Artificial intelligence has mastered text and images, but without a deep understanding of three-dimensional space, it cannot move into the physical world. Spatial intelligence is emerging as the next competitive frontier, opening possibilities for robotics, manufacturing, and immersive commerce while challenging executives to rethink data strategies and infrastructure.
Artificial intelligence has moved in impressive but incomplete waves. It began with language, when machines learned to read, reason, and converse with striking fluency. The next breakthrough was visual, as systems started to create images and video so lifelike that viewers often could not tell the difference. Yet these triumphs remain essentially two-dimensional. Models may detect edges and textures, but they still lack a proper grasp of geometry, physics, and the way objects occupy and interact in real space. Without that depth of understanding, AI reaches the edge of the physical world and stops.
Human life is shaped by spatial awareness. We navigate crowded streets, design machinery, build homes and factories, and make decisions with an instinctive sense of distance, volume, and proportion. “AI today is trained mainly on two-dimensional content because that is what is most available online,” Alex de Vigan explains. “Images and videos do not contain the consistent spatial information that models need to understand geometry or physics. Without that, you cannot guarantee size, proportion, or the interaction of objects in a generated scene.” The lack of spatial intelligence means even the most advanced models struggle to train warehouse robots or plan a production line with complete reliability.
Data is the critical differentiator
The obstacle is not clever algorithms but the absence of rich data. Large language models thrive because text is abundant and relatively easy to label. Visual models draw on billions of photographs and video clips. High-quality three-dimensional information, carefully curated and consistently annotated, is far harder to find. “The amount of data matters, but quality matters more,” de Vigan says. “Models need data that is realistic, specific to the use case, and perfectly labelled. Poor or biased data ruins performance. Human involvement in curation is critical, and in 3D it may be even more important than for text or images because of the complexity.”
Synthetic data is emerging as a practical answer. Developers can generate photorealistic 3D environments and objects to create the immense training sets that industrial AI demands. Defence and robotics companies already use synthetic datasets to simulate scenarios that would be impossible to capture in the real world. “You cannot film tanks moving in every possible environment or replicate every factory layout,” de Vigan says. “By digitising these objects and spaces, you can create millions of variations for training while controlling fidelity and conditions.”
Producing this data is demanding work. High-resolution 3D assets can be created in small numbers, but scaling them without losing accuracy requires a sophisticated pipeline and constant refinement. De Vigan describes the balance between fidelity, scale and computational efficiency as a moving target rather than a single achievement. The goal is to push all three forward together so that quality, speed and cost improve in tandem. For senior executives, this means investing in new infrastructure and skills while setting realistic expectations about the time, budget and energy such projects require.
Forward-looking organisations are starting to build hybrid strategies that mix captured scans of real factories, warehouses or city districts with procedurally generated objects and scenes. Combining the two creates diverse training sets and helps avoid model collapse, where AI trained on its own synthetic output slowly degrades. These programmes bring together simulation engineers, data scientists and industry specialists in new interdisciplinary teams that bridge the gap between digital modelling and physical engineering.
Industrial robotics takes the lead
Industrial robotics is the most advanced commercial field for spatial AI. Robots must understand their environment with exact precision to navigate, grasp and manipulate objects safely. “Robotics is the biggest near-term opportunity for spatial AI,” de Vigan observes. “Industrial robots must understand their environment with exact accuracy to improve efficiency and performance. Training them requires realistic 3D data and endless permutations of movement, lighting and obstacles.”
Factories and warehouses are already testing digital twins of their production lines. High-fidelity models allow engineers to plan thousands of robotic trajectories or optimise grasping strategies long before a machine moves on the floor. These simulations cut downtime and accelerate deployment, lowering costs and improving safety. Similar methods are being applied to automated ports, smart-city infrastructure and autonomous vehicles, where spatially aware AI can improve routing, maintenance and energy use.
Healthcare and construction illustrate the wider potential. Surgical robots need precise models of human anatomy to plan complex procedures. Architects and builders can use spatial simulations to test structures against extreme weather or seismic stress. In retail and entertainment, immersive 3D content is becoming a powerful way to engage customers and shorten design cycles. Energy and logistics companies are exploring spatial AI to optimise offshore installations, plan maintenance schedules and reduce risk across extended supply networks. All these examples share a dependence on accurate 3D data and the ability to run countless permutations before committing to costly physical changes.
Beyond the factory floor
Spatial AI is also beginning to influence areas not immediately associated with robotics or construction. Supply chain design, predictive maintenance, urban planning and environmental modelling all benefit from a detailed understanding of how objects and systems interact in space. Logistics networks can be modelled to forecast bottlenecks under different demand scenarios. City planners can simulate traffic patterns or flooding events to test resilience strategies before breaking ground. Even agriculture is experimenting with spatial data to optimise planting layouts and track crop health in real time.
Ethical and regulatory questions follow close behind. Ownership of training data, intellectual property rights and the risk of feedback loops need careful oversight. “You cannot simply feed AI-generated data back into the system without safeguards,” de Vigan warns. “It risks creating feedback loops that degrade model performance. Human oversight remains essential to maintain realism and trust. Organisations must also consider energy demands. High-fidelity modelling consumes significant compute power, and sustainability goals will influence how companies scale these projects.”
From edge to expectation
For now, spatial intelligence offers a decisive competitive edge. The ability to integrate accurate three-dimensional understanding into AI systems differentiates robotics platforms, design tools and immersive commerce applications. Over time, de Vigan expects it to become standard. “Eventually, spatial understanding will be table stakes,” he predicts. “Everything we do, from how we work to how we create, has a spatial component. Even two-dimensional information ultimately reflects three-dimensional reality.”
That trajectory mirrors the rise of language models. Once revolutionary, they are now commoditised. Strategic advantage will shift toward proprietary datasets and domain-specific applications. “The difference will be in the specific application, and it will depend on the data,” de Vigan says. “Enterprises that assemble high-quality vertical data and build the infrastructure to manage it will lead as foundational models converge in capability.”
The implications for strategy are clear. Investment must reach beyond algorithms to the entire stack of spatial data: capture, generation, curation and governance. Compute demands will climb, energy planning will matter, and partnerships with specialised data providers will become central. Boards should treat spatial intelligence as a core capability, not a side project. Early movers that secure reliable data sources and develop expertise in spatial modelling will shape market standards and attract the best talent.
Preparing for a spatial future
Spatial AI is no longer a distant prospect. Autonomous vehicles already navigate dense urban traffic. Industrial robots increasingly share workspaces with human colleagues. Marketing and entertainment companies experiment with immersive 3D content that blurs virtual and physical boundaries. As these systems mature, the ability to model reality at scale will decide which organisations lead and which follow.
Enterprises that recognise this shift are integrating spatial data teams into their AI roadmaps and rethinking cloud and edge infrastructure to handle heavier computational loads. They are forging alliances with universities, simulation specialists and niche 3D data providers to secure diverse sources of training material. Sustainability goals are becoming part of the planning as energy demands rise. These steps may appear incremental, but together they build the foundation for lasting advantage.
AI began with words, learned to see and now must learn to move. Spatial intelligence will enable machines to share our environment and collaborate on tasks that require awareness of depth and physical constraints. For business leaders, understanding and investing in this capability is no longer optional. It is the next logical step in the evolution of artificial intelligence and the key to unlocking its full commercial and industrial potential.
