The latest World Humanoid Robot Games, held in Beijing from August 22–26 last year, brought together 2,056 robots from 666 teams. A year earlier, those figures stood at 500 robots and 280 teams, respectively. The event offered further evidence of how quickly the industry is advancing, along with an opportunity to see robots in action running, dancing, jumping, boxing and competing in impressive challenges.
But companies in industries such as manufacturing, logistics, energy and services may not be particularly interested in whether a robot can win a race. They’re more likely to ask a different question: Can it operate reliably throughout an entire shift?
This is where the difference between a demonstration and real innovation becomes clear. A demonstration needs only three minutes to impress. Real innovation, however, demands eight hours of consistent performance, safety, maintenance, systems integration, workforce acceptance and economic returns.
A Gartner forecast published in January 2026 provides a revealing picture of where the market stands. By 2028, fewer than 100 companies will move humanoid robot proofs of concept beyond experimentation, and fewer than 20 will reach production in supply chain and manufacturing use cases.
The market has no shortage of ideas. The challenge is turning them into production-ready solutions that can scale, address real business problems and deliver tangible impact.
Physical AI and tangible innovation
In many organizations, pilot projects begin with an exciting technology, followed by an effort to find a suitable application. The result is often an interesting showcase that’s difficult to scale.
Tangible innovation takes the opposite approach. It starts with the process itself operational friction, costs, safety, talent shortages, operational variability or areas where small improvements can generate significant cumulative benefits.
Physical AI integrates artificial intelligence into machines, processes and physical environments, enabling them to sense, make decisions and act in real time.
At NTT DATA, we see Physical AI as the combination of real-time intelligence, edge AI and enterprise automation applied to factories, warehouses, cities and critical infrastructure, with benefits including reduced latency, improved safety and greater operational scale.
Making this a reality requires moving beyond the fascination with demonstrations. A robot is one component of a broader physical operating system. As such, it depends on reliable data, stable connectivity, computer vision, edge computing, integration with enterprise resource planning (ERP), manufacturing execution systems (MES) or warehouse management systems (WMS), as well as safety protocols, predictive maintenance, risk governance and teams prepared to work alongside the technology.
Bringing robotics into production therefore means creating the conditions around it that allow the technology to operate reliably over time.
This is not a challenge that organizations can easily address in isolation. It requires combining research and development (R&D) with collaboration across startups, academic institutions, accelerators and technology partners.
An ecosystem-driven approach therefore plays a key role in moving emerging technologies from initial ideas to experimentation and, from there, to solutions that deliver real-world impact.
Designing pilots with production in mind
This shift toward ecosystem collaboration and tangible, outcome-driven solutions also changes how organizations should design pilot projects.
If the ultimate goal is production, the minimum requirements for scaling should be established from the outset. These include expected uptime, cost per task, the level of human supervision required, safety risks, integration with actual workflows, maintenance plans, workforce acceptance and business metrics.
If these factors aren’t addressed during the pilot phase, they’re likely to emerge later as barriers to scaling.
Another important consideration is recognizing that not every solution needs to be humanoid.
Gartner points out that, in many supply chain applications, polyfunctional robots with less human-like designs may offer better performance, adaptability and return on investment.
A robot’s design should be driven by the operational challenge it needs to solve, not by the appeal of resembling a human.
In my view, real innovation happens when a robot whether visually impressive or not makes an industrial facility safer, reduces downtime, improves traceability, eases physically demanding tasks or enables operations in environments where people shouldn’t be exposed to risk.
Success, therefore, isn’t about demonstrating that a robot can do something extraordinary. It’s about proving that it can perform that task reliably within a real operational environment.
The pilot that truly matters is the one that stands up to real-world conditions. It’s the one that can operate throughout an entire shift, integrate into existing processes without disrupting them and free up people’s time and capacity for higher-value activities.
It’s also the pilot that provides enough evidence for the executive committee making the investment decision to invest not in a promise, but in a proven capability.
In Beijing, we were amazed to watch robots demonstrate their abilities on an athletics track. But the true future of robotics is taking shape elsewhere in warehouses, factories, hospitals, ports, airports, mines and critical infrastructure networks.