Flex is both the manufacturer and largest customer of Teradyne Robotics' Universal Robots and Mobile Industrial Robots, creating an internal feedback loop where deployment learnings directly inform product development. This supplier customer duality model aims to solve the…
Flex is the company that makes the robots. It is also the company running them.
That is the actual story in the Flex and Teradyne Robotics partnership announced April 22 — not another press release about the promise of physical AI, not another announcement that two companies will collaborate. According to Flex's investor relations release, Flex is simultaneously Teradyne's largest component manufacturer for its Universal Robots cobots and Mobile Industrial Robot autonomous mobile robots, and its single biggest customer deploying those machines across its own global footprint. One hundred facilities. Thirty countries. One hundred and forty thousand employees. The company that makes the robots is running them on its own factory floors, then feeding everything it learns back into the next generation of the product.
The robotics industry has a scale problem. Not a capability problem — the robots themselves have gotten genuinely good. A problem of translation: getting those robots to work reliably, consistently, and profitably across a customer's actual operations, in an actual factory, with actual workers, in Malaysia and Mexico and Ohio on the same day. BCG's physical AI analysis puts it plainly: the gap between what robots can do in a demo and what they do in a live production environment is where most automation investments quietly die. The Flex-Teradyne model is a bet that the fastest path through that gap is to build the robot, run the robot, and debug the robot inside the same company.
The dual-track strategy is structurally unusual. Teradyne Robotics — the parent of Universal Robots and MiR — has other customers who are deploying its cobots and AMRs. Those customers buy from Teradyne, deploy in their own facilities, and figure out the operational kinks on their own. Flex is doing something different: supplier and customer simultaneously. The feedback loop that other manufacturers negotiate across a vendor relationship happens internally. Problems Flex encounters deploying MiR AMRs in its own facilities get incorporated into the next manufacturing run of MiR components that Flex itself is producing. There is no handoff friction. There is also no independent validation.
The 20-year history between these two companies matters here. This is not a new marriage. Flex has manufactured semiconductor test equipment for Teradyne since the early 2000s, a business where precision and uptime are existential. The trust required to extend that relationship from test equipment into live factory floor robotics is not trivial. Dennis Kirkpatrick, Flex's president for lifestyle, consumer devices, and core industrial, said in the press release: "Expanding our relationship into robotics and intelligent automation builds on a strong foundation." The foundation is a two-decade track record of delivering complex hardware at scale. The new part is that Flex is now also the proving ground.
Jean-Pierre Hathout, president of the Teradyne Robotics Group, framed it as a validation play. "Flex's experience in manufacturing complex products across industries, combined with its global scale and resilient supply chain, makes it an ideal partner for advancing intelligent automation," he said in the announcement. The word "ideal" is doing work there. What Teradyne gets is a reference customer running its robots across a genuine global footprint — not a pilot program, not a press-release deployment, but production use across Flex's actual facilities. That is a different kind of proof point than a demo or a case study.
The physical AI context is worth sitting with. Physical AI — a term popularized by Jensen Huang at Nvidia — refers to systems that perceive, reason about, and act in the physical world, as opposed to generating content or processing text. A cobot that adjusts its grip based on the shape of an object it has never seen before. An AMR that reroutes itself around an unexpected obstacle without pausing production. These are not science fiction capabilities, but they are also not solved problems in every environment. BCG's analysis notes that improvements in perception have advanced rapidly, while dexterous manipulation — the kind of fine motor control humans find trivial — has proven far harder to scale. The consequence is a robotics landscape full of impressive demos and a much smaller number of systems operating reliably at commercial scale.
The humanoid market projections illustrate the uncertainty. BCG cites analyst projections for humanoid robotics by 2030 ranging from under one million annual units to more than six million. That is not a forecast — it is a measurement of how completely the experts disagree about what is actually going to happen. Flex and Teradyne are not betting on humanoids. They are betting that the more immediate opportunity — getting today's cobots and AMRs to work consistently across a real global manufacturing operation — is solvable without waiting for the humanoid question to resolve. It is a lower-risk, lower-ceiling bet than the humanoid race. It may also be the more honest one.
The skeptical case is straightforward and the press release does not answer it: how many robots are actually running in Flex facilities right now? What is the uptime? What specific design changes has the internal deployment produced? The announcement language is aspirational — continuous operational feedback, validating robotics technologies at scale, faster replication of successful automation workflows. These are the phrases of a marketing department, not an operations report. The actual numbers are not in this press release and may not exist in any public filing. The closed feedback loop is a compelling structural argument. Whether it is also a genuine operational reality is a question the press release leaves open.
The broader implication is where this gets interesting for the industry. If Flex's model works — if the builder-deployer convergence genuinely accelerates the translation from pilot to production-scale deployment — it reshapes the competitive landscape in industrial automation. The bottleneck stops being robot capability and becomes access to real-world operational data at scale. Pure-play robotics vendors without captive deployment networks would face a structural disadvantage: they can sell into Flex's model, but they cannot replicate Flex's data advantage. Contract manufacturers like Jabil and Foxconn would face pressure to develop similar internal deployment capabilities. The robotics industry's scale problem, reframed as a data access problem, produces a very different set of winners than the scale problem reframed as a hardware capability problem.
For now, the story is simpler than that. Two decades into a relationship, a contract manufacturer and a robotics company have decided that the best way to prove robots work is to build them and run them in the same place. Whether that makes Flex the Rosetta Stone of industrial automation or just a very well-marketed guinea pig is the question its factory floors will eventually answer.