The next decade may look like a hardware boom
Physical AI is likely to attract enormous capital into bodies, actuators, sensors, factories and compute. That is necessary infrastructure. But infrastructure build-out and durable economic power are not the same thing. The companies financing the first wave of machines may create the installed base on which later software and application companies build higher-margin businesses.
Hardware profits are possible, but hardware-only moats are fragile
FlyPig AI's thesis is deliberately narrower than saying hardware never makes money. Exceptional hardware companies can create strong margins through scale, brand, proprietary technology or ecosystem control. The risk appears when a category matures, supply expands and buyers can compare increasingly interchangeable alternatives. At that point capital intensity remains high while differentiation can fall.
The stack is separating
Physical AI can be viewed as a layered system: foundation models, runtime and middleware, edge compute, robot bodies, reusable skills, agents and vertical applications. A company may integrate several layers, but the analytical question is where switching costs, data advantages, developer adoption and customer workflow ownership accumulate as adjacent layers standardize.
Vertical context is difficult to commoditize
A mine inspection workflow, warehouse exception process and healthcare assistance task have different safety, integration and operating requirements. The company that understands the job, connects the robot to existing systems, captures exceptions and improves the workflow over time can build an advantage that is not easily reproduced by a generic hardware vendor.
FlyPig is Physical AI, but not merely physical
FlyPig AI is interested in robot hardware because hardware determines what is physically possible. But our longer-term focus is the intelligence layer above it: architecture selection, reusable capabilities, agent orchestration, deployment knowledge and software that turns machines into repeatable operating outcomes. We do not want to predict only which robot body wins. We want to understand who becomes the Android of Physical AI, who defines the skill layer, and which applications become the first category-defining businesses.
The investment question is therefore different
Instead of asking only which humanoid, quadruped or mobile robot will ship the most units, the more durable question may be which platform becomes the common development surface for all of them. If that layer emerges, the hardware boom will have created something more consequential than machines: a new installed base for software.