AI Infrastructure: chart data from AI Adoption (https://aiadoption.org/ai-infrastructure-statistics/)

Data from the AI Index Report 2026 (Stanford HAI) and the partners named below. Licence: CC BY-ND 4.0, as for the AI Index.
Each CSV is the data behind one chart. Cite the chart page and its original source.

ai-chip-peak-performance.csv  The fastest AI chips now reach 2.5 petaFLOP/s at 16-bit precision
    Source: Epoch AI, 2026
    Chart: https://aiadoption.org/data/ai-chip-peak-performance/
ai-models-trained-by-chip.csv  84 notable AI models have been trained on Nvidia A100 chips, more than on any other accelerator
    Source: Epoch AI, 2026
    Chart: https://aiadoption.org/data/ai-models-trained-by-chip/
global-ai-compute-capacity.csv  AI chips worldwide reached 17.1 million H100-equivalents by end-2025; Nvidia supplied two-thirds
    Source: Epoch AI, 2026
    Chart: https://aiadoption.org/data/global-ai-compute-capacity/
gpu-computation-cost-decline.csv  GPU computation is 99.8% cheaper than in 2006
    Source: International Energy Agency (2025), Energy and AI
    Chart: https://aiadoption.org/data/gpu-computation-cost-decline/
ai-data-center-power-capacity.csv  AI data centre power capacity reached 29.6 GW by end-2025, close to New York state's peak demand
    Source: Epoch AI, 2026
    Chart: https://aiadoption.org/data/ai-data-center-power-capacity/
data-centers-world-map.csv  The US has 47% of the world's tracked data centres: 5,427 of 11,614
    Source: Cloudscene, 2025
    Chart: https://aiadoption.org/data/data-centers-world-map/
data-centers-by-country.csv  The US has 5,427 data centres, more than ten times Germany, the next largest
    Source: Cloudscene, 2025
    Chart: https://aiadoption.org/data/data-centers-by-country/
public-ai-supercomputers-by-region.csv  China had 85 public or public-private AI supercomputers in 2025, as many as Europe and North America combined
    Source: Epoch AI, 2026
    Chart: https://aiadoption.org/data/public-ai-supercomputers-by-region/
ai-chip-energy-efficiency.csv  The most efficient AI chips now do 2.5 trillion operations per second per watt
    Source: Epoch AI, 2026
    Chart: https://aiadoption.org/data/ai-chip-energy-efficiency/
power-draw-to-train-frontier-models.csv  Training Grok 3 drew an estimated 110 MW, the most of any model with published data
    Source: Epoch AI, 2026
    Chart: https://aiadoption.org/data/power-draw-to-train-frontier-models/
ai-model-energy-per-prompt.csv  The most energy-hungry 2025 models used up to 23 Wh per medium-length prompt
    Source: Jegham et al., 2025
    Chart: https://aiadoption.org/data/ai-model-energy-per-prompt/
chatgpt-query-energy-vs-everyday.csv  A day of eight medium GPT-4o queries uses about as much energy as charging two phones
    Source: AI Index, 2025
    Chart: https://aiadoption.org/data/chatgpt-query-energy-vs-everyday/
ai-accelerator-power-demand.csv  AI accelerator chips sold in 2023–24 draw an estimated 5,235 MW; Nvidia's account for half
    Source: de Vries-Gao, 2025
    Chart: https://aiadoption.org/data/ai-accelerator-power-demand/
ai-systems-power-demand.csv  Complete AI systems shipped in 2023–24 draw an estimated 9,357 MW, more than Austria's national demand
    Source: de Vries-Gao, 2025
    Chart: https://aiadoption.org/data/ai-systems-power-demand/
ai-power-demand-vs-countries.csv  AI hardware shipped in 2023–24 draws more power than Austria; bitcoin mining draws twice as much
    Source: de Vries-Gao, 2025
    Chart: https://aiadoption.org/data/ai-power-demand-vs-countries/
carbon-emissions-training-ai-models.csv  Training Grok 4 emitted an estimated 72,816 tonnes of CO₂, against 0.01 tonnes for AlexNet in 2012
    Source: AI Index, 2026; Strubell et al., 2019
    Chart: https://aiadoption.org/data/carbon-emissions-training-ai-models/
model-size-vs-training-emissions.csv  Training DeepSeek v3 emitted a fifteenth of the CO₂ of Llama 3.1 405B, despite more parameters
    Source: AI Index, 2026
    Chart: https://aiadoption.org/data/model-size-vs-training-emissions/
chatgpt-water-use.csv  GPT-4o's inference uses an estimated 1.3–1.6 million kilolitres of water a year
    Source: AI Index, 2025
    Chart: https://aiadoption.org/data/chatgpt-water-use/
