# Fei-Fei Li goes from 15 million images to 40 people _Fei-Fei Li told Hoover Institution that 15M-image ImageNet birthed modern AI, now 40-person World Labs is chasing spatial intelligence as the next frontier._ **Published:** 2026-10-27 **Source:** https://www.startuphub.ai/ai-news/artificial-intelligence/2026/fei-fei-li-goes-from-15-million-images-to-40-people --- Fei-Fei Li put a number on the next frontier when she joined the [Hoover Institution](https://www.youtube.com/watch?v=hHlpUt7Cqg8) for a conversation with historian Stephen Kotkin: 15 million images built ImageNet and the modern AI era, now 40 people at World Labs are trying to build its spatial successor. That comparison is the point. Li, Stanford professor and co-founder of [World Labs](https://www.startuphub.ai/ai-news/investors-news/2026/fei-fei-li-clarifies-world-models), said the company she started in 2024 with three other co-founders is a frontier model company for spatial intelligence, a term she described as the next chapter after language [models](https://www.startuphub.ai/ai-news/investors-news/2026/fei-fei-li-clarifies-world-models) like ChatGPT that took hold around 2022. The lab has 40 employees as of the conversation, Li said, and aims to empower creators, robotics, healthcare, education and enterprise users. It is a deliberately small team against the scale of ImageNet, the 15-million-image dataset Li built that helped birth modern computer vision. Kotkin pressed Li on curiosity and audacity, the themes of the Hoover discussion, and she tied both to biography. Li said she immigrated to the United States at 15, that her parents arrived with $20, and that San Francisco's Chinatown dates from 1850, part of a longer story she cited of about 48 million foreign-born people now living in the U.S. Her parents gave unconditional support rather than academic pressure, she said, and she first loved physics through Albert Einstein before concluding that intelligence was the audacious question worth pursuing. Intelligence, for Li, is multi-dimensional. She said it involves sensing, perception, planning, talking, comprehension and reasoning in complex 3D worlds, plus the ability to imagine and predict spatially and semantically. Language, she noted, is a few hundred thousand years old as an evolutionary phenomenon, while animal sensing and perception reflect a half-billion-year journey. That framing matters for World Labs, because spatial intelligence is not just generative 3D, it is prediction and interaction in physical space. Li co-founded Stanford's Institute for Human-Centered AI to address how AI systems affect people and to steer them toward positive use while limiting harm, she said. The institute's work extends beyond her lab. Its AI Index is an independent initiative at Stanford HAI led by an interdisciplinary steering committee of experts from academia and industry, according to Stanford HAI. The HAI program more broadly covers technical fundamentals, business implications and societal considerations. The numbers leave gaps. Li did not disclose funding, compute, model architecture or evaluation for World Labs, and the 40-person figure is a point-in-time headcount with no breakdown of research versus product roles. What the data does not show is whether a small frontier lab can sustain training and data costs that larger labs absorb. One disclosure would change the read: benchmark results or a reproducible demo showing spatial prediction, not just generation. Li also offered two judgments that frame the policy debate around scale. She said curiosity needs freedom to flourish, and that science is the pursuit of truth using rigorous methodology, debate and review. She said improved productivity absolutely does not guarantee shared prosperity, and that the most important moral responsibility as a scientist is to pursue truth, a line she delivered verbatim. The open question the numbers leave is whether 15 million labeled images was the hard part, or whether training a spatial foundation model with a 40-person team will require an even larger and less curated data engine. --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory. © StartupHub.ai. All rights reserved. You may not republish this article in full without a license. Search engines and AI research tools may crawl and summarize for reference. Bulk reproduction or model training on this content requires a license. See https://www.startuphub.ai/terms.