Tuesday, October 20, 2026
Issue No. 5 Robotics · Physical AI September 28, 2026

Issue No. 5 ·

Five Million Robots, and the Money Around Them

The IFR counts five million industrial robots at work. The same week, the big money went to the eyes and software around them, and a $29,990 two-armed robot came out of stealth.

News window: September 21–28, 2026 Feather's $29,990 robot HoldSoftBank × ASI Hold

News window: September 21–28, 2026

Editor’s Note

The International Federation of Robotics released its annual census this week, and the headline number is five million industrial robots now at work in factories. Almost none of them are humanoids. They’re arms bolted to the floor, welding, handling and assembling, and they’re the part of this industry that already works. The interesting money this week didn’t go to new robots at all. Cognex paid $500 million for RealSense’s depth cameras and Qualcomm agreed to buy PickNik, the company behind the open-source MoveIt motion-planning software. Buyers with real revenue are paying for the eyes and the software that make the existing fleet more useful.

Stress Test

Feather Robotics — a $29,990 two-armed robot that ships now

Claim
California startup Feather Robotics came out of stealth with a wheeled, two-armed robot priced at $29,990. It has two seven-axis arms on an omnidirectional base, 600 mm of vertical torso travel and about 10 hours of battery. It ships with an SDK and Python/ROS 2 support, and buyers choose their own onboard compute and AI model. The company raised a $7.6 million pre-seed round led by Gradient and says it already has more than $1 million in revenue, with units used in manufacturing and food service over the past year.
Demonstrated vs. marketing
The price is real and checkable, though it’s a promotional price against a listed $32,990. The revenue figure and the deployment history are the company’s own, with no named customers. “Humanoid” is a generous label for a wheeled two-armed mobile manipulator. There’s no task-level performance data, because the product is sold as a platform for developers to build on, not as a solution to a specific job. That’s an honest pitch, but it means there’s nothing yet to test against a production task.
Deployment reality
Feather is upfront that it’s selling a development platform, and that’s the right way to read it. A general-purpose two-armed base is a generalist by design. For any specific task a plant has in mind, a purpose-built cell will almost always do it better, faster and with less integration risk. The $30k price matters most to labs and developers, where it undercuts comparable research platforms by more than half. For a plant buyer, the real question is still what task, and whether a specialist tool already does it.
VerdictHold
Hold.

SoftBank × Autonomous Solutions — autonomous fleets for heavy construction

Claim
SoftBank Group put $225 million into Autonomous Solutions Inc. (ASI) and formed a joint venture to develop and commercialize autonomous construction equipment for large infrastructure projects: roads, airports, rail and waste sites. ASI’s Mobius software runs fleets of self-driving work vehicles and, per the company, works across machines from multiple manufacturers, so a contractor can automate equipment it already owns.
Demonstrated vs. marketing
The retrofit approach is the most credible part of the pitch, because it doesn’t ask a contractor to buy a new fleet. Everything else is still a structure, not a result. The joint venture’s funding, ownership split and name weren’t disclosed, no customer or project was named, and there are no productivity or uptime figures. The most detailed account is ASI’s own press release, not independent reporting.
Deployment reality
A jobsite is far less structured than any factory, so autonomous equipment there will see more downtime than the factory benchmarks buyers usually hear quoted. Even automotive plants, about the most controlled automation environment there is, run around 85 to 95 percent uptime. Construction buyers also have less experience pushing back on robotics uptime claims than manufacturing buyers do. The question to ask ASI is what availability its fleets actually hold on an active site, and who carries the cost when they stop.
VerdictHold
Hold.

Scorecard

ClaimStatusRevisit
Low-cost wheeled two-armed robots sold as developer platforms · Feather RoboticsStandby~Q2 2027
Retrofit autonomy for mixed heavy-construction fleets · ASI / SoftBank GroupStandby~Q3 2027

Spec Check

IFR: “Five million robots now operate in factories globally.”

The IFR’s World Robotics 2026 report puts the operational stock at 5 million at the end of 2025, up 9 percent, with 603,307 new installations during the year.

Run those two numbers against each other. If the stock grew 9 percent to 5 million, it started the year at about 4.59 million, a net gain of roughly 410,000 robots. But 603,000 were installed. The gap, about 190,000 robots, is the number that came out of service in the same year, about 4 percent of the fleet. The stock figure is an estimate built on service-life assumptions, not a headcount, so treat the retirement number as approximate. It still says something useful: a meaningful share of every year’s installations is replacement, not expansion.

The regional split is lopsided. China installed 354,000 robots in 2025, 59 percent of the world total. The US installed about 38,500, up 12 percent and its third-highest year on record. That’s still roughly one American installation for every nine Chinese ones. Germany fell 8 percent, to under 25,000.

Funding vs. Function

This week’s biggest checks went to acquisitions, not venture rounds, and that’s a different kind of signal.

Cognex is paying $500 million in cash for RealSense, which it expects to bring in $80 to $90 million in revenue this year. That’s roughly six times revenue for a business Intel spun out only last year. Cognex itself sizes the whole robotic perception market at about $600 million today, so it is paying close to the size of the current market for a position in it. Investors weren’t sure: the stock fell almost 5 percent on the news. Qualcomm didn’t disclose what it’s paying for PickNik, but it committed to keeping MoveIt open source, which is the part the robotics community actually depends on.

The week’s largest private round was TEKEVER’s $580 million Series D at a $6.4 billion valuation. TEKEVER builds autonomous aircraft and has reported defense-sector ties: the round followed its selection for a UK Ministry of Defence surveillance program worth up to £400 million.

Quick Hits