Let’s Get Physical AI

Why the next wave of AI isn’t happening in the cloud, but on your warehouse dock.

By Eric Ragan – Director, Robotic Systems & Automation, Goddard Product Development

AI has been living in the Digital World. It’s time to get physical.

For the past three years, artificial intelligence has dominated every conference agenda, trade publication, and executive briefing in nearly every industry. And for most of that time, the conversation has centered on the same category of tools: chatbots, content generators, code assistants, and data analytics platforms. Useful? Absolutely. Transformative for your DC floor? Well, maybe not so much.

Meanwhile, your warehouse is still running the same AGV logic it had when installed 10 years ago. Your picking lines still depend on tribal knowledge and headcount. Your automation roadmap is slowly becoming a shadow on your office whiteboard.

The uncomfortable truth is that AI has largely been navel-gazing, generating impressive outputs in the digital world while the physical world of warehousing, distribution, and fulfillment has waited for something it can actually put to work.

That wait is ending… maybe.

| “The ChatGPT moment for physical AI is here.” — Jensen Huang, NVIDIA CEO, CES 2026

When NVIDIA’s Jensen Huang made that declaration at CES in January 2026, it wasn’t just a product announcement. It was a signal that AI once confined to research labs and tech demos is crossing into mainstream commercial deployment and the proving ground is your factory floor, your fulfillment center, and your distribution hub. But what the heck is he talking about?

What Exactly Is “Physical AI?”

Before we go further, let’s be direct about the idea of a definition, my background in academia has taught me terms get thrown around in ways that obscure more than they explain most of the time.

So, here is my, not official but dang close definition. Physical AI is artificial intelligence that doesn’t just think, it acts. It perceives the real world through cameras, sensors, and depth systems, reasons about what it observes, and takes physical action, moving, gripping, navigating, and adapting, without being explicitly pre-programmed for every scenario it might encounter.

The contrast to traditional industrial automation is important. Industrial robots of the past were rule-based: explicitly programmed to execute the same task with precision and speed, in exactly the same environment, every time. For high-volume, low-variability applications, automotive stamping, electronics assembly, palletizing standard cases. Those systems remain effective and will continue to evolve into the future.

But the moment something changes, a new SKU, a repositioned conveyor, an irregularly shaped package, traditional automation gets in trouble. The environment must conform to the robot.

Physical AI inverts that relationship. Machines learn from simulated and real-world experiences, making them adaptable to mid-volume and even non-repetitive tasks. Their training can be virtualized, dramatically reducing deployment time and widening the scope of what can be automated.

The shorthand: old robots followed a script. Physical AI robots read the room.

The evolution moves through three distinct stages:

• Programmed Automation: Rule-based robots that execute fixed programs in structured environments.
• Physical AI (Training-Based): Training-based robots that learn from simulation and real-world experience, enabling adaptability to variation.
• Physical AI (Context-Based): Context-based robots equipped with foundation models that integrate vision, language, and action, capable of receiving a goal (“unload this trailer”) and planning the steps autonomously. I know it’s magical!

The warehouse industry is actively moving through all three stages simultaneously, and the gap between them is closing faster than most operators realize.

Your Distribution and Fulfillment Warehouse Is the Ideal Proving Ground

When analysts look across industries for where physical AI is gaining the most traction, the answer keeps coming back to manufacturing, logistics, and distribution. Not because these sectors are the most glamorous, but because they have a structural advantage that’s easy to overlook.

The distribution center is a controlled environment. It’s not a public street, a hospital corridor, or an unstructured outdoor space. The operating domain is defined, the tasks are repeated at scale, and the ROI pressure is immediate and measurable. Physical AI applications in these environments progress significantly faster than deployments in open, unpredictable settings, and that’s exactly why warehousing and fulfillment are leading adoption globally.

58% of companies are already using physical AI today in at least limited use. That number is projected to reach 80% within two years, at least in limited use.

Source: Deloitte, State of AI in the Enterprise, January 2026

Manufacturing, logistics, and defense are driving that adoption curve, and the underlying market reflects it. The global AI in warehousing market was valued at approximately $5–$8 billion in 2025 and is projected to reach $80–$100 billion by 2035, as more facilities embed AI into robots, software, and sensing systems.

But here’s the insight that separates operators who get this right from those who chase the wrong things: the future of warehouse automation is less about which robot you buy than about the software ecosystem orchestrating your entire fleet. Hardware is increasingly becoming commoditized. Intelligence, how your systems perceive, decide, and coordinate in real time, is quickly becoming the differentiator.

One company described it this way: they’re granting warehouse robots more autonomy to decide where and how to store items to maximize floor space, not just executing a predetermined put away sequence, but actively optimizing based on current conditions. That is the shift from automation to intelligence. These things are happening or will be happening across all areas of warehouse operations.

Why Adoption Still Lags, and What Most Organizations Get Wrong

Any honest practitioner assessment of physical AI must acknowledge the gap between the headline results and what most organizations actually experience when they try to deploy. That gap is real, it’s significant, and it’s largely self-inflicted. I know, I hate to break it to you.

The Pilot Trap

Most warehouse operations today are running AI pilots. They feel like progress. They rarely turn into production deployments. The reason is a fundamental mismatch between what it takes to run a pilot and what it takes to scale one. This is not ubiquitous to warehousing, it spans all industries, academia and public and private institutions.

A pilot runs with a small team, cleansed data, a controlled environment, and flexible timelines. Production deployment requires infrastructure investment, integration with existing WMS and WCS systems, security reviews, compliance checks, monitoring systems, and ongoing maintenance. Use cases estimated to take three months routinely stretch to 18 months once integration complexity is exposed. No wonder things fall apart in the end!

“If there is no coherent AI strategy, you are likely to see pilot fatigue. You’re chasing the next shiny object, pressured to do something with AI without a real plan.” — AI Leader, Major Logistics Organization

The result is a vicious cycle: organizations fund new pilots, which are relatively low-cost and low-risk, rather than facing the harder work of scaling existing successes. Pilot fatigue sets in. Budgets get redirected. The technology gets blamed for an organizational failure. And we rinse and repeat.

The True Cost of Physical AI

The second thing most organizations underestimate is the total cost of ownership. It’s easy to focus on the AI model, the robot unit price, or the software license. The harder number to face is everything else.

A warehouse automation project might require hundreds of thousands of dollars in AI development, but millions of dollars in physical infrastructure, robotic systems, facility modifications, systems integration, and maintenance infrastructure. Companies that scope the project around the AI cost and discover the full picture mid-implementation are the ones that stall out or abandon the project entirely. Organizations fail to evaluate TCO from the start. The business case needs to hold at full cost, not demo cost.

The Reliability Bar Is Non-Negotiable

In distribution and fulfillment, there is no partial credit. A system that works 70% of the time, impressive in a lab, isn’t going to cut it on the production floor. The bar for physical AI in a live DC environment is 99%-plus reliability, consistent across shifts, SKU mixes, environmental conditions, and the thousand small deviations from the ideal that happen every day in a real operation.

Vendors pitching physical AI solutions should be able to answer directly: what does this perform at in a live, uncontrolled environment, not in your controlled demo? If the answer is vague, there’s your sign.

The Workforce Gap Nobody Is Talking About

Perhaps the most underreported barrier to physical AI adoption isn’t technical; it’s organizational. According to Deloitte’s 2026 State of AI in the Enterprise survey of over 3,200 business and IT leaders globally, 84% of companies have not redesigned jobs or the nature of work itself around AI capabilities. Insufficient worker skills are identified as the single biggest barrier to integrating AI into existing workflows, yet fewer than half of organizations are making significant adjustments to their talent strategies.

84% of companies have not redesigned jobs around AI capabilities.

Source: Deloitte, State of AI in the Enterprise, January 2026

For warehouse operators, this translates directly: you can buy the robots, but if you haven’t redesigned the roles around them, you’re automating a broken process, much faster. The technology doesn’t fix organizational dysfunction. It amplifies it.

The good news: companies that are getting this right are finding that physical AI creates new skilled roles rather than simply eliminating existing ones. Robot technicians, fleet coordinators, predictive maintenance specialists, AI operations managers, these are the jobs that grow alongside intelligent automation. But they require deliberate investment, not just a deployment budget.

What Leaders Should Actually Be Doing Right Now

Given everything above, here is the practitioner’s checklist, opinionated, specific, and grounded in what separates the deployments that work from the ones that don’t.

• Stop evaluating robots. Start evaluating ecosystems. The platform matters more than the robot. Evaluate the orchestration software, data pipeline, and integration layer alongside the hardware. That’s where the long-term value, and the switching costs, actually live.
• Think in degrees of autonomy, not binary switches. Your operation doesn’t flip from manual to fully automated overnight. Map where you are on the autonomy spectrum today, and design your next stage deliberately, not toward a distant ideal that requires ripping out everything that works.
• Pilot in the messy middle. Not your highest-volume lines (already optimized) and not your most chaotic exceptions (not ready). Find the 20–30% of your operation that is variable-but-predictable. That’s where physical AI earns its ROI fastest and builds organizational confidence and creates a runway for scaling.
• Build the real TCO model upfront. Scope the infrastructure, integration, facility modification, and maintenance costs before the business case goes to leadership. The projects that fail at implementation are almost always the ones that got approved on incomplete cost models.
• Invest in the talent stack alongside the tech stack. Machine operators become robot technicians. Logistics coordinators become fleet managers. Maintenance teams shift to predictive models. These transitions require deliberate investment, training programs, new career pathways, and workforce communication that brings people along rather than leaving them anxious.
• Design for deployment from day one. Organizations that treat pilots as stepping stones to production, with clear criteria for scale, defined integration requirements, and executive accountability, see far higher adoption rates and faster, more durable impact than those running disconnected experiments.

Get Physical or Get Left Behind

The AI revolution that has dominated three years of headlines, the one happening in language models, content generators, and digital assistants, is becoming table stakes. The productivity gains are real, but they are not structural. Every competitor has access to the same tools.

The next wave is different. Physical AI is not a software upgrade. It is a fundamental change in what machines can do, how operations are designed, and what competitive advantage looks like in distribution and fulfillment. The organizations that understand this now, that invest in the right ecosystems, build the right talent, and scale beyond the pilot, will operate at a structural cost and throughput advantage that is genuinely difficult to close.

The gap between ambition and activation is real, and it is largely organizational. But the underlying technology is ready, the deployments are live, and the results are documented. The only remaining variable is whether your organization chooses to lead this transition or react to it.

Physical AI isn’t coming. It’s already on your dock. The question is whether you’re deploying it or scrambling to catch up when your competitor does.

SOURCES & REFERENCES

Deloitte — State of AI in the Enterprise, January 2026 | deloitte.com

Deloitte — Tech Trends 2026: Physical AI & Humanoid Robots | deloitte.com

World Economic Forum — Physical AI: Powering the New Age of Industrial Operations (2025) | weforum.org

SupplyChainBrain — How Physical AI Will Reshape the Warehouse | supplychainbrain.com

Manufacturing Dive — The Physical AI Craze and Other Automation Trends to Watch in 2026 | manufacturingdive.com

Manufacturing Dive — Physical AI Is the Convergence of Robotics and Intelligence | manufacturingdive.com

International Federation of Robotics — Top 5 Global Robotics Trends 2026 | ifr.org

Citi Research — Embodied Intelligence: The Rise of Physical AI | citigroup.com

Automate / NVIDIA — What Physical AI Means for Robotics (2025) | automateshow.com

AI News — Physical AI Is Having Its Moment (March 2026) | artificialintelligence-news.com