbook. Tesla is targeting $20,000–$30,000 per Optimus unit on lines designed for a million robots a year. China's State Grid Corporation has just allocated roughly $1 billion to deploy 8,500 robots across its power infrastructure in 2026. Goldman Sachs has lifted its robotics market forecast sixfold in the past year. The structure is forming up exactly the way the EV market did: a handful of vertically integrated giants chasing scale, Chinese manufacturers pressing on price (Beijing already has more than 150 humanoid companies versus around twenty in the United States), Western players betting on software differentiation, and now Big Tech (Meta, Amazon, Nvidia) trying to own the operating-system layer underneath the hardware.
The problem — and this is the angle worth dwelling on — is that the robotics industry is being born with a defect the automotive industry took a century to develop: the closed-by-design product. Almost every humanoid being commercialized is conceived as a black box: locked firmware, parts paired by cryptographic signature, diagnostics restricted to the manufacturer, non-standard spares. The philosophy is iPhone, not VW Beetle. The buyer pays for the device but doesn't really pay for the ability to maintain it. If a joint fails, the official path is to ship it back to the factory or, worse, swap the entire unit — exactly the way Tesla still replaces $16,000 battery packs rather than fix a cracked fitting. Meta's platform move makes it worse, not better: a single proprietary intelligence stack running across dozens of hardware brands means the company that controls the software controls who is allowed to repair the hardware.
That this isn't alarmism is now legally documented. Last month John Deere paid $99 million to settle a class action over blocking farmers' access to its diagnostic software — the first time repair restrictions translated into a nine-figure check, with an FTC suit still open. Washington State's Right to Repair Act took effect on January 1st, joining New York, California, Colorado, Oregon, and Minnesota. The federal REPAIR Act (H.R. 1566) cleared a House subcommittee on February 10th. The Repair Association's 2026 model legislation explicitly bans parts pairing and software locks as repair barriers, and senators Hawley and Luján — an unusual political pairing — are championing the bill. Even small-business lobbies are on board: 89% of NFIB members list right-to-repair as a top 2026 legislative priority. Yet none of this regulation, as currently drafted, mentions humanoid robots.
That gap is the whole point. A €25,000 home robot is not a phone you replace every three years; it's a durable good that ought to live in your house for a decade or more, like a car or a refrigerator. If it's born closed — running Meta's stack on Tesla's chassis with Figure's actuators, none of which you can legally inspect, repair, or swap — you depend entirely on the manufacturer continuing to exist, continuing to support that specific model, and continuing to have commercial reasons not to obsolete it. The Haneda trial this month is the first commercial deployment most travelers will physically encounter; the first home units arrive at the end of this year. The window to write the rules before the installed base reaches millions is measured in months, not years. We took far too long to demand it from tractors and cars: standardized parts, open service manuals, accessible diagnostics, the basic right to take your machine to the shop down the street. If we don't demand it from robots now — before Optimus, NEO, and Figure 03 ship at volume — we won't be buying mechanical helpers. We'll be renting them from the manufacturer for life.
The 94th mkdev dispatch will arrive on Friday, May 22nd. See you next time!