At SXSW 2025, I attended a featured session titled “From Cages to the Real World: The Dawn of Physical AI.” This panel discussion focused on the transition from the recent boom in AI infrastructure to an era of widespread autonomy driven by physical AI. This change is expected to make robotics more accessible and promote broader adoption of AI in consumer and industrial applications.
The session was moderated by venture capitalist Jenny Stojkovic and featured insights from Dr. Anirudh Devgan, CEO of Cadence; Sir Tim Berners-Lee, inventor of the World Wide Web; Tye Brady, chief technologist at Amazon Robotics; and Dr. Andrea Thomaz, CEO of Diligent Robotics. The panel explored the opportunities and challenges within this emerging field.
(Note: Amazon and Cadence are advisory clients of mine at Moor Insights & Strategy.)
Physical AI Defined
Physical AI encompasses far more than just robotics; it extends to autonomous vehicles, drones and various other autonomous systems that can interact with and manipulate the physical environment. During the introduction of the panel session, Devgan laid out the progression of AI development through distinct evolutionary phases; the current infrastructure buildout phase is focused on datacenters, and is already being followed by the physical AI phase, which is expected to mature within two to seven years. Eventually we can expect a sciences AI phase, projected for five to ten years from now. This progression represents what many describe as the union of “mind and body” — the culmination of 50 years of parallel development in computing and physical systems. As Devgan noted, “This is the moment where AI moves from cages to the real world.”
The shift from purely digital AI to systems that can perceive, reason about and interact with physical environments marks a transformative development, one with multi-trillion-dollar implications across industries, including automotive, healthcare, defense and manufacturing. However, this transition also raises concerns about potential negative impacts.
Applications Of Physical AI Already In Use
Physical AI is already being used in multiple industries, with real-world applications producing substantial operational benefits. For example, Amazon’s extensive robotics ecosystem, with more than 750,000 drive units globally, processes approximately 75% of all Amazon packages using robotics and automation. Brady described this integration as a “symphony of humans and machines” collaborating to achieve greater efficiency, and emphasized that this partnership is crucial for Amazon’s operational scale.
Although there is a widespread fear of robots and AI replacing humans in the workforce, Brady stated his belief that “We have to put people at the center of the robotics universe.” This means that we need to “reframe [our]
mindset with machines — that machines allow us to be more human [and] should allow us to be more capable.” He added that the burden falls to roboticists like him “to make [machines] simple and easy to use, to make them have utility that allows you to perform.” The goal is ultimately to foster a “harmonious relationship of augmenting human capability with machines to do the job.”
In healthcare, Diligent Robotics has sought to strike that balance by deploying its “Moxi” robots to reduce the workload on healthcare workers. Moxi is a robot designed to perform delivery tasks in hospitals. The company’s fleet of 100 robots has completed about a million transport tasks, equivalent to 1.5 billion steps that nurses and pharmacists didn’t have to take.
Thomaz described Moxi as a robot with wheels, storage drawers, a torso and a “cute head” designed to be “socially appropriate and functional.” The Moxi robots leverage generative AI to perform transport tasks autonomously. Similar to how models such as ChatGPT process language and image data from the Internet, robotic AI processes data from multiple onboard sensors. These robots use multimodal sensors such as cameras, LiDAR and other perception tools positioned in ways that capture unique perspectives not typically found in online data. This fits with a broader trend for physical AI, because training robotic AI models commonly involves gathering and processing data specific to each robot’s environment and tasks.
As robotic AI advances, robots are becoming more capable of perceiving, understanding and making quick decisions in complex, dynamic environments. This technological shift is enabling robots to be deployed in spaces with high human activity, where they must interact seamlessly with people. One of the most promising applications is workforce assistance, particularly in industries facing labor shortages, such as healthcare. By working alongside humans rather than replacing them, robots have the potential to enhance productivity and improve essential services. “Robots plus people working together,” Thomaz said. “That is going to be a big part of how robots can have a huge impact on society.”
Economic Impact And Potential Industry Transformation
The transition from digital AI to physical AI systems interacting with their environments is a transformative development with multi-trillion-dollar implications. This shift includes significant changes in the automotive sector, where autonomous vehicles and related technologies could generate $3 to $4 trillion in annual economic value. There are also significant national security implications. For example, AI-powered drones can enhance situational awareness, surveillance and threat response with increased speed and precision while reducing risks to human life.
Devgan predicts that in the second phase of physical AI, new applications will need to yield economic impacts measured in the trillions of dollars, given that technology companies are already investing hundreds of billions of dollars. He also believes that there is a significant possibility that physical AI will indeed become that substantial in scale.
Tackling The Technical Challenges Of The Physical AI Ecosystem
While all of this sounds good at a high level, there are countless knotty problems of engineering to solve along the way to the physical AI future. Cadence is a good company to talk with in this respect, given that its design software — driven by math, physics and computer science that’s about as complex as it gets — is used to design electronic and mechanical systems ranging from semiconductors to aerospace components. After SXSW, I had a chance to speak with KT Moore, Cadence’s head of marketing, about the company’s perspective on the computational demands of physical AI systems.
Moore emphasized the importance of vendors like Cadence supporting the computational infrastructure required for physical AI. The growing prevalence of physical AI applications is expected to increase compute requirements, and Moore pointed out that this trend could in turn increase demand for silicon, systems, memory and other machines. “As long as people are building machines,” he said, “we [at Cadence] have a job and a market.” Besides the growth tied directly to physical AI, there could be expansion into related sectors such as drug discovery.
For Cadence specifically, this is likely to play out in new uses for its design tools; for example, its ability to generate synthesizable RTL for microcontrollers and developing test verification benches can provide a foundation for AI hardware development. We’re also likely to see the creation of specialized solutions for domain-specific AI applications. This could extend to the use of its digital twin technologies for new physical AI uses in the aerospace, defense and automotive sectors.
Governance Considerations For Physical AI
As physical AI integrates into daily life, ethical considerations such as data security, equitable integration and robotic decision making must be addressed. This is why well-positioned industry observers including Berners-Lee have called for a unified global framework for AI governance to ensure that AI serves individuals and society responsibly. As he cautioned at SXSW, the key question is, “Who does the AI work for?” His point emphasized the importance of AI serving individual users’ interests rather than only corporate or commercial agendas.
The absence of a unified global framework for AI governance presents a significant challenge to the responsible deployment of physical AI. Without standardized protocols, it’s hard to achieve the reliability required for mission-critical applications where even a fraction of a percentage point in downtime can have severe consequences. Data privacy and security issues further complicate this landscape, again pointing to the need for robust safeguards to prevent misuse. AI systems must operate with clear accountability to truly meet human and societal needs, and in this context trust and transparency must be foundational to realize AI’s full potential while mitigating systemic risks.
The Physical Future
The emerging field of physical AI offers many business opportunities, potentially generating trillions of dollars and transforming industries. However, companies like Cadence, Amazon and Diligent Robotics must continue to approach this economic growth with a clear awareness of the significant societal changes it will bring.
As robots become more integrated into our daily lives, from healthcare to logistics, we must proactively address both the technological challenges and the human aspects involved. During the SXSW panel, all the participants agreed that it is crucial to understand and facilitate the changing relationship between people and robots, ensure equitable access and establish strong ethical frameworks.
The companies that will lead this revolution must innovate technologically while also effectively navigating the complex interactions between business potential and societal impact. Doing so should help create a future where physical AI progress is both profitable and centered on human values.
Moor Insights & Strategy provides or has provided paid services to technology companies, like all tech industry research and analyst firms. These services include research, analysis, advising, consulting, benchmarking, acquisition matchmaking and video and speaking sponsorships. Of the companies mentioned in this article, Moor Insights & Strategy currently has (or has had) a paid business relationship with Amazon and Cadence.
