Every fact, every receipt

Applied Digital had signed long-term hyperscaler leases for 1,410 megawatts of AI data-centre capacity across five campuses as of May 31, 2026, worth approximately $36 billion in contracted revenue over the initial 15-year lease terms 1.41 gigawatts of AI data-centre capacity leased for $36 billion

The total computing power of the stock of AI chips is growing about 3.4× per year

Core Scientific has contracted about 1.1 gigawatts of leased data-center power capacity to AI and HPC customers, including CoreWeave and AMD, representing more than $24 billion of potential revenue, the company said in its second-quarter 2026 results Core Scientific's leased AI data-center capacity: 1.1 gigawatts, over $24 billion in contracts

Data-center landlords are racing to keep pace with AI and cloud demand. Digital Realty's second-quarter 2026 financial supplement to investors disclosed a global construction pipeline of Digital Realty's 1,402 megawatts of IT capacity under construction, 54% already pre-leased

The 84 AI data centers tracked by Epoch AI have a combined real-world power footprint of 16.9 gigawatts as of August 2026, more than New York City's 11 GW peak demand 16.9 GW

Epoch AI rebuilt the cost structure of three AI companies — Anthropic, MiniMax and Z.ai — from IPO filings and reporting. Staff, despite the industry's headline pay packages, take under a quarter of spending. Research and inference compute together take 54% to 62% of an AI company's costs

Epoch AI estimates the single largest owner of AI computing power worldwide is not a chipmaker or a cloud reseller but Google, which held, measured in H100-equivalents as of the fourth quarter of 2025, about one quarter of all global AI compute, mostly from its own custom TPU chips rather than Nvidia GPUs

Traditional computing powers like Germany, Japan and France now play marginal roles in AI infrastructure. Epoch AI's tracking of global GPU cluster performance in May 2025 put China second at 15%, behind a United States hosting about three-quarters of the world's GPU cluster performance

Amazon, Google, Meta, Microsoft and Oracle together held, measured in H100-equivalents as of Q4 2025, an estimated 71% of the world's AI compute

Epoch AI models the world's AI inference capacity, at fixed model size and context length, as growing 3.4× a year — but it puts token demand growth at around 10× a year, three times faster than supply

IREN, a bitcoin miner turned AI cloud operator, told investors in its quarter ended 31 March 2026 that it now holds executed grid connection agreements across seven data-centre sites in Texas, Oklahoma and British Columbia for a combined 4,510 megawatts of AI data-centre power capacity

Epoch AI tracks the largest AI data centre in the world by computing capacity, and the record has changed hands repeatedly since August 2024 — Colossus, Anthropic and Amazon's New Carlisle, Microsoft's Fairwater Atlanta, Meta's Prometheus. Each time, the bar has risen at the same pace: a doubling every seven months

The largest AI data centre in Epoch AI's tracker, Colossus 2 in Memphis, Tennessee, holds a computing capacity of 1,112,000 H100-equivalent AI chips

SpaceX listed on Nasdaq in June 2026, and its first quarterly report as a public company discloses a metric no rocket company used to publish: installed GPUs multiplied by their all-in power draw, before any cooling or facility overhead. At 30 June 2026 that came to 1.4 gigawatts of installed GPU power draw, 3.5× a year earlier

Tesla reports its AI training compute the way it reports factory capacity — in installed megawatts. At the end of the second quarter of 2026 both Texas clusters, Cortex 1 and Cortex 2, were in production, and the compute installed there had more than doubled in six months, to over 205 MW of AI training capacity

Compute used to train frontier language models has grown roughly 5× per year since 2020