The Stanford Institute for Human‑Centered AI has released its ninth annual report on the state of artificial intelligence — and the picture it paints is both impressive and unsettling. This isn’t just a dry collection of figures; it’s a real snapshot of an era when AI has stopped being a “technology of the future” and has become everyday infrastructure, complete with its benefits, risks, and paradoxes.

The Growth Rate Isn’t Slowing Down — and That’s News in Itself

The report’s authors emphasize that AI isn’t plateauing; its capabilities continue to grow at a rapid pace. More than 90 % of cutting‑edge models were created in 2025 alone, and some of them are already solving scientific problems at a level comparable to humans — and in some cases, surpassing it. This isn’t about “writing better emails”; it’s about real research tasks where AI is becoming a full‑fledged assistant to scientists.

At the same time, AI development remains strikingly contradictory. For example, the Gemini Deep Think model won a gold medal at the International Mathematical Olympiad, while the best current models correctly tell the time from an analog clock only about half the time. This reminds us that AI’s “intelligence” isn’t universal — it’s highly specialized: it excels at certain tasks and stumbles over things that humans find trivial.

Who Uses AI and How: Numbers That Speak for Themselves

AI adoption has become widespread: 88 % of organizations use it, and four out of five university students use generative AI. Within three years, the penetration rate of generative AI reached 53 % of the population, outpacing personal computers and the internet at a comparable stage of their development.

However, the geography of adoption varies greatly. In Singapore, the AI usage rate is 61 %, in the UAE — 54 %, while the US ranks only 24th with 28.3 %. This shows that leadership in development doesn’t always mean leadership in application: countries with well‑thought‑out digital policies and infrastructure can integrate AI into everyday life more quickly.

The Leaders’ Race: The US and China Are Almost Neck‑and‑Neck

The gap between the US and China in terms of model quality has virtually disappeared. Since early 2025, the two countries have repeatedly swapped the lead position, and in March 2026, the Anthropic model — considered the best at the time — outperformed its rivals by only 2.7 %, a difference that could be seen as a statistical margin of error at the industry scale.

Their areas of specialization remain different:

• The US leads in the number of advanced models and high‑impact patents and remains the main hub for AI infrastructure: the country hosts over 5,400 data centers — more than ten times as many as any other country.
• China dominates in the volume of scientific publications, citations, patents, and industrial robot installations.
• South Korea holds the global lead in the number of AI‑related patents per capita.

There’s also a structural vulnerability: nearly all leading AI chips are produced by a single company — TSMC (Taiwan). This makes the global supply chain critically dependent on one manufacturer.

The Economics of AI: There’s Value Even When the Service Is Free

Even though many AI services are available for free, users derive real economic value from them. In the US, this value is estimated at $172 billion per year, and the median value per user has tripled between 2025 and 2026. So even “free” AI use delivers tangible benefits — in the form of time savings, improved work quality, and new opportunities for businesses and education.

Investments are also reaching record levels: in 2025, private investments in AI in the US reached 285.9 billion — more than 23 times higher than China’s 285.9 (12.4 billion). However, the US is losing its appeal to talent: the number of professionals moving to the country has dropped by 89 % since 2017. This creates a paradox: money is flowing into the industry, but talent is redistributing elsewhere.

Safety and Responsibility: Lagging Behind the Progress

The development of AI capabilities is noticeably outpacing the development of safety and accountability mechanisms. Almost all developers publish results from capability tests, but data on safety testing remains limited. The number of recorded AI‑related incidents rose from 233 in 2024 to 362 in 2025 — a worrying signal that as models become more powerful, they also carry greater potential risks.

AI agents have also made significant progress: they’re getting better at handling computer‑based tasks, but they still fail in about one out of every three attempts on structured tests. This means automation is becoming more realistic, but it’s still too early to fully rely on AI for critical processes.

The Cost of Progress: Energy, Water, and Emissions

The rapid development of AI also has an environmental dimension. Training a single model (for example, Grok 4) can generate 72,800 tons of carbon dioxide emissions. The power capacity of data centers supporting AI operations has reached 29.6 GW — comparable to the peak consumption of the state of New York.

Water consumption is equally important: the annual water usage required to run GPT‑4o may exceed the amount of drinking water needed for 12 million people. These figures prompt us to consider how sustainable AI growth will be in the long term and what measures need to be taken now.

Differing Views on the Future: Experts vs. the Public

Attitudes toward AI differ markedly between experts and the general public. A positive impact on work is expected by 73 % of experts — but only 23 % of the population. This gap reflects a serious disconnect in how the technology is understood: professionals see the potential, while the public sees risks and uncertainty. This disconnect is one of the key challenges for policymakers, educational institutions, and tech companies: not only to develop AI but also to explain how it works and how to use it safely.

Overall, the Stanford center’s report paints a picture of a world where AI is no longer an experiment but a powerful, fast‑growing, and heterogeneous force. It offers enormous opportunities but requires a mature approach — to safety, to the environment, to the distribution of benefits, and to risk management. And it’s the balance between progress and responsibility that will determine how beneficial AI will be to society in the coming years.

Powered by Froala Editor