Stanford AI100 Report: Scaling Era Hits Data and Cost Limits

Stanford HAI's third One Hundred Year Study on Artificial Intelligence, announced November 10, 2026, concludes that the scaling strategy behind generative AI is running into data scarcity and prohibitive frontier-training costs just as AI's effects spread through science, security and public institutions.

The report, titled AI Goes Mainstream: The Scaling Era and the Rise of Generative AI, follows editions from 2016 and 2021. AI100 launched in 2014 and convenes a study panel every five years. University of Oxford professor Michael Wooldridge chaired the multidisciplinary panel, which answered 12 standing questions and added two deep dives — one on AI geopolitics, one on scaling. The panel finalized the text in September 2026.

What ends the scaling era?

The panel credits increased training data and computation for generative AI's chief advances over the past five years. But it finds two hard constraints: scarce high-quality data and the prohibitive cost of frontier training runs.

Transformer and diffusion architectures became consequential only when paired with massive datasets and compute. ChatGPT's public release in November 2022 then carried generative tools to millions of users.

The next phase, the panel expects, will lean on smarter algorithms rather than raw size:

• Potentially smaller models, rebuilt less frequently
• Better context and memory
• Specialized reasoning training
• More inference-time computation
• Tools and agents for complex tasks

Reasoning models generate intermediate steps and burn additional computation while producing an answer, unlike a conventional model's single forward pass. Capabilities remain uneven: sophisticated performance sits alongside fragile reasoning and failures on simple tasks. The gains are real, but so are the gaps.

Where is the money actually showing up?

The panel says AI is compressing some scientific research timelines from years to months or weeks, and forecasts that AI-powered laboratories will become normal across disciplines within a decade. It calls for principled validation of machine-generated discoveries alongside continued human verification of AI-produced research. How much say humans should keep in experimental design remains open.

The investment numbers are striking. The panel says the magnitude and speed of spending on AI data centers and chips surpassed earlier investment booms, straining energy and freshwater supplies, raising carbon emissions and shaping national energy policy. At the report's time of writing, Nvidia's valuation exceeded $5 trillion, and more than 15% of U.S. stock-market gains in 2025 were attributed to Nvidia's stock alone.

Yet beyond valuations and capital expenditure, the report concludes AI's economic impact was not yet evident in aggregate macroeconomic data or companies' returns on investment. That distinction — visible financial activity versus demonstrated economy-wide and company-level returns — is the central qualification the panel places on the boom. The market has moved; the measured economy largely hasn't.

How serious are the security risks?

Cybersecurity works both ways, the panel finds. It cites demonstrated vulnerability discovery and cyberattacks conducted by teams of AI agents during the reporting period. Lethal targeting decisions have been made without humans in the loop during active warfare. Synthetic images, cloned voices and video weaken the evidentiary assumptions underlying journalism, law and democratic institutions.

Governance hasn't kept pace. The panel describes international governance as fragmented and increasingly framed through national security, with export controls and restrictions on chips and model access becoming instruments of statecraft. International summits and declarations have advanced principles and scientific or diplomatic discussion, but produced few practical, legally binding multilateral measures.

Although AI affects billions of people, the panel finds the public has had little meaningful input into its development and deployment. Concerns span jobs, education, the environment, mental health, democratic institutions and critical infrastructure. The recommended response: treat AI's direction as a collective societal choice, backed by meaningful public input and action from governments and technology providers.

The panel declines to predict whether rapid capability growth will continue or flatten against limits in investment, semiconductors and data. Under either path, it concludes, AI's scientific and financial footprint is already visible — while reliable validation, broad economic returns and meaningful public accountability remain unresolved. Five years from now, the next AI100 panel will judge how many of those gaps closed.