Robots need a lot of compute to learn in virtual environments, and they also need fast, reliable chips inside the machine so they can see, decide and move in real time, without waiting on the cloud.
As we move into 2026, markets are shifting their focus from the architects of AI (i.e. model labs and hyperscalers) to the enablers (infrastructure) and adopters (applications). Market leadership so far this year shows this trend, with a continued rotation away from the Mag 7 into other parts of the market as investors seek the next layer of opportunity.
Robotics may be the next frontier for AI, where intelligence leaves the webpage and enters the physical world, creating transformative new use cases. But the question for investors is primarily whether and when it becomes commercially viable.
The technological unlock
Historically, the binding constraint in robotics wasn’t hardware, but data.
Unlike LLMs that consume vast data from the internet, robots couldn't learn to fold laundry or turn a doorknob from a vast database. Robots had to be taught one task at a time through painstaking human demonstration1, but that bottleneck is finally breaking:
- With AI-enabled simulation training, robots can now "practice" in virtual environments millions of times before ever touching a physical object.
- We have moved away from programming specific actions to training general-purpose motion. A model that learns how to wipe a counter can now generalize that skill to scrubbing a window.
- Agentic AI allows robots to navigate and adapt to new environments, like a busy warehouse or a messy kitchen, with greater autonomy.
The economics are also improving. Humanoid robot costs declined an estimated 40% between 2023 and 20242. In an environment shaped by labor scarcity and pressures to reshore manufacturing to the U.S., robotics can help boost efficiency and protect margins—particularly given U.S. manufacturing wages are roughly four times higher than China’s3.
The commercialization challenge
Even with technical progress, meaningful revenue from AI-powered robots is still likely years away. Scaling production will depend largely on fragile supply chains, manufacturing capacity, safety considerations and the economics around deployment and maintenance.
Tesla, for instance, has fallen short of its own ambitious Optimus production goals4, and progress in autonomous vehicles, despite recent commercial success, remains constrained by practical bottlenecks like vehicle supply and integration logistics.
What’s investable now
For now, the investable part of “physical AI” is still largely at the infrastructure layer. Robots need a lot of compute to learn in virtual environments, and they also need fast, reliable chips inside the machine so they can see, decide and move in real time, without waiting on the cloud. The software behind physical AI training is also proliferating, while manufacturing and healthcare firms have been the strongest early adopters.
Elsewhere, a growing number of U.S. tech companies and startups are pursuing AI robotics, but uncertainty around timelines, unit economics and commercial viability make it premature to pursue as a standalone theme in portfolios. It does however offer investors another reason to invest internationally, as Asia remains a key leader with its manufacturing prowess and head start advantage5.
The promise of robotics also highlights how early AI deployment still is. The most transformative business models in this space may not even exist yet. Investors would do well to maintain broad exposure to the enablers and adopters this year, rather than try to pick winners prematurely, as this landscape evolves.