If you happen to’re following AI information, you’re in all probability getting whiplash. AI is a gold rush. AI is a bubble. AI is taking your job. AI can’t even learn a clock. The 2026 AI Index from Stanford College’s Institute for Human-Centered Synthetic Intelligence, AI’s annual report card, comes out right now and cuts via a few of that noise.
Regardless of predictions that AI improvement could hit a wall, the report says that the highest fashions simply preserve getting higher. Persons are adopting AI quicker than they picked up the private laptop or the web. AI corporations are producing income quicker than corporations in any earlier expertise increase, however they’re additionally spending a whole bunch of billions of {dollars} on information facilities and chips. The benchmarks designed to measure AI, the insurance policies meant to control it, and the job market are struggling to maintain up. AI is sprinting, and the remainder of us are looking for our sneakers.
All that velocity comes at a value. AI information facilities all over the world can now draw 29.6 gigawatts of energy, sufficient to run the complete state of New York at peak demand. Annual water use from operating OpenAI’s GPT-4o alone could exceed the ingesting water wants of 12 million folks. On the similar time, the availability chain for chips is alarmingly fragile. The US hosts a lot of the world’s AI information facilities, and one firm in Taiwan, TSMC, fabricates nearly each main AI chip.
The information reveals a expertise evolving quicker than we will handle. Right here’s a take a look at a few of the key factors from this yr’s report.
The US and China are almost tied
In a protracted, heated race with immense geopolitical stakes, the US and China are nearly neck and neck on AI mannequin efficiency, in keeping with Enviornment, a community-driven rating platform that enables customers to match the outputs of huge language fashions on similar prompts. In early 2023, OpenAI had a lead with ChatGPT, however this hole narrowed in 2024 as Google and Anthropic launched their very own fashions. In February 2025, R1, an AI mannequin constructed by the Chinese language lab DeepSeek, briefly matched the highest US mannequin, ChatGPT. As of March 2026, Anthropic leads, trailed intently by xAI, Google, and OpenAI. Chinese language fashions like DeepSeek and Alibaba lag solely modestly. With the most effective AI fashions separated within the rankings by razor-thin margins, they’re now competing on value, reliability, and real-world usefulness.
The index notes that the US and China have completely different AI benefits. Whereas the US has extra highly effective AI fashions, extra capital, and an estimated 5,427 information facilities (greater than 10 instances as many as another nation), China leads in AI analysis publications, patents, and robotics.
As competitors intensifies, corporations like OpenAI, Anthropic, and Google not disclose their coaching code, parameter counts, or data-set sizes. “We don’t know a lot of things about predicting model behaviors,” says Yolanda Gil, a pc scientist on the College of Southern California who coauthored the report. This lack of transparency makes it tough for unbiased researchers to review how you can make AI fashions safer, she says.
AI fashions are advancing tremendous quick
Regardless of predictions that improvement will plateau, AI fashions preserve getting higher and higher. By some measures, they now meet or exceed the efficiency of human consultants on assessments that goal to measure PhD-level science, math, and language understanding. SWE-bench Verified, a software program engineering benchmark for AI fashions, noticed high scores bounce from round 60% in 2024 to nearly 100% in 2025. In 2025, an AI system produced a climate forecast by itself.
“I am stunned that this technology continues to improve, and it’s just not plateauing in any way,” says Gil.
Nevertheless, AI nonetheless struggles in loads of different areas. As a result of the fashions be taught by processing monumental quantities of textual content and pictures fairly than by experiencing the bodily world, AI reveals “jagged intelligence.” Robots are nonetheless of their early days and reach solely 12% of family duties. Self-driving automobiles are farther alongside: Waymos are actually roaming throughout 5 US cities, and Baidu’s Apollo Go automobiles are shuttling riders round in China. AI can be increasing into skilled domains like regulation and finance, however no mannequin dominates the sphere but.
However the best way we check AI is damaged
These stories of progress must be taken with a grain of salt. The benchmarks designed to trace AI progress are struggling to maintain up as fashions rapidly blow previous their ceilings, the Stanford report says. Some are poorly constructed—a preferred benchmark that assessments a mannequin’s math talents has a 42% error price. Others could be gamed: when fashions are educated on benchmark check information, for instance, they’ll be taught to attain properly with out getting smarter.
AI corporations are additionally sharing much less about how their fashions are educated, and unbiased testing typically tells a unique story from what they report. “A lot of companies are not releasing how their models do in certain benchmarks, particularly the responsible-AI benchmarks,” says Gil. “The absence of how your model is doing on a benchmark maybe says something.”
AI is beginning to have an effect on jobs
Inside three years of going mainstream, AI is now utilized by greater than half of individuals all over the world, a price of adoption quicker than the private laptop or the web. An estimated 88% of organizations now use AI, and 4 in 5 college college students use it.
It’s early days for deployment, and AI’s impression on jobs is difficult to measure. Nonetheless, some research recommend AI is starting to have an effect on younger employees in sure professions. In keeping with a 2025 research by economists at Stanford, employment for software program builders aged 22 to 25 has fallen almost 20% since 2022. The decline won’t be pinned on AI alone, as broader macroeconomic circumstances may very well be responsible, however AI seems to be taking part in an element.
Employers say that hiring could proceed to tighten. In keeping with a 2025 survey performed by McKinsey & Firm, a 3rd of organizations count on AI to shrink their workforce within the coming yr, notably in service and provide chain operations and software program engineering. AI is boosting productiveness by 14% in customer support and 26% in software program improvement, in keeping with analysis cited by the index, however such positive aspects will not be seen in duties requiring extra judgment. General, it’s nonetheless too early to know the larger financial impression of AI.
Individuals have sophisticated emotions about AI
World wide, folks really feel each optimistic and anxious about AI: 59% of individuals suppose that it’ll present extra advantages than drawbacks, whereas 52% say that it makes them nervous, in keeping with an Ipsos survey cited within the index.
Notably, consultants and the general public see the way forward for AI very in a different way, in keeping with a Pew survey. The largest hole is round the way forward for work: Whereas 73% of consultants suppose that AI can have a optimistic impression on how folks do their jobs, solely 23% of the American public thinks so. Consultants are additionally extra optimistic than the general public about AI’s impression on training and medical care, however they agree that AI will harm elections and private relationships.
Amongst all nations surveyed, People belief their authorities least to control AI appropriately, in keeping with one other Ipsos survey. Extra People fear federal AI regulation gained’t go far sufficient than fear it is going to go too far.
Governments are struggling to control AI
Governments all over the world are struggling to control AI, however there have been some minor successes final yr. The EU AI Act’s first prohibitions, which ban the usage of AI in predictive policing and emotion recognition, took impact. Japan, South Korea, and Italy additionally handed nationwide AI legal guidelines. In the meantime, the US federal authorities moved towards deregulation, with President Trump issuing an govt order in search of to handcuff states from regulating AI.
Regardless of this federal motion, state legislatures within the US handed a report 150 AI-related payments. California enacted landmark laws, together with SB 53, which mandates security disclosures and whistleblower protections for builders of AI fashions. New York handed the RAISE Act, requiring AI corporations to publish security protocols and report essential security incidents.
However for all of the legislative exercise, Gil says, regulation is operating behind the expertise as a result of we don’t actually perceive the way it works. “Governments are cautious to regulate AI because … we don’t understand many things very well,” she says. “We don’t have a good handle on those systems.”
