What does the 2026 AI Index say about US vs China?

Nearly tied on model performance per Arena rankings, with US leading in data centers and funding, China in papers, patents, robots.

Are AI models really plateauing?

No — scores keep soaring on benchmarks like SWE-bench (60% to 100%), but real-world gaps persist, especially robotics.

Why is AI power and water use exploding?

Data centers hit 29.6 GW power draw; GPT-4o alone may use water for 12 million people annually.

🏛️ Governance & Ethics

[2026 AI Index] US-China Neck-and-Neck, Benchmarks Broken

AI's not slowing down—it's sprinting past our grasp. Stanford's 2026 AI Index charts the chaos: neck-and-neck US-China race, power guzzling data centers, and benchmarks in tatters.

Legal AI Beat Apr 16, 2026 3 min read

Charts from Stanford's 2026 AI Index showing US-China model rankings, power usage, and benchmark scores

⚡ Key Takeaways

US and China nearly tied in AI model performance, shifting competition to cost and reliability. 𝕏
AI models excel on benchmarks but show 'jagged intelligence'—strong in text, weak in physical tasks. 𝕏
Benchmarks are flawed, gamed, and outdated; transparency from companies is vanishing. 𝕏
Massive costs: 29.6 GW power, huge water use, fragile TSMC chip supply chain. 𝕏