Stanford HAI AI Index 2025: Global AI Trends Amid Technological Acceleration and Unfinished Governance

In April 2026, Stanford AI Index Report, the most authoritative global AI monitoring tool, was published. The 2026 report is structured into nine thematic chapters—Research and Development, Technical Performance, Responsible AI, Economy, Science, Medicine, Education, Policy and Governance, and Public Opinion—and introduces, for the first time, a chapter dedicated to the impact of AI on science.

The data covers the period up to March 2026 and offers an updated picture at a time when the gap between technical progress and governance capacity is widening at an alarming rate.

Summary data

Indicator	Value
Private AI Investment in the US in 2025	$285.9 billion (23× China)
Organizations using AI in 2025	88% (it was 78% in 2024)
Documented AI-related incidents in 2025	362 (there were 233 in 2024)

Trend	Description
Trend 1	AI capability isn't plateauing: it's accelerating - Over 90% of frontier models in 2025. - Outperforms humans on doctoral-level scientific questions. - Performance on SWE-bench Verified from 60% to almost 100% in one year. - Four out of five college students use generative AI.
Trend 2	The US-China modeling gap has effectively closed - Alternation at the top between the US and China from 2025. - DeepSeek-R1 reached the US flagship model in February 2025. - As of March 2026, Anthropic model with a 2.7 point lead. - US produces more high-impact patents; China leads in publication volume and total patents.
Trend 3	AI Infrastructure: Concentration and Geopolitical Fragility - USA with 5,427 data centers, over 10 times more than other countries. - TSMC makes nearly all of the flagship AI chips. - TSMC expansion in the US started in 2025.
Trend 4	The "jagged frontier" of AI - Gemini Deep Think wins gold medal at the International Mathematical Olympiad. - Top model reads an analog clock correctly 50.1% of the time. - AI agents from 12% to about 66% on OSWorld tasks.
Trend 5	Responsible AI is not keeping pace with capabilities - Results on published capability benchmarks, but gaps on responsible AI. - Documented accidents rose to 362 in 2025 (+55% on 2024).
Trend 6	USA leads in investment, but loses talent - US private investment $285.9 billion in 2025. - 1,953 new AI companies funded (10× the next country). - Researcher and developer transfers have dropped by 89% since 2017.
Trend 7	Adoption at historic speed, growing consumer value - Global adoption of generative AI reaches 53% in three years. - Estimated value of generative AI tools for US consumers: $172 billion annually. - USA ranks 24th for adoption (28.3%).
Trend 8	Deep gap between experts and public opinion - 73% of US experts view the impact of AI on jobs positively; only 23% of the public do. - US has the lowest level of trust in government to regulate AI (31%).

Job trends: a fact that deserves specific attention

The report clearly documents for the first time the overlap between AI productivity gains and declining youth employment in the most exposed sectors. The cited studies show productivity increases of 14–26% in customer support and software development, with weaker or negative effects in more judgment-intensive tasks. Meanwhile, in the US, software developers aged 22 to 25 are projected to experience a nearly 20% decline in employment by 2024, while growth continues for older age groups.

This finding has direct relevance for public policy: it suggests that AI productivity gains could be concentrated in the early stages of career paths, with implications for university education and labor market entry mechanisms.

Analytical observations for research on regulation and sustainability

From the perspective of regulatory quality, the 2026 report introduces a new element compared to previous editions: the growing tension between speed of adoption and institutional capacity to respond. The fact that 88% of organizations already use AI, while half of US middle and high schools still have no AI policies, vividly illustrates this gap.

In terms of sustainability—a central theme for SUSTAIN-AI-REG—the report offers worrying data on infrastructure: the US consumes more energy than any other country for AI data centers, and reliance on a single foundry (TSMC) for chip manufacturing introduces significant geopolitical fragilities. TSMC's expansion in the US, which began in 2025, represents a first step toward diversification, but the question of absolute energy consumption remains open.

Finally, the “jagged frontier” described in the report—models that win Olympic medals in mathematics but fail on elementary tasks—raises fundamental questions about the reliability of high-risk AI systems in the public sector.

The AI Index 2026 report documents a turning point: AI is now a mainstream technology, but public governance—both nationally and internationally—is struggling to keep pace. The findings on employment, institutional trust, and energy infrastructure offer precise empirical grounding for the European and Italian regulatory debate.

Source: Stanford HAI, AI Index Report 2026 .