The hype around agentic AI pales in comparison to physical AI. Industry experts have described the space, encompassing everything from warehouse robotics and AI-native factories to energy infrastructure automation, as a roughly $50 trillion opportunity, an order of magnitude larger than AI for “knowledge worker” use cases. That projection may well prove directionally right, but investors and executives still need to be realistic about the timelines involved. This is where a patient capital mindset matters most.
Over the course of my career, I’ve gained perspective on the transformative potential of industrial technology from both frontline and executive vantage points. Early on, I worked in the field deploying software updates to industrial vehicles. As I reached under the hood to manually attach my laptop to the Electronic Control Module (ECM), I knew there had to be a better way, yet the enabling infrastructure simply did not exist. Two decades later, we now take for granted software updates delivered over-the-air to electric vehicles, a capability that has helped unlock consumer EV adoption at scale.
This is a clear example of how technology enables equipment upgrades without needing to rip and replace the underlying hardware asset.
As an investor, I now evaluate startups working to drive similar transformation in a wide range of industrial environments. I’ve seen the dramatic difference innovation can make in these settings, and I’m also well aware of the complexity and risk that comes with changing physical equipment, processes, and operating conditions.
I’m entirely bullish about the physical AI opportunity, but clarity about industry fit, CapEx cycles, and what it actually takes to get technology embedded in industrial environments will help investors, executives, and founders understand where and how to invest today.
Physical AI is fundamentally different
Enterprise software environments are built with a certain margin for error. Systems can be debugged, patched, reconfigured, and tuned relatively easily. A serious failure might cause significant financial damage and operational disruption, but a correction path, as well as data backups, generally exist.
In industrial environments, on the other hand, industrial data is often on-premises and failure modes can be physically irreversible. An AI model hallucination causing a robotic arm to misjudge the position of a 500-pound steel beam poses a massive safety liability that can be detrimental to workforce conditions, day-to-day operations, and company reputation.
The production workflows where industrial AI can be deployed are equally unforgiving. Facilities like factories, power plants, and beverage bottling lines are built to exact specifications with finely tailored control mechanisms. Once embedded in production workflows, changes to automation systems are expensive, slow, and high-risk. In a white collar setting, professionals and teams can start slowly with new tools and ways of working and learn as they go. When you’re dealing with heavy equipment and precision engineering, you need to get it right the first time.
The balance sheet doesn’t care about the hype cycle
For all the futuristic promise of industrial AI, its adoption will be shaped in large part by the seemingly mundane realities of capital expenditure and physical infrastructure. Industrial assets like processing plants, commercial HVAC systems, and core food and beverage production lines are durable investments, not software subscriptions. Their project infrastructure and financing are designed around lifecycles as long as 20–30 years, and their operational ecosystem includes dependencies on supply chains for parts covered under warranty. Abrupt changes invite unintended consequences, and when failures happen, you can’t simply roll back an update or apply a patch as you would in enterprise software.
The dramatic efficiency gains driving the $50 trillion narrative will only fully materialize as CapEx cycles renew and facilities are redesigned around AI from the ground up. For heavy industrial infrastructure, that can mean waiting for a quarter-century. The near-term opportunity is the shorter cycle nested inside that longer one: control systems, instrumentation, and digital layers turn over every 5 to 15 years, creating recurring upgrade windows where insertion into existing operations is feasible without requiring a full infrastructure overhaul.
Targeting the brownfield window
For executives building technology strategies and investors allocating capital, the greenfield narrative where AI-native factories are built from scratch describes a small fraction of the actual market. The overwhelming majority of industrial infrastructure is brownfield and will remain so for decades. The near-term value lies in identifying the companies successfully embedding themselves in today’s operations, not in betting on fully transformed facilities that don’t yet exist. These companies are delivering real efficiency within existing constraints, building trust inside industrial organizations, and collecting field data that compounds in value over time. By proving value in today’s operating conditions, they are earning the right to redesign the workflows of tomorrow.
SE Ventures’ portfolio companies Augury, UnitX, and Axion illustrate this model. Each company has embedded AI into existing factory operations, Augury for predictive machine health, UnitX for in-line manufacturing inspection, and Axion for product quality issue detection. Rather than designing systems predicated on an infrastructure overhaul, they created relevance within existing customer environments. The customer relationships and proprietary data assets they’re building now will position them as the natural technology providers when renewal cycles arrive.
Deployment in existing production environments also generates rich training data that lab-optimized competitors can’t replicate. For example, the foundation model for the industrial robots built by SE Ventures’ portfolio company Skild AI is trained on both large-scale simulation data and targeted real-world data. This grounding in actual operating conditions helps the model develop real physical awareness and respond more effectively to hardware changes or failures.
This staged approach also creates space to address the human side of AI transformation as well. Industrial AI can’t adopt itself; organizations will still need human workers to deploy and work side-by-side with these systems. Members of the current factory workforce can upskill into new roles as knowledge workers for the next generation of industrial technology.
Moving today to capture tomorrow
While top-to-bottom transformation will play out over decades, not years, companies that take a more targeted approach can drive value much sooner while earning trust and credibility for the long term.
For industrial AI founders, short-term success will depend on understanding the budget cycles of your customers and designing your GTM approach accordingly. Real-world training data is key; the sooner you can embed in customer environments, the better. But watch out for the proof-of-concept graveyard. If you find yourself in a holding pattern for six months or more, the odds of moving forward are fading.
For CXOs seeking to deploy industrial AI, it’s important not to lose momentum waiting for the perfect solution to materialize. A risk-averse mindset can leave you stalled in the status quo while competitors iterate quickly with new technology and gain ground. Instead, identify the pain point you want to address, gain experience as you proceed, and build on your learnings as you expand the role of industrial AI in your operations.
The $50 trillion industrial AI opportunity that capital markets are chasing will take longer to play out than anyone wants to admit, and many companies designed for sweeping transformation will struggle to reach that horizon. But patience does not imply inaction. For founders and CXOs who take a strategic approach grounded in the practical realities of these operating environments, the future begins today.
The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.