Leju leverages Pangu via HarmonyOS

The Huawei tie up matters because it lets Leju buy time instead of burn it. In humanoid robotics, getting a robot to hear a spoken command, understand a scene, and turn that into a task sequence usually requires either a large in house AI team or an expensive proprietary stack. By running KUAVO on HarmonyOS and using Pangu for voice, vision, and planning, Leju can focus its own engineering on motors, balance, and deployment workflows where it already has an advantage.

• This is especially valuable in Leju's actual sales motion. Industrial robots are sold through pilots and systems integrators, so faster integration matters more than theoretical model leadership. A customer wants a robot that can follow instructions, connect to building or factory systems, and be reprogrammed quickly, not a bespoke AI stack that takes months to tune.
  1 sacra
• The trade off is that Leju is using a shared intelligence layer while rivals like Figure are pouring capital into proprietary robot models such as Helix. That can make Figure harder to copy, but it also requires much heavier spending. Leju's approach is more like assembling a working commercial product from proven components and then learning from field data.
  1 sacra 3 figure 4 figure 5 figure
• That field data is the real prize. Across humanoids, the key contest is no longer just who builds the robot body, it is who gets robots into real settings soon enough to collect examples of failed grasps, unclear commands, and messy environments. Leju's lower software cost helps it deploy earlier and feed that loop faster.
  1 sacra 2 sacra

The next step is that HarmonyOS and Pangu stop being just a cost saver and become a distribution wedge. If Leju can turn education installs, factory pilots, and smart building deployments into a steady stream of usage data, it can keep the Huawei layer for general reasoning while training more proprietary robot behavior on top, which is where durable advantage in humanoids is likely to form.