The best robot might lose: how "where it's built" overtook "how well it works"

For most of its history, the autonomy industry ran on a simple faith: the best system wins. Lowest intervention rate, highest uptime, best perception stack — get the engineering right and the market follows. That faith is now obsolete, and the past two weeks buried it.

Consider three events, each covered in these pages, that are really one story. The US Federal Communications Commission added foreign-produced advanced robotic devices to its Covered List, cutting new foreign humanoids, quadrupeds and mobile robots off from the certification they need to be sold in America — regardless of price or performance. An Israeli defence-tech firm, XTEND, announced a $1.5 billion merger whose central asset is not a better drone but a US-based, NDAA-compliant manufacturing identity. And Kraken, a five-year-old British boatmaker, entered series production of its uncrewed vessels in Germany through a Rheinmetall joint venture, while supplying three allied navies.

Read together, these are not three deals. They are three data points on one curve: provenance has overtaken capability as the primary determinant of who gets to sell an autonomous system. Where a robot is built, and who controls its software and supply chain, now matters more to a growing share of buyers than how well it performs.

Why this happened, and why it's rational

It is tempting to dismiss this as protectionism dressed in a security costume, and some of it is. But the underlying logic is sound, and it is specific to autonomy in a way it isn't for, say, steel.

An autonomous system is not a passive product. It is networked, sensor-laden, software-defined and frequently updatable from afar. The FCC's supporting determination pointed to a reported incident in which a remote actor accessed thousands of foreign-made robots operating inside American homes. That is the nightmare that makes autonomy different: a compromised robot is not a stolen blueprint but a live actuator in your warehouse, your port, your home, taking instructions from somewhere you can't see. When the product can be remotely commanded, "who ultimately controls it" stops being a philosophical question and becomes a threat model.

Once you accept that, provenance is not xenophobia. It is a legitimate — if bluntly applied — proxy for a real risk. The blunt application is the problem: an FCC listing that is nominally country-neutral but understood by everyone, Beijing included, to target China; carve-outs at 2 kg and 200 kbps that are engineering trivia, not security logic; an installed base of grandfathered hardware whose risk didn't change the day the rule did.

The cost nobody is pricing

Here is what the security framing obscures: this movement has a bill, and buyers will pay it. Chinese platforms became the default in labs, warehouses and inspection fleets because they cost a third to a half of Western equivalents. Cutting off that pipeline doesn't make the Western alternatives cheaper; it makes the buyer's automation roadmap more expensive, slower, and — for smaller operators — sometimes unaffordable. There is a real productivity tax embedded in supply-chain nationalism, and it falls hardest on exactly the mid-sized firms that automation was supposed to help most.

There is also a capability cost. When procurement rewards the flag over the spec, the incentive to build the best system weakens at the margin. A market that guarantees demand to allied-built hardware is a market that tolerates allied-built hardware being a generation behind. That is a dangerous trade to make in a field moving as fast as this one.

The new scorecard

None of this is going to reverse; the security logic is too politically durable. So the honest response is not to pretend capability still rules, but to buy with both scorecards open. For any operator specifying an autonomous system now, three provenance questions belong next to the performance ones. Where is it manufactured, and does that satisfy the procurement regime you'll be selling into or operating under? Who controls the software supply chain and the update mechanism — and can that party be compelled by a foreign state? And what is your exposure on the installed base you already run: grandfathered today, but supportable in three years?

The uncomfortable truth for engineers is that the best robot can now lose to a worse one with a better passport. The uncomfortable truth for policymakers is that every point of security they buy this way is paid for in productivity and, eventually, in capability. Both truths are real. Pretending either away is how you make bad decisions in the sector that will define the next decade of industry.