Fei-Fei Li Says AI’s Next Frontier Is Spatial Intelligence

How World Labs’ Marble fits the vision

Marble is a multimodal world model that can generate spatially consistent, high-fidelity, persistent 3D environments from inputs including text, images, and video. The resulting worlds can be explored and edited rather than existing only as a fleeting image or a text response.

That persistence is central to the product concept. A stable digital environment can be used as a setting for storytelling and creative work, or as a simulated space in which systems can learn, test decisions, and evaluate behavior. World Labs has specifically highlighted applications spanning creativity, simulation, robotics, and scientific discovery.

Why robotics is the clearest early use case

Robots need more than verbal instructions. They must interpret space, understand objects and surfaces, anticipate movement, and connect perception to action. A navigable 3D world could give developers a way to simulate environments and test robotic behavior before deploying machines in the physical world.

This does not mean Marble has solved general-purpose robotics. The stronger, source-supported conclusion is narrower: spatial world models are being developed as infrastructure that could help robotics training, simulation, and evaluation.

Healthcare imaging and manufacturing are often discussed as plausible areas for spatial AI because they involve 3D structures and physical processes. However, the available reporting on the August 19 The Circuit segment does not provide enough evidence to attribute detailed promises about those sectors directly to Li in that interview. The practical opportunity is real enough to discuss as a possibility, but it should not be presented as a confirmed interview commitment.

World Labs’ funding and AI 50 recognition

World Labs has raised more than $1 billion, according to Forbes’ 2026 AI 50 coverage. Forbes also identified it as one of four female-led companies on that list and described its focus as spatial intelligence.

That level of funding signals investor interest in world models as a potential layer of AI infrastructure beyond today’s language-model race. It does not, by itself, establish that the technology will become a dominant platform or that its proposed applications are commercially proven.

Li’s warning about how AI is discussed

In the Bloomberg interview, Li said technologists need to communicate AI’s concrete benefits more effectively. She warned that growing public opposition in the United States could have consequences beyond the country because of the US role in the global technology ecosystem.

Her position is a call for more disciplined communication—not simply more optimistic messaging. The useful question is what a system can demonstrably do, what risks follow from those capabilities, and who gets access to the benefits.

The broader policy message: access, investment, and responsibility

That approach is consistent with Li’s longer-standing human-centered view of AI. She has argued for evidence-based policymaking, public-sector leadership, global collaboration, and responsible development.

She has also emphasized the importance of broad access to computing resources, data, research infrastructure, and education. Stanford’s Institute for Human-Centered Artificial Intelligence has described equitable access to powerful computational resources and large-scale datasets as important for universities and nonprofits, while Li’s work with AI4ALL has focused on inclusion and diversity in AI education.

The underlying concern is that AI’s gains—and the ability to shape the technology—could become concentrated among a small number of companies or communities. Public investment and inclusive education are therefore not side issues in her framework; they are part of how AI’s benefits can be distributed more widely.

The takeaway

Li’s vision extends AI from systems that mainly talk about the world to systems that can model and act within it. World Labs and Marble are an early expression of that bet: persistent, editable 3D environments that may support creative tools, simulation, and robotics.

The broader message is more cautious than the headline “beyond chatbots” might suggest. Spatial intelligence is a proposed next frontier, not proof that language models have reached their endpoint. And Li’s policy argument is that progress should be judged through measurable capabilities, real-world benefits, accountable development, and equitable access—not through either utopian promises or apocalyptic predictions.