BTC: Mining against AI Data Center Leases per Megawatt
Shown as its 2019–2022 sample; the latest readings and full history are included with Plus.
What a megawatt earns in a day under the AI data center leases listed bitcoin miners have signed, against what it earns mining bitcoin with the network's average profitable machine or the most efficient machine.
Why use it. Listed miners are turning their power sites into data centers for AI companies, and this gap is why: a lease fixes a higher revenue per megawatt for 10 to 25 years, while mining revenue halves at each halving and moves with the price and difficulty. How to read it. Read the lease line against the mining lines. The lease is rent per megawatt of critical IT capacity under triple-net leases, where the tenant also pays for the power; mining revenue is before the power bill and counts every megawatt the machines draw. The lease line steps when a new lease is announced.
Calculation. Lease: each lease's base-term contract value ÷ its term in days ÷ its critical IT megawatts, as the company states them in the release it filed with the SEC; the line is the median over the leases announced up to each day (Core Scientific with CoreWeave; TeraWulf with Core42, a Fluidstack joint venture and Anthropic; Cipher with Fluidstack and Amazon Web Services; Hut 8 with Fluidstack and two at Beacon Point; Riot with AMD and two AI labs; Bitdeer at Tydal; the list with each filing is at /api/v1/mining/ai-leases). Contract values are as announced, not revenue earned yet; renewal options, GPU cloud contracts and programme totals are left out. Mining: the energy hashprice × 24, as in Mining Revenue per Megawatt. Each day uses the seven UTC days ending with it, to smooth block luck: hash rate is those blocks' work (difficulty × 2^32 each) over the window's seconds, and revenue per unit of hash rate is their claimed coinbase BTC (subsidy plus fees) over that work, at the day's BTC/USD close. A machine counts from the middle of the quarter its first units shipped, for four years (machines are written off over three to five years). It is profitable when its revenue covers its electricity at $0.05/kWh with a power usage effectiveness (PUE, the site's overhead for cooling and power conversion) of 1.10. The best guess is the plain average efficiency (J/TH) of the profitable machines, each model counted once; the lower bound puts all hash rate on the most efficient machine available, the upper bound on the least efficient profitable one; with no profitable machine all three use the most efficient. The model starts on 1 January 2014, when ASICs had replaced GPU and FPGA mining. It knows no real fleet: it assumes miners run machines that pay for their power and that every model is equally common, so cheap or free power, machines not yet installed, curtailment and each miner's real tariff move the true figure; the bounds show a reasonable range. It is an estimate, not a measurement. Leases: the companies' own releases filed with the SEC (EDGAR), read by CryptoKeyhole. Mining: A model computed by CryptoKeyhole from its own Bitcoin node (each block's difficulty, claimed coinbase value, height and header time), the site's admitted BTC/USD daily closes, and a list of the main Bitcoin mining machines of Bitmain, MicroBT, Canaan, Bitdeer and Innosilicon from the first ASICs of 2013 to today, each with its manufacturer's nominal rating (TH/s and watts) and the quarter its first units shipped (the list: /api/v1/mining/hardware). Days are UTC days by block header time; the open day is partial. BTC/USD prices are the site's admitted boundary prices: daily closes from The Bitcoin Time Machine (CC BY 4.0) until 2 October 2026, then quotes recorded from Alternative.me at each boundary (within 15 minutes, never carried forward). Hourly values start on 3 October 2026, when hourly quotes begin.