AI Investment and Adoption Are Outrunning Institutions’ Capacity to Respond

AI investment and adoption are accelerating at a pace that strains the systems meant to evaluate and absorb them. James Landay, director of Stanford’s Institute for Human-Centered Artificial Intelligence (HAI), said global corporate investment in AI exceeded all of 2025’s total by 46% in the first half of 2026. He said the year’s total could therefore rise above $2 trillion. A midyear update to Stanford HAI’s AI Index tracks that pace more frequently than the institute’s annual report because, Landay said, AI is moving too fast to document only once a year.

Adoption is spreading just as quickly. Landay cited AI use in 88% of surveyed organizations and said population-level adoption of generative AI had outpaced the PC and the internet. Schools, governments, and evaluation systems were not built to move at that speed; feeling overwhelmed, he argued, is understandable.

But the pace is not the only source of unease. AI development is concentrated: the United States and China accounted for more than 90% of notable AI models released so far in 2026, according to the figures Landay presented. The systems people use therefore reflect choices made in a small number of countries.

Public enthusiasm also varies sharply by country. The survey data Landay showed put excitement about AI products and services at 83.1% in China, compared with 32.7% in the United States; the global average was 51%. Landay said researchers need to understand why people in the U.S. feel less certain about AI’s place in their future, without dismissing their concerns.

The concentration of model development is one part of a broader question Landay raised: who gets to shape AI, and whose needs do its systems recognize? He asked what control people have when a system does not understand their language or the way their society sees the world.

HAI faculty and its policy team study language gaps. Landay’s own lab examines assumptions models make about what a human is and what matters—assumptions that may not hold across cultures. These concerns, he said, help explain calls for AI sovereignty: countries want real choices about the systems on which they depend. One possible route is genuinely open-source AI that countries can adapt and examine for their own needs.

The issue, as Landay framed it, is not just whether a model works, but who can inspect it, whose language it serves, and whose assumptions it carries.

For students and parents, Landay said, the immediate fear is whether the next generation will still find a way into a career. He pointed to research by Erik Brynjolfsson and colleagues at HAI’s Digital Economy Lab, which finds an employment gap for young workers in AI-exposed occupations such as software engineering. The researchers attribute it to employers hiring fewer young workers in those fields.

Landay stressed the limit of that finding: it is not yet evidence of economy-wide job loss. He described it instead as an early warning. HAI faculty affiliate Diyi Yang studies how people can build skills for working with AI; Landay presented that work alongside employment monitoring as part of a response to changing jobs.

Landay pointed to incidents involving OpenAI and Hugging Face, and Anthropic models bypassing government website controls, as reasons to take current capabilities seriously. His emphasis was on harms people can cause with powerful AI today. Addressing those harms, he said, requires technology, policy, and science—not distraction by science-fiction speculation about superintelligence.

That response depends on independent evaluation. HAI is launching a center for AI measurement science, AIMS, led by Sanmi Koyejo. One of its questions is whether benchmark scores predict real-world outcomes. HAI senior fellow Mykel Kochenderfer develops methods to search for rare failures in AI systems before people depend on them. Together, those efforts focus on what tests can tell us about deployment, and whether they can reveal problems that ordinary measures miss.

Landay argued that confronting AI’s risks matters in part because the technology also has substantial potential. He pointed to three Stanford projects built through collaboration between AI researchers and people with domain expertise.

In bioengineering, Brian Hie’s Evo 2 learns patterns in DNA to help predict genetic changes and design new sequences. In government, Dan Ho and collaborators in the Law School scanned 500 million words of state law to identify reporting requirements that may no longer serve a purpose. In education, Dora Demszky combines computer science and learning science to improve tutoring systems, including how they respond to questions that reveal a learner’s reasoning in the moment.

For Landay, these examples show why AI’s consequences cannot be understood through technical expertise alone. The relevant context might be biology, public administration, or how a student learns. HAI began in 2019 on the premise that AI would become pervasive and that computer scientists could not build it alone in a way that benefits society. In May 2026, Stanford brought HAI and Stanford Data Science together under the HAI name, joining communities that share an emphasis on research across disciplines.

Landay assigned universities a particular responsibility: pursue open, curiosity-driven inquiry and bring independent, long-term thinking to AI. That means testing claims—whether optimistic or alarming—and letting evidence guide conclusions. It also means bringing relevant disciplines in from the start: understanding effects on a learner, patient, or community requires people who know those settings.

Openness is part of that responsibility. Universities should share methods, data, and tools so others can examine the work, challenge it, and build on it. Landay described Stanford HAI as positioned to connect technical depth with the breadth needed to understand AI’s effects on people.

HAI’s next chapter, he said, will concentrate on three aims: helping researchers across disciplines use AI and data science to pursue questions that were previously beyond reach; rethinking education for a world where AI is part of nearly every profession; and translating research into evidence for policymakers while working with industry and public institutions on how AI is developed and adopted.