Physical AI: The New Frontier in Robotics or an Overvalued Promise?
Reliability in robotics has traditionally been viewed as a hardware challenge. However, the real unresolved issue lies in achieving true autonomy within environments engineers cannot fully control. This distinction is vividly illustrated by Skild AI’s recent funding round, which has captured significant attention. Founded in 2023 and based in Pittsburgh, Skild AI recently raised nearly $1.4 billion in a financing round led by SoftBank, catapulting its valuation beyond $14 billion. This staggering figure comes despite the company generating just around $30 million in revenue in 2025 and not yet delivering a commercially available product in the conventional sense.
Skild AI isn’t building a traditional robot; instead, it is developing the “Skild Brain,” an artificial intelligence system designed to control a wide variety of robots. This approach embodies the emerging concept of “physical AI,” a term used within the industry to describe the ambition of creating a general AI model for robots—akin to how large language models revolutionized natural language processing. Sergey Levine, co-founder of the San Francisco-based Physical Intelligence lab, succinctly described this vision as “ChatGPT, but for robots.” This analogy has helped attract massive investment, with valuations of labs like Physical Intelligence nearly doubling to around $11 billion within a few months. Similarly, humanoid robot startups such as Figure and 1X are securing substantial funding, while technology giants like Nvidia, Google DeepMind, and Tesla are aggressively entering this space.
Data Scarcity: The First Major Hurdle
Despite the excitement, physical AI faces significant challenges from the outset. Unlike language models, which benefit from a vast and growing corpus of human text freely available on the internet, there is no comparable “internet of robot movement.” The data necessary to train these AI systems at scale simply does not exist. To work around this, companies like Skild AI utilize a combination of human video footage and sophisticated physics simulations. However, video clips of humans performing tasks—such as pouring coffee—do not fully capture the complex interactions a robot faces, including friction, weight, and unexpected failures. Simulations, while useful, are imperfect approximations of real-world conditions. Crucially, the most valuable data—the nuanced experiences of robots operating in unpredictable environments like a typical kitchen—have yet to be collected comprehensively.
The Gap Between Demos and Real-World Autonomy
The second, and more critical, challenge lies in the gap between impressive demonstrations and reliable, autonomous operation. Many physical AI companies showcase their technology through polished videos, but some of these have proven misleading. For example, 1X’s NEO humanoid robot was shown performing household chores; however, much of the footage was later revealed to involve teleoperation, where a human remotely controlled the robot. Similarly, a video linked to a Google DeepMind partner was admitted to be entirely computer-generated. This distinction between a controlled demonstration and a fully autonomous, commercial product is well known within robotics circles. Transitioning from a successful lab prototype to a dependable robot capable of functioning consistently in strangers’ homes remains a formidable obstacle. Currently, there are almost no independent, third-party evaluations verifying these robots’ autonomous performance in everyday settings, and the extent to which teleoperation supplements autonomy in early deployments remains largely unquantified.
Why the Promise Remains Worthwhile Despite Challenges
None of these concerns negate the potential of physical AI. Genuine autonomous demonstrations do exist, such as Nvidia’s GR00T humanoid prototype, showcased at their developer conference, which runs its control systems directly on the robot’s hardware. Moreover, research progress in this field is accelerating, driven by real-world demand. Manufacturers and logistics companies facing labor shortages are eager for capable machines to handle repetitive, heavy, or dangerous tasks that human workers increasingly avoid. The opportunity is concrete and urgent.
The crux of the issue lies in valuation and expectations. Wall Street and venture capitalists are pricing these startups as if the most difficult problem—achieving reliable autonomy in complex, uncontrolled environments—has already been solved or is simply a matter of scaling existing solutions. Market analyses estimate addressable markets in the tens of trillions of dollars, reflecting extraordinary optimism. Yet, robotics has a history of teaching caution. The pace of capital infusion, moving at the speed of software investment cycles, often clashes with the slower, physics-bound pace of real-world robotic development. A robot that adapts well in a lab setting may still fail in the unpredictable thousands of unique situations found in a typical household. Additionally, mistakes by robots—such as dropping sharp objects or accidents near children—carry far more serious consequences than errors in software applications like chatbots.
A European Perspective: The Strategic Question
For European observers and policymakers, there is a pressing strategic question. Most leading companies in physical AI are based in the United States or China. Should this technology become the next dominant computing platform, the likelihood is that Europe will primarily be a consumer rather than a creator of this technology. What pathways exist for Europe to assert itself in this pivotal technological arena? As of early 2026, what has been proven is not that autonomous robots are broadly functional, but that capital markets believe they will be soon. European stakeholders must decide whether to patiently wait and observe which of these outcomes materializes or acknowledge that by the time clear proof emerges, critical decisions and investments may have already been made elsewhere.
In summary, the physical AI revolution in robotics holds immense promise but remains in an early, uncertain phase. The immense valuations and capital influx reflect a strong belief in the technology rather than its current capabilities. As robotics continues to mature, the interplay of engineering challenges, market expectations, and geopolitical considerations will shape who ultimately leads in this transformative field.
Source: Here
