Samsung's $13B robot unit aims at its own factories first

Samsung folded its scattered robotics teams into one $13B division on July 21. The first factories it plans to automate are its own.

Samsung's $13B robot unit aims at its own factories first

On July 21, Samsung Electronics stopped pretending its robotics work was a side project. It folded a decade of scattered R&D — the Bixby-era gadget teams, the industrial-arm groups, its majority stake in Rainbow Robotics — into a single unit called the Robotics eXperience Business Promotion Office, or RX, reporting straight to device chief TM Roh. The number attached to it is ₩19 trillion, roughly $13 billion, with a new manufacturing base in Gumi. That is not a research budget. That is a factory budget.

And the factory RX is built to automate first is Samsung’s own.

The customer is also the guinea pig

Samsung’s stated goal is to turn every one of its production sites into an “AI autonomous factory” by 2030, then package what it learns and sell it B2B to other manufacturers. Read that sequence carefully, because the order matters. Samsung is not planning to sell humanoids and then, incidentally, use some in-house. It is planning to automate its own plants as the demo — and only once the demo works does the sales pitch exist.

This is the honest version of a story most robotics companies tell backwards. Figure and Agility talk about “labor shortages” and “jobs nobody wants.” Samsung has quietly conceded the actual business model: to sell a machine that replaces factory labor, you first have to replace factory labor somewhere, and the cheapest place to run that experiment is on workers who already report to you. Samsung’s own line staff aren’t a market for the product. They’re the R&D data set.

Shallow-π and the speed problem

The reason this isn’t just another press release is that Samsung showed up with a working brain, not a concept video. Samsung Research has published Shallow-π, a Vision-Language-Action model it says compresses the control pipeline to about one-third of what comparable systems need and makes roughly 17 decisions per second. Those two numbers matter more than the $13 billion.

Humanoids don’t fail in the demo because they can’t lift a box. They fail because the loop between “camera sees the box” and “hand closes on the box” is too slow and too brittle for a real production line running a real takt time. Every fraction you shave off that loop is the difference between a robot that photographs well at a trade show and one that can stand at a station for a full shift without falling behind the belt. If Shallow-π does what Samsung claims — and “already passed factory tests” is the load-bearing phrase in the announcement — the company is claiming it has solved the boring part, which is the only part that was ever hard.

Two chaebols, one roadmap

The person running RX strategy is Dongkun Lee, an executive vice president who joined Samsung in May. His last job was leading robotics strategy at Hyundai — the group that owns Boston Dynamics and its Atlas humanoid. So the two largest industrial conglomerates in Korea are now both building humanoid roadmaps, and one man has now helped write both of them.

Which is worth remembering alongside a story we covered five days ago: 35,000 Hyundai workers escalated to four-hour strikes on July 20, and the sticking point wasn’t wages — it was a demand that no robot enter a Hyundai plant without a prior labor agreement. Those workers were striking against Atlas, a machine scheduled for 2028. They were bargaining early, on purpose, while they still had leverage.

Samsung has just handed every Korean union the next thing to bargain about. The RX plan puts a date on it — 2030 — and a mechanism — “AI autonomous factory” — and a first target: Samsung’s own floors. The unions at Ulsan treated a rollout schedule as the layoff notice it functionally is. Samsung just published one. The only question is whether its own workers read it the same way Hyundai’s did.

Sources

Keep reading

Figure Raised Under $2B. Its New Compute Bill Is $3.5B. Robotics

Figure Raised Under $2B. Its New Compute Bill Is $3.5B.

Figure has raised a little under $2B in equity across its history. On September 3 it signed a compute deal worth $3.5B, expandable past $6B, from a cloud provider that is also becoming its shareholder. The bill for training a robot now exceeds the bill for building the company.

#figure-ai#nscale#humanoid-robots
Tesla Launches Cybercab in Austin With No Manual Controls At All Robotics

Tesla Launches Cybercab in Austin With No Manual Controls At All

Tesla launched Cybercab in Austin on September 3: two seats, no steering wheel, no pedals, no physical way for anyone inside to take over. Texas's automated-vehicle registry hit 420 cars statewide the same day, while Tesla's own July disclosures showed paid robotaxi miles going in reverse, not up.

#tesla#waymo#robotaxi
Mitsubishi ran the robot 1,000 hours, then led its funding round Robotics

Mitsubishi ran the robot 1,000 hours, then led its funding round

Lumos Robotics launched the MOS 2 on September 1. The load-bearing number is not the payload. It is the 1,000 failure-free hours the previous model logged on a live PLC line at Mitsubishi Electric's Changshu plant, after which Mitsubishi led the funding round.

#lumos-robotics#mitsubishi-electric#humanoid-robots