How Humanoids Are Moving from Prototypes to Production: Jabil's View

From Demos to Factories

While the media focuses on flashy videos and massive funding rounds, the real work of bringing humanoids into production is happening quietly. Jabil Inc., a design and manufacturing services company, shared how robots are transitioning from prototypes to large-scale output.

Tesla, 1X Technologies, XPeng, and UBTECH are all active in this space. Jabil consults for developers and helps bring products to market at high volumes. The company works especially closely with Apptronik on the Apollo humanoid in a manufacturing context.

Moving into the Tens-of-Thousands Phase

When volumes grow from low (on-site manual assembly) into the tens-of-thousands phase, all the rules change. You need new approaches to cost reduction, manufacturability, and quality. "When a robot is very expensive, you can produce a limited run to collect AI training data, but you still need to validate real-world applications. We help lower the cost of the hardware and reach the required volumes," explained Thomas Brown, Senior Director of Engineering and Technology at Jabil.

Growing compute power is opening up new possibilities. NVIDIA Jetson is a popular platform among humanoid developers. But edge computing remains a bottleneck: new levels of performance and memory are needed, and pricing remains a scaling challenge.

When Hyundai, Tesla, and BMW fund humanoid programs, they bring automotive engineering experience with them. Wiring harnesses need to combine automotive-grade reliability with the micro-scale sizing of consumer electronics. Developers need partners capable of fast iteration and a wide range of swappable power source options.

Where Humanoids Make Sense

In manufacturing and warehousing, humanoids change the economics. Existing automation is specialized, but the versatility of humanoids allows processes to be reconfigured through software instead of retooling entire lines. Companies partner with clients to gather clear requirements — in the consumer segment, requirements remain vague, like "a robot that folds laundry."

Do humanoids need fully human-level capabilities? No. Brown believes the right approach is building a superset of capabilities tailored to different markets. Some companies already use wheeled bases instead of bipedal walking — it's simpler and cheaper. Companies that master the entire stack will be able to scale everywhere; specialists in a single component will fall behind. The market will support several players: manufacturing, domestic use, hospitality, food service. Each niche demands its own approach.

Mobile manipulators have their own place too. Traditional automation running at 1.5 m/s (3.3 mph) still works efficiently. Needs vary — a humanoid is versatile, but it's not always the right choice. High-speed automation will retain its place for years to come.

Consumer robots are possible within five to twenty years. If the current pace of model and compute development continues, the AI and electronics side will be ready. Variation in performance levels and tool types leaves plenty of room for specialization.

Once humanoid fleets become widespread, factory line design approaches will change. Layouts could become programmable — send instructions, and robots will reconfigure stations, swap hand tools, and retrieve materials themselves. The line will stop being the key constraint — it will become the computational substrate of manufacturing.

Safety and Standards

Robotic arms already carry safety certifications, distance norms, and guarding requirements. Humanoids are a new class that requires new standards. The industry needs to work through failure modes, certification methods under new conditions, and behavior across various scenarios. The industry has begun tracking regulatory developments and drafting standards for both commercial and consumer applications. It's a long process, but a necessary one.