12 Charts That Explain the State of AI in 2026

Based on the Stanford AI Index 2026 and IEEE Spectrum commentary, here are 12 charts summarizing the state of AI in 2026.

• The United States still leads in notable AI model releases, but China is clearly closing the gap.
• Industry dominates. Over 90% of notable models in 2025 came from industry.
• China leads in robotics deployment. In 2024, industrial robot installations reached 295,000 units, far ahead of Japan (44,500) and the United States (34,200).
• Global AI compute capacity has grown 3.3x annually since 2022, and is now 30x larger than in 2021. Nvidia GPUs currently account for over 60% of the total.
• Carbon emissions from frontier LLM training are rising. Estimates for Grok 4 exceed 72,000 tons CO2e, while a separate estimate from Epoch AI puts the figure as high as roughly 140,000 tons.
• Inference efficiency also varies widely. The least efficient models emit over 10x more than the most efficient ones.
• Benchmark performance is rising rapidly. In particular, agentic AI and multimodal LLMs have improved sharply on metrics like OSWorld and SWE-Bench Verified.
• Humanity's Last Exam scores rose from 8.8% for OpenAI o1 in 2025 to 38.3% in 2026, with some of the latest models surpassing 50%.
• AI research is expanding in healthcare. Drug discovery papers have more than doubled in two years, and multimodal biomedical AI papers have increased 2.7x.
• By contrast, everyday cognitive tasks remain weak. On ClockBench, even the top-performing GPT-5.4 achieved only about 50% accuracy in reading analog clocks.
• 2025 AI investment hit an all-time high of $581B, far surpassing 2024's $253B, with the United States absorbing over $344B.
• The number of AI-related projects on GitHub has grown to 5.58M, roughly 5x the 2020 level. Stars and high-star projects have grown at a similar pace, suggesting genuine human engagement is being sustained.
• The employment impact remains unclear. Entry-level roles have declined in software development and customer support, but overall unemployment patterns cannot be explained by AI exposure alone.
• Public opinion has improved slightly. In an Ipsos survey, 59% of respondents said AI's benefits outweigh its drawbacks, up from 55% in 2024, though 52% still said they find AI products unsettling.