Amidst the current climate surrounding artificial intelligence, resembling a gold rush or a bubble, the 2026 AI Index from Stanford University's Institute for Human-Centric Artificial Intelligence has recently been released. This annual AI report aims to dispel some of the hype surrounding its developments and prospects, highlighting

steady improvement .

Despite some predictions that AI development may have reached a standstill, the report indicates that the best models continue to improve. Today, people are adopting AI at a faster pace than they adopted personal computers or the internet. AI companies are generating revenue faster than companies in any previous technological boom. Yet, at the same time, they are spending hundreds of billions of dollars on data centers and chips.

Meanwhile, standards designed to measure AI, policies regulating it, and the job market are struggling to keep up. AI is certainly advancing at breakneck speed, and we are struggling to adapt. Of course, all this speed comes at a price. Currently, AI data centers around the world consume 29.6 gigawatts of power—enough to supply the entire state of New York with electricity at peak times. The annual water consumption to run OpenAI’s GPT-40 model alone may exceed the drinking water needs of 12 million people. At the same time, the chip supply chain is alarmingly fragile.

While the United States hosts most of the world’s AI data centers, Taiwan’s TSMC manufactures virtually all of the leading AI chips.

A US-China rivalry is unfolding, and the data reveals technology evolving faster than we can keep up. Here’s a look at some of the key takeaways from this year’s report.

In a long and heated race with enormous geopolitical stakes, the US and China are locked in a fierce competition over the performance of their AI models, according to Arena, a community-based ranking platform that allows users to compare the output of large language models based on identical texts.

* Model Performance: In early 2023, OpenAI held a lead thanks to ChatGBT, but that gap narrowed in 2024 with the release of Google and Anthropic’s own models.

In February 2025, R1, an AI model developed by China’s DeepSec Lab, briefly performed as well as ChatGBT, the top US model. By March 2026, Anthropic topped the list, followed closely by XAI, Google, and OpenAI. Chinese models, such as DeepSec and Alibaba, lagged behind.

* Cost, Capital, and Research. As the top AI models narrowed the rankings, the competition now centered on cost, reliability, and practicality.

The index revealed that the US and China have different advantages in AI. While the US boasts more robust AI models, greater capital, and an estimated 5,427 data centers (more than ten times that of any other country), China leads in AI research publications, patents, and robotics.

As competition intensified, companies like OpenAI, Anthropic, and Google began withholding their training code, the number of transactions in their models, and the size of their datasets. Technology Review quoted Yolanda Gill, a computer scientist at the University of Southern California and co-author of the report, as saying, "We don't know much about predicting model behavior," adding that this lack of transparency makes it difficult for independent researchers to study how to make AI models more secure.

Rapid Evolution

* High Performance and "Flickering Intelligence." Despite predictions that AI development would eventually plateau, AI models continue to improve. By some measures, they now match or surpass the performance of human experts on tests designed to measure PhD-level understanding of science, mathematics, and language. The SWE-bench Verified test, a software engineering benchmark for artificial intelligence models, saw a significant jump in top scores, from around 60 percent in 2024 to nearly 100 percent in 2025. An AI system independently produced weather forecasts,

and Jill expressed her amazement at "the continued development of this technology; it never stops." However, this doesn't mean that AI is without its challenges in many other areas. Because the models learn by processing vast amounts of text and images, rather than experiencing the physical world, AI exhibits "inconsistent intelligence."

* Robotics' success. Today, robots are still in their infancy, successfully performing only about 12 percent of household tasks. Self-driving cars have come a long way: Waymo vehicles now operate in five US cities, and Baidu's Apollo Go vehicles transport passengers across China. Artificial intelligence is also expanding into professional fields such as law and finance, but no single model has yet become dominant in these areas.