AI Spread Faster Than the Internet, But America Barely Uses It

There’s a number floating around this week that should bother you: 28.3%. That’s the percentage of Americans who regularly use generative AI, according to the 2026 Stanford AI Index, the most comprehensive annual report on the state of artificial intelligence. The country that builds the most AI models, hosts the most data centers, and pours the most money into AI research ranks 24th in the world for actually using the stuff.

Singapore is at 61%. The UAE hits 54%. Meanwhile, the nation that spent $285.9 billion in private AI investment last year can’t get a third of its population to try ChatGPT.

AI Adoption Outpaced the Internet (Yes, Really)

The headline finding from Stanford’s report is staggering: generative AI reached 53% global adoption in just three years. For context, the personal computer took over a decade to hit similar numbers. The internet needed roughly seven years. AI did it in three, and the curve is still pointing up.

But “global adoption” hides a messy reality. The countries where people are most excited about AI aren’t the ones building it. In China, Malaysia, Thailand, Indonesia, and Singapore, over 80% of people expect AI to profoundly change their lives within the next five years. In the US, the mood is more like cautious curiosity mixed with a heavy dose of skepticism.

This tracks with something we’ve been watching closely. When bots started outnumbering humans online, the reaction in the West was alarm. In Southeast Asia, the reaction was closer to “how do I get one working for my business?”

The $172 Billion Nobody Notices

Here’s where it gets interesting. Despite low adoption rates, Stanford estimates that generative AI tools deliver $172 billion in annual value to US consumers alone. The median value per user tripled between 2025 and 2026. So the Americans who do use AI are getting enormous value from it. The rest of the country just hasn’t caught on yet.

The report doesn’t speculate much on why, but you can connect some dots. AI literacy in US schools is basically nonexistent. Four out of five high school and college students use AI for schoolwork, but only half of schools have any AI policy at all. Just 6% of teachers say those policies are clear. Students are adopting the technology faster than institutions can figure out what to do about it.

If you’ve been curious about running AI tools yourself without relying on cloud services, we put together a complete guide to running AI locally on your own machine. It’s easier than most people think.

China Closed the Gap (and Nobody Panicked)

The other bombshell in the report: the US-China AI gap has “effectively closed.” US and Chinese models have been trading the top spot on benchmarks since early 2025. In February 2025, DeepSeek-R1 briefly matched the best American model. As of March 2026, Anthropic leads, but only by 2.7%.

The numbers behind the competition tell a split story. The US still produces more top-tier models and higher-impact patents. China leads in publication volume, citations, total patent output, and industrial robot installations. The US hosts 5,427 data centers, more than ten times any other country. But China is catching up on capability while spending a fraction of the money.

US private AI investment hit $285.9 billion in 2025. China invested $12.4 billion. That’s a 23x gap in spending for a razor-thin gap in performance. Either the US is wildly inefficient, or China has figured out how to do more with less. Probably both.

The Talent Drain Nobody’s Talking About

Buried in the report is a number that should set off alarms: the flow of AI researchers and developers moving to the US has dropped 89% since 2017. An 80% decline happened in the last year alone. The country that built its AI dominance partly on importing global talent is watching that pipeline dry up.

Combined with the adoption gap, this paints a picture of a country that’s winning the AI production race but losing the adoption and talent races simultaneously. It’s like building the world’s best cars while your population prefers to walk.

The Safety Gap Is Even Wider

Perhaps the most uncomfortable finding: responsible AI reporting is falling further behind capability advances. Stanford’s benchmark table for safety and responsible AI is, according to the report, mostly empty. Companies are shipping models faster than anyone can evaluate whether those models are safe.

This isn’t a problem for tomorrow. AI is already finding hundreds of zero-day vulnerabilities in critical software. The question isn’t whether AI will cause problems. It’s whether we’ll have the frameworks in place to catch them when it does.

What This Actually Means

Stanford’s report is 300+ pages of data, and the takeaway is surprisingly simple: AI is spreading faster than any technology in history, but the countries building it aren’t the ones adopting it fastest. The safety infrastructure lags years behind the capability curve. And the talent pipeline that fueled American AI dominance is shrinking fast.

The optimistic read: 53% global adoption in three years means AI is genuinely useful to ordinary people, not just tech workers. The pessimistic read: we’re deploying a world-changing technology with almost no safety guardrails, uneven global adoption, and a shrinking talent base in the country that leads development.

The realistic read? All of the above, happening at the same time. Welcome to 2026.