<p>Stanford’s annual AI Index dropped its 2026 edition, and the picture it paints is complicated. Capability is racing ahead. Investment is at record highs. And the gap between what AI insiders believe and what everyone else feels has never been wider.</p>
<h2>The US still leads, but China is closing fast</h2>
<p>In May 2023, the performance gap between top US and Chinese AI models sat somewhere between 17 and 31 percentage points. Today it’s 2.7%. China got there while spending 23 times less on private AI investment — $12.4 billion versus the US’s $285.9 billion.</p>
<p>China also leads in AI patents (69.7% of global filings) and publications (23.2%), and installs industrial robots at nine times the US rate. The US still dominates in notable model releases — 50 in 2025 — and proprietary labs hold a 3.3 percentage point edge over open-source models. But the trajectory is hard to ignore.</p>
<p>One number that stands out: AI talent migration to the US has dropped 89% since 2017, with 80% of that decline happening in the last year alone.</p>
<h2>Benchmarks are falling fast</h2>
<p>A year ago, the best AI models scored 8.8% on Humanity’s Last Exam, a test designed to stump frontier models with graduate-level questions. Now they’re above 50%. Models are approaching 100% on SWE-bench (software engineering tasks) and hitting 93% on graduate-level science.</p>
<p>Real-world reliability still lags behind benchmark performance, but the benchmark numbers are moving fast enough that the gap between “impressive in testing” and “useful in practice” is narrowing.</p>
<h2>Investment hit $581 billion — more than double 2024</h2>
<p>Global AI private investment reached $581 billion in 2025, more than double the 2024 figure. GitHub AI projects hit 5.58 million. Global compute capacity has grown 3.3x annually since 2022, with Nvidia holding over 60% market share.</p>
<p>Generative AI reached 53% population adoption in three years, a faster uptake than most technologies in recent memory.</p>
<h2>The environmental costs are getting harder to ignore</h2>
<p>Training Grok 4 emitted an estimated 72,816 tonnes of CO₂ equivalent. That’s not a rounding error — it’s a real and growing cost that the industry hasn’t figured out how to talk about honestly. The report flags that AI training emissions are rising sharply, and responsible AI development is not keeping pace with capability advances.</p>
<p>AI-related incidents also rose to 362 in 2025, up from 233 the year before.</p>
<h2>The expert-public divide is stark</h2>
<p>Here’s where the report gets uncomfortable. Among AI experts, 56% expect AI to positively impact the US over the next 20 years. Among the general American public, only 10% say they’re more excited than concerned.</p>
<p>The gap is widest on specific domains:</p>
<ul>
<li><strong>Jobs:</strong> 73% of experts are positive, versus 23% of the public</li>
<li><strong>Medical care:</strong> 84% of experts versus 44% of the public</li>
<li><strong>Economy:</strong> 69% of experts versus 21% of the public</li>
</ul>
<p>Gen Z sentiment is shifting in the wrong direction. Excitement dropped from 36% to 22% between 2025 and 2026, while anger rose from 22% to 31%. Globally, the share of people who see AI’s net benefits improved slightly, from 55% to 59% — but that aggregate masks real anxiety in specific demographics.</p>
<p>US trust in government to regulate AI responsibly sits at 31%, the lowest among all surveyed countries.</p>
<h2>Employment picture is mixed</h2>
<p>Entry-level jobs are declining while senior roles are holding steady. The report doesn’t offer a clean narrative here — the data is genuinely mixed, and anyone claiming certainty about AI’s employment impact is probably overstating what we know.</p>
<h2>Worth noting</h2>
<p>The report was funded in part by Google and OpenAI. That doesn’t automatically invalidate the findings, but it’s worth keeping in mind when reading sections that touch on industry self-regulation and responsible AI development.</p>


Collections

Stanford's 2026 AI Index reveals a rapidly shifting landscape: China has closed the US model performance gap to just 2.7% while spending 23x less on private investment. Global AI private investment hit $581 billion in 2025, more than doubling the prior year. Benchmark performance is accelerating — models now score above 50% on Humanity's Last Exam, up from 8.8% a year ago. Environmental costs are rising sharply, with Grok 4 training emitting over 72,000 tonnes of CO₂. A stark expert-public divide persists: 56% of AI experts expect positive US impact over 20 years, versus only 10% of the general public. Gen Z sentiment is worsening, and US trust in government AI regulation is the lowest among all surveyed nations. Employment data remains mixed, with entry-level roles declining while senior positions hold steady.

Stanford's 2026 AI Index finds a wide gap between AI experts and the public