Stanford's Human-Centered AI Institute released its annual AI Index report yesterday. Two findings stand out: China has effectively closed the performance gap with the United States, and the leading AI companies now share much less about how their models are built.

The US lead over China is almost gone
The lead of the top US model over the top Chinese model narrowed from 9.26% in January 2024 to 1.70% by February 2025. US and Chinese models now trade places at the top of major benchmarks.
The US keeps real advantages in capital, infrastructure and chip development. American companies attract more private investment and have access to the latest hardware from NVIDIA and AMD. On model capability, though, the gap has essentially closed. China leads on patents, publications and autonomous robotics, which the industry now calls "physical AI".
For Gulf governments building sovereign AI programmes, this widens the field of credible technology partners beyond Silicon Valley.
Companies are disclosing less
The Foundation Model Transparency Index measures how openly companies share information about their AI systems. Its average score fell from 58 points last year to 40 in 2026. A fall that size looks like a deliberate choice by the developers.
Google, Anthropic and OpenAI have all stopped disclosing their latest models' dataset sizes and training duration. Of the 95 most notable models launched in 2025, 80 were released without their training code. Over 90% of notable AI models now come from private companies rather than academic institutions.
This makes life harder for regulators assessing safety, for researchers checking capability claims, and for any organisation trying to choose a model on evidence.
Adoption is faster than any earlier technology
According to the report, 53% of the global population now uses generative AI regularly. That makes it the fastest-adopted technology in history, ahead of smartphones, the internet and electricity.
The regional differences are large. In China, Malaysia, Thailand, Indonesia and Singapore, over 80% of citizens expect AI to profoundly affect their lives within three to five years. The United States, home to most AI development, ranks only 24th in adoption, with 28.3% regular usage.
Consumer surplus from generative AI in the US alone reached $172 billion in 2026, and corporate investment has grown 40-fold since 2013.
The report also records the environmental cost. Training xAI's Grok 4 produced over 72,000 tons of CO2, and running GPT-4o inference uses enough water to sustain 12 million people. The UAE and its neighbours are investing heavily in AI infrastructure, so these figures belong in their planning. The region has the resources to try more sustainable approaches, such as solar-powered data centres and the efficiency gains now coming out of research labs.
Public trust lags behind
Only 31% of US citizens trust their government to regulate AI properly. In China the figure is 27%. The EU leads at 53%, which may reflect the AI Act and more visible regulatory work.
Optimism and anxiety are rising together. The share of people who feel optimistic about AI's benefits rose to 59% (from 52%), while nervousness about the technology also rose, to 52%.
Jobs data adds to the unease. Employment for software developers aged 22 to 25 has fallen nearly 20% since 2022. Whether that comes from AI or from wider economic conditions is still debated.
A practical response for UAE organisations
Two points from the report matter most for teams buying AI here. Strong models now come from more than one country, so it is reasonable to compare Chinese, American and domestic options on your own tasks. And since the big developers have stopped volunteering information, transparency will have to be requested, by regulators and by buyers.
Write disclosure into procurement. Before signing, ask the vendor for what the developer will share about training data, data handling and evaluation results, and record what they decline to provide. That record will be useful when regulators start asking the same questions.