Stanford’s AI Index 2025 report reveals faster, cheaper, and more accessible AI, driven by open models and mass enterprise adoption.
The “AI Index Report 2025,” published by Stanford University’s Institute for Human-Centered Artificial Intelligence (HAI), is a definitive reference on the current state of AI. Conducted by an international team of experts led by Yolanda Gil and Raymond Perrault, this annual study compiles and analyzes global AI trends across various fields: economics, technological performance, research, and accessibility. Here are four particularly striking takeaways for digital professionals in 2025.
Record corporate investments in AI, a major lever for productivity
In 2024, US private investment in artificial intelligence reached a historic high: $109.1 billion, nearly 12 times more than China ($9.3 billion) and 24 times that of the United Kingdom ($4.5 billion). A strong trend is emerging around generative AI, which alone attracted “$33.9 billion in private investment globally, an increase of 18.7% compared to 2023,” the report states.
At the same time, AI is now adopted by 78% of organizations worldwide, compared to only 55% in 2023. This acceleration is explained by the tangible benefits observed: a significant improvement in productivity and an effective reduction in skills gaps within teams. Clearly, companies that do not intelligently take the AI turn today risk quickly finding themselves left behind.
The United States maintains its lead in AI models, but China is catching up
While the United States still largely dominates the market in terms of the number of “notable” AI models produced (40 models in 2024, compared to 15 for China and only 3 in Europe), their qualitative lead is melting like snow in the sun. The report highlights that “performance differences on major benchmarks such as MMLU and HumanEval have gone from double-digit gaps in 2023 to near parity in 2024.” China is also confirming its status as a leader in AI patents and scientific publications, marking a clear strategy of technological conquest.
Global dynamics are also evolving: the development of artificial intelligence models is becoming increasingly international, with major innovations coming from the Middle East, Latin America, and Southeast Asia. This globalized context should strongly challenge Europe, which is still struggling to establish its own technological champions in the face of the two American and Chinese giants.
AI becomes more accessible thanks to lower costs and greater efficiency
Artificial intelligence is becoming more affordable every year, significantly expanding its scope of application. The cost of running a high-performance model equivalent to GPT-3.5 fell “more than 280 times between November 2022 and October 2024 ,” according to the study. This dramatic drop in inference costs is compounded by a 30% annual reduction in hardware costs and a 40% improvement in infrastructure energy efficiency. As a result, economic barriers are falling, allowing even SMEs to access solutions once reserved for tech giants.
Furthermore, the study observes that open-weight models have caught up with closed models in just one year, reducing the performance gap “from 8% to only 1.7% on certain benchmarks .” This development promotes an unprecedented democratization of advanced technologies, making powerful tools that were previously exclusive to the general public accessible.
Industry dominates AI research, but the technology frontier is becoming highly competitive
Industry is now taking almost complete control of producing the most advanced AI models. By 2024, nearly 90% of notable models will come from the private sector, compared to only 60% in 2023. However, the academic sector remains crucial for producing “highly cited scientific research ,” the report notes, and thus contributes to the fundamental innovation essential for future technological advances.
Paradoxically, the intense technological race observed in recent years is not widening the gaps: on the contrary, it is reducing them. “The Elo score gap between the best model and the tenth best-ranked model has fallen from 11.9% to 5.4% in just one year,” the study indicates (see cover image). And today, the two best AI models are separated by only a minimal performance gap (0.7%). This increasingly tight competition could represent a strategic opportunity for European players, particularly in France, to target very specific niches with high added value.
