Two Spot robots have walked nearly 900 kilometers inside GlobalFoundries' Singapore plant since April. They completed more than 90,000 inspections on the sub-fab floors, the levels beneath the cleanroom where pumps, gas lines and utility equipment feed the production tools. Boston Dynamics puts the mission success rate at 99%.
That record is why a humanoid is now in the room. On November 17, Boston Dynamics and GlobalFoundries said Atlas will begin on-site testing at GF fabs in early 2027. GF becomes the first semiconductor manufacturer in the Atlas early adopter program, and the first customer outside Hyundai, Boston Dynamics' owner.
Our read: Spot earned this trial. Atlas has not earned anything yet.
What four years of Spot bought
GF has run Spot at its Burlington, Vermont fab for more than four years. Boston Dynamics published a case study on that deployment in 2022, built around a simple argument: sensorizing every pump and gauge in a vast facility is uneconomical, so the sensors should travel to the equipment instead.
The new agreement spreads the quadruped to Malta, New York; Dresden, Germany; and Singapore. The Singapore numbers are the ones a fab operator will study. Two robots, roughly 900 kilometers, 90,000 inspections, 99% mission success. GF also moved 875 routine inspections onto Spot in two weeks and says it can more than double that count with the robot's visual inspection features.
Spot has helped us shift from reactive to predictive maintenance, collecting data across thousands of inspection points to prevent downtime and keep our fabs running more efficiently.— Pradip Singh, chief manufacturing officer, GlobalFoundries
Inspection is a narrow job with a clean metric. A robot walks a route, reads a gauge, records a thermal image, and files the result. The payoff shows up as avoided downtime, not as extra output. Nobody expects Spot to fix the pump.
Atlas arrives with a blank task list
The Atlas program starts as a workshop. The two companies will evaluate use cases, check integration requirements and customer readiness, then train the robot on specific tasks. Early 2027 is a testing date, not a deployment date.
GF has not said how many Atlas robots will be involved, which fab hosts the first one, what job it will try, or what the program costs. Those omissions are the story.
As we wrote in November, the humanoid buyer's guide is a five-subsystem trade-off sheet. The GF program does not yet answer any of the five.
Open: Atlas unit count, first task, host fab, price, production timetable.
Atlas has spent 2026 inside Hyundai's orbit. Boston Dynamics said at CES in January that the year's production run was committed to Hyundai Motor Group and Google DeepMind, which is integrating Gemini Robotics models into the robot. The Hyundai Metaplant application center opened in September, with parts sequencing and material handling as the first training tasks. Hyundai has described a plan for 25,000 Atlas units across its plants, with component assembly as a 2030 target.
The hardware is still moving. In October, Boston Dynamics showed a new Atlas hand with four fingers and 13 degrees of freedom, up from seven. The company also named Rohit Prasad, formerly of Amazon's Alexa team, as chief executive that month. A platform that changes hands and end effectors while it enters its first external trial is still being defined.
Why a chip fab is a harder first customer
Automotive plants tolerate a robot that stops and asks for help. A fab running around the clock has less room for that. Cleanroom particle limits, chemical handling, and process tools that cannot be idled without cost all raise the price of a mistake.
That cuts both ways. If Atlas can work in a fab, the integration story gets credible fast. If every site needs a bespoke safety case and a custom task library, the humanoid business starts to look like systems integration with a hardware margin attached.
GF's own framing is cautious. Singh calls the Atlas test the next step in "AI-enabled manufacturing," a phrase that promises learning, not output.
The first non-automotive customer is a data point, and the useful questions are unglamorous. How often does a person intervene? What happens after a fault? Does the robot finish a shift without adding work for the technicians around it?
Spot answered those questions over four years, one inspection route at a time. Atlas gets a workshop and a date. The invitation is real. The job description is still blank.