Why physical AI is the next profitable frontier

The fastest growing market categories have humble beginnings. Physical AI, the combination of robotics and advanced artificial intelligence, is just the latest example.

ChatGPT started as a pitch at an accelerator before becoming a household name in 2023. Before cloud computing became a boardroom imperative, the founders were building it, server rack by server rack, cluster by cluster. And, of course, before fintech rewired retail banking, it showed up as a surge in funding in deal flow data that most banks weren’t reading.

You see the pattern, and I’ve seen it repeat itself over several years of evaluating startups. The startup ecosystem is the most underutilized business intelligence system in the world.

Incumbents may accumulate wealth and power, but it is the founders who choose to build. Which industries attract a sudden wave of founding activity is a three- to five-year leading indicator of where business disruption is headed.

As AI agents garner the lion’s share of media attention and VC funding, deal flow suggests the next hypergrowth market will be of the metallic persuasion. This is physical AI, which includes autonomous machines and smart manufacturing, among other sensor-laden systems.

According to PitchBook, robotics and physical AI startups raised a record $27.6 billion in 1,009 deals in 2025, which is more than double the previous year. Genesis AI, which emerged in stealth mode with a $105 million seed round, recently demonstrated a robot on wheels capable of bringing physical AI into commercially relevant environments without the need to replicate the human body.

Physical AI: internal intelligence

Physical AI systems can sense environments, understand spatial context, make operational decisions, and act autonomously in real time in industrial and other contexts. Physical AI systems move beyond static automation to adaptive operations that continuously improve through interaction with the real world.

Like most technology trends, physical AI is driven by the coalescence of several technologies that matured simultaneously. These include: multimodal sensing, edge computing, simulation environments, basic robotics models, and world modeling systems. These technologies allow machines to gather information from the physical environment, interpret it in context, simulate outcomes before acting, and continually refine performance through operational feedback loops.

Additionally, the costs of several components that make physical AI possible have fallen, with lidars, actuators, depth cameras and sensors becoming more affordable.

For example, Waymo said its sixth-generation Driver system significantly reduced costs while increasing resolution, range and processing power. The autonomous vehicle system used a suite of 13 cameras, four lidars and six radars. Together, technological maturation and cost reduction create the right conditions for physical AI to flourish.

The manufacturing, automotive, logistics and healthcare sectors have emerged as major areas of capital concentration, with organizations actively piloting robotics to address labor costs and shortages of skilled workers, and deploying them in challenging environments even as productivity needs grow.

Figure AI, whose humanoid robots assemble car parts in BMW factories, has reached a valuation of $39 billion in three years, the fastest path to unicorn status in the history of robotics. Investors are betting on general-purpose robot models that do for hardware what large language models have done for software. Why shouldn’t they? The business flow is there.

Three phases of a disturbance wave

Over the past few years, as I’ve read through deal flow, I’ve identified a consistent three-phase pattern that manifests itself in every major technology disruption.

Phase 1: The Founders’ Surge

A new capability emerges, attracting a disproportionate number of high-quality founding teams. Capital is concentrated. Failure rates soar even as survivors define the category. As I mentioned earlier, generative AI has hit the stars in 2023. Agentic AI is now going through this phase, while physical AI is just entering it.

Dyna Robotics, which is building basic models that allow a robot to learn service tasks for hotels, restaurants and laundromats, raised a $120 million Series A funding round within about a year of its founding.

Phase 2: The corporate adoption wall

Businesses begin to experiment with the technology, but struggle to move from experimentation to production. Startups that survive the first phase hit sourcing cycles, integration complexity, and organizational resistance. According to Deloitte, we’re seeing this right now with organizations implementing GenAI and AI tools without adapting their operating models.

Some organizations break through this wall. After recognizing that generative AI would fundamentally reshape customer service, Intercom canceled non-AI projects, reorganized product teams, and committed $100 million to retool the business around AI. This transformation produced its Fin agent which now resolves millions of customer service requests.

Another company, Maven Robotics, has used mixed palletizing in warehouses and is going through the wall one customer at a time.

Phase 3: industrialization

A subset of companies crack the distribution code. Technology becomes operational infrastructure. The window for competitive entry is closing. Cloud computing reached this stage around 2015. GenAI and agent AI will arrive at this stage within 18 to 24 months as organizations redesign their processes and double down on change management practices.

Kiva Systems’ warehouse robots, purchased by Amazon in 2012, have grown into a fleet of more than 1 million. What was once a startup capability is now a sophisticated infrastructure that new entrants cannot easily match.

The practical implication is simple: by the time a technology reaches Phase 3, the structural advantages of early distributors are largely locked in. Companies that used AWS in 2010 to save money ultimately built strong engineering cultures, data architectures, and product velocity in the years since.

Physical AI today, something else tomorrow

The same dynamic is playing out in AI today, and deal flow tells us exactly which categories are on that curve. For physical AI, the competitive advantage lies in the ability to build continuous operational learning loops where physical interaction generates data, data improves simulation, and improved models and models improve real-world performance.

As the deployment of these intelligent systems increases, the benefits through infrastructure integration, operational feedback and real-world learning increase. More data, operationalized appropriately, produces better results.

Most business leaders treat startup deal flow as noise. They may view it as too early, too speculative, or too distant from their operational reality.

But instead of looking at the flow of startup deals to find your next supplier, look at it to understand where the next generation of your competitive environment will be built. Then calibrate your development and strategic roadmap accordingly. The signs are there; you just need to know how to read them.

About the author

Janis Skriveris is a principal at Plug and Play Ventures and director of its Deeptech platform. Hand researches, analyzes and drives investments in deeptech and artificial intelligence companies shaping the next generation of industrial and urban systems.

Skriveris focuses on applied artificial intelligence, robotics and autonomy, next-generation mobility, industrial technologies and the built environment: real estate, construction and infrastructure. Helps companies validate business traction, refine go-to-market strategies, and access Plug and Play’s global business network. Since 2017, Skriveris has led 37 investments, including unicorn Einride.

Prior to Plug and Play, Skriveris worked in EY’s Assurance and Consulting department, where he developed a deep understanding of various industries and honed his skills in strategic planning, risk management and business transformation. He holds a bachelor’s degree in economics and entrepreneurship, graduating with honors from the University of Latvia. Janis holds a master’s degree in finance and financial management services from the Faculty of Business and Economics of the same university.