The third edition of the century-long study documents generative AI’s leap into the mainstream while flagging data, compute, and governance constraints
Stanford’s 100-year study of artificial intelligence has published its third report. Its message is sobering: AI’s explosive progress runs on fuel that may not last forever.
The report is titled AI Goes Mainstream: The Scaling Era and the Rise of Generative AI. It argues that bigger computers and more data powered the field’s recent breakthroughs. It also warns that both inputs are becoming harder and costlier to get.
The authors also concede the document may already be outdated. They used AI to help write and review it.
What the AI100 report actually says
The One Hundred Year Study on Artificial Intelligence, known as AI100, launched at Stanford in 2014. The plan is to publish an assessment every five years for a century. The two earlier editions came out in 2016 and 2021.
This latest edition covers 2021 to 2026. During that stretch, generative AI moved from a niche tool to something billions of people use. The report examines how that shift has reshaped work, science, public life, and governance.
Professor Sheila McIlraith chairs the AI100 Standing Committee. She framed the period in sweeping terms.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
We respect your privacy. Unsubscribe anytime.
“The last five years constitute a watershed moment in the history of AI.”
The central finding is about how progress happened. According to the report, recent advances came mainly from scaling up computation and the amount of training data. The report says future gains will run into limits on available data and the rising cost of compute.
Concentration, control, and who gets a say
The report also looks at who builds frontier AI. It finds development concentrated in a small number of entities in the US and China. The authors see that concentration as a governance problem. They call for clearer governance structures and more public involvement in deciding how AI gets deployed. They also push for broader education about how these systems work.
The panel brings together experts from fields including computer science and philosophy.
A report about AI, partly written with AI
One of the more interesting details concerns how the report was made. The authors say AI tools handled jobs like transcription, summarization, and early drafts. They stress that people did the core analysis and writing.
The authors call for “clear-eyed scrutiny” of AI’s benefits and risks.
Why the five-year rhythm matters
AI100 was designed as a slow, steady record of a fast-moving field. The goal is a consistent snapshot every half decade that future readers can compare across generations. When a report warns that it may be outdated on release, the mismatch between cadence and pace is clear. The panel appears to know this and says so openly.
What this means for the AI industry and its regulators
The most important takeaway may be the warning about scaling limits. The report suggests that assumption faces growing pressure. The report’s call for public involvement reads partly as a response to the risk that only a few US and Chinese entities can afford frontier work. For regulators, the report adds weight to arguments for clearer rules and more transparency. It does not lay out a single legislative blueprint. It does make the case that leaving deployment choices to a few builders is a problem worth solving.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.
Next in TechnologyBitdeer posts record $320.7 million Q3 revenue as AI lease pipeline grows
