Explainer

Who is ahead in AI

The File desk · fetched filings and named boards

Public comparable numbers. Estimates are marked. We only chart when the unit is the same — n/a is omitted, not a zero bar.

Compiled Wednesday, September 2, 2026

Updated Wednesday, September 2, 2026

Energy

TWh · Ember yearly electricity · 2025 Est.

This is electricity made each year, not chips. China’s line roughly doubled. The US line stayed almost flat, then ticked up. India rose, still far below both. 2025 is an estimate.

More detail

Ember Yearly Electricity Data via OWID electricity-generation extract. 2015–2024 Reported; 2025 Estimate from monthly data, per Ember’s supporting materials.

Other Ember 2025 Estimates, not on the chart: Russia 1,193 TWh; Japan 1,030 TWh; Brazil 751 TWh; Canada 652 TWh; South Korea 625 TWh.

Do not mix Energy Institute 2025 onto these lines. EI US 4,772 TWh is gross, about 250 TWh above Ember’s US 4,520. EI China 10,575; India 2,055.

EIA check, not a second line: Electric Power Monthly US 2024 4,309 TWh final; 2025 about 4,430 TWh preliminary.

Data-centre use, different unit: DOE / LBNL US 58 TWh in 2014 to 176 TWh in 2023 (Reported). IEA Estimate, world: 415 TWh in 2024; 485 TWh in 2025 to 950 TWh in 2030 (Base Case). DOE hub Estimate: 11.8% of U.S. electricity by 2030 (9.5–15.3%).

Compute

TWh · IEA data-centre electricity · 2024 Est. · 2030 Base Case proj.

This is electricity used by data centres, not how many chips anyone owns. The US line starts almost twice China’s. IEA thinks both rise by 2030; China’s climbs faster in percent, the US still has the higher bar. 2024 is modelled. 2030 is their Base Case, not a measurement. We do not have the years in between as a table, so those years are not drawn.

World about 415 TWh in 2024, 1.5% of world electricity — caption, not a fourth line. 2030 Europe is about 115 TWh; IEA said more than +45 TWh vs 2024. The 2020–2030 regional chart was not released as a downloadable table.

More detail

Companies — money for chips and buildings, 2026

Billions of dollars · 0 to the longest bar

This is money these companies plan to spend this year on buildings and machines. Amazon’s bar is tens of billions longer. They do not count every dollar the same way.

DOE / LBNL is a different US-only study — not on the IEA lines: 58 TWh in 2014 to 176 TWh in 2023.

Accelerators are not installed stock: Amazon plans more than 1 million NVIDIA GPUs starting 2026 and 2 million more in 2027–28 (Plan); Anthropic has access to more than 220,000 at Colossus 1 (Access); Nvidia’s $89.0 billion Data Center quarter is sold chips.

TSMC Q2 $40.20 billion; 77% of wafer revenue at 7-nanometer and below; SK hynix HBM4 mass shipments in Q2 2026. Factory dollars and memory shipments are not one scale.

Power contracts, not one scale: Anthropic Amazon up to 5 GW and Google 5 GW TPUs from 2027; Microsoft +1 GW this quarter; Oracle more than 1.2 GW delivered in FY26 (Partially confirmed).

IEA Estimate, extra data-centre electricity 2024–30: United States about +240 TWh, China about +175 TWh, Europe more than +45 TWh.

Cloud share, Synergy Q2 Estimate: AWS 28, Azure 20, Google Cloud 15. Same-quarter dollars: AWS $42.2 billion vs Google Cloud $24.8 billion, up 82%.

Models

Hard public test (Terminal-Bench 4.0)

Percent of 66 tasks · 0 to the longest bar

Anthropic Opus 5 — United States51.8%src
Zhipu GLM-5.3 — China41.8%src
OpenAI GPT-5.6 Sol — United States37.3%src
xAI Grok 4.6 — United States20.3%src

A hard test of 66 computer tasks. Three bars are US labs. China’s GLM-5.3 beats OpenAI’s Sol here and still trails Anthropic. Longer bar = more tasks solved.

More detail

Fable 5.1 55.8% / Mythos 5.1 60.9% are Anthropic-reported — not on this independent Snorkel board, so not on the bars.

API list price (input $ per million tokens), not training cost: DeepSeek V4-Pro $1.32; Grok 4.6 $2; Kimi K3 $3; Sol promo $4 through Nov. 21, 2026; Fable 5.1 $10. Gemini list prices did not load.

Snorkel full-run bill, not training: Opus 5 $6.0k; Fable 5 $7.3k; GLM-5.3 $2.7k; Sol $2.5k; Luna $0.3k (same board).

Usage, not capability: Gemini App 950 million monthly users in Q2 2026. OpenAI user posts did not load.