Andy Lonsberry is the CEO and Co-Founder of Path Robotics.
Not long ago, I sat across from an energy manufacturer weighing whether to walk away from a nine-figure contract because they simply couldn’t staff a second shift.
I’ve had versions of that conversation dozens of times. The details change—sometimes it’s a defense contractor, sometimes it’s a power infrastructure company—but the punchline is always the same: The orders are there, but the people aren’t.
The problem isn’t a lack of demand or capital, but that America’s capacity to turn both into physical output is constrained by a shrinking skilled workforce. We are asking the physical economy to scale faster at precisely the moment the workforce needed to build it is becoming harder to find.
This story is becoming increasingly common, and it should fundamentally change how we think about the AI race.
The Role Of Physical AI
Today, the AI conversation is dominated by chips, LLMs and compute. But the real bottleneck is the physical infrastructure behind it all: data centers, grid systems and industrial equipment that must be built to power the AI economy.
We don’t have the workforce to build it fast enough. A Center for Strategic and International Studies analysis found that AI infrastructure could require upwards of 140,000 additional skilled trade workers.
That is where a new category of technology, physical AI, becomes strategically important. Unlike AI systems that operate primarily in the digital world, physical AI enables machines to perceive, reason and act in real-world environments. For manufacturers, that means robots that can take on increasingly complex work and help companies expand production without requiring a one-for-one increase in labor.
I was at CES earlier this year, then NVIDIA GTC. Both times, everyone was talking about physical AI—intelligence in the real world, not just in a model.
But what struck me was how few people in that room had been on a factory floor to see the sheer volume of work required to build the infrastructure this AI boom demands.
The U.S. committed $1.2 trillion to infrastructure modernization with the Infrastructure Investment and Jobs Act. Hyperscalers are approaching $1 trillion annually in data center and AI infrastructure spend. Goldman Sachs estimates data centers could increase power demand globally 50% by 2027, and increase demand by 165% by 2030.
All of that infrastructure has to be physically built. The AI race won’t be decided by who trains the best models, but by who can build the infrastructure to run them.
Trillion-Dollar Demand Meets A Labor Wall
We’re at a once-in-a-generation convergence of demand: grid modernization, hyperscale data centers, defense and shipbuilding and industrial reshoring.
Each is a massive investment wave, and they are all competing for the same materials, supply chains and labor. Take welding as a proxy. It’s where we’ve spent the last eight years building and deploying physical AI at Path Robotics. By 2029, the American Welding Society projects the U.S. needs 320,500 new welding professionals to keep up.
Meanwhile, hyperscalers are generating more than $2.8 trillion in annual revenue. This is real demand, backed by real businesses, driving sustained investment. And we’re shifting from the training phase of AI to the inference phase where these systems actually run at scale. That transition only increases demand for compute and the infrastructure behind it.
But the pipeline isn’t there. Companies are turning away millions in orders because they lack capacity to produce. They can’t hire fast enough or staff second shifts. They have the capital and the contracts, but don’t have the throughput.
When that infrastructure is delayed, consequences are immediate. Transformer lead times, for example, have stretched from 30 to 60 weeks a few years ago to three to four years today. These delays push data center timelines, drive up costs and can trigger financial penalties and market reactions when commitments aren’t met.
The question, then, is not simply how to train more workers. It’s how to increase what each worker and each facility can produce.
A Scalable Path Forward: Physical AI
There is growing discussion about the importance of robotics in this race. That’s directionally right, but it misses a deeper constraint. For decades, industrial robots have excelled in highly structured, repeatable environments.
But most real-world manufacturing doesn’t look like that—parts vary, conditions change and materials behave unpredictably.
I watch a lot of companies in this space give impressive demos. But getting a robot to run reliably in a live production facility is where most of the industry fails.
A different approach is needed: intelligence built into the physical world.
Physical AI can unlock work that was previously considered too complex or variable to automate. It can increase output per worker, reduce rework and enable manufacturers to scale in ways that weren’t possible.
This isn’t about replacing skilled workers. It’s about giving them leverage. When companies can automate more of the work around their skilled employees, those employees can produce more, facilities can run more shifts and manufacturers can take on work they otherwise would have to turn away.
The Factory Floor Is The New Frontline
The AI race is no longer just about compute; it’s about capacity. Chips and models create potential, but it’s factories that determine outcomes.
Over the next decade, the defining advantage in AI will go to whoever builds faster. That means aligning policy, capital and technology around a single goal: expanding the nation’s ability to manufacture at scale.
The opportunity is enormous. So is the risk. The biggest remaining friction point for physical AI is moving it from demo to production. Physical AI has made enormous progress, but manufacturing has an incredibly high reliability bar.
When something fails on a factory floor, the impact on cost and productivity can be catastrophic. A system that works 70% or 80% of the time might make for an impressive demo, but it doesn’t work in production. You need 99%+ reliability across all the variability of the real world.
That’s where the hard work is happening now—making these systems reliable, repeatable and robust enough to operate at industrial scale.
However, if the U.S. wants to lead in AI, it must lead in building the physical systems that power it. That means embracing physical AI, accelerating automation adoption and treating manufacturing capacity as the strategic asset it’s become.
Because in the end, the future of AI won’t be decided in code alone. It will be decided on the factory floor.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?