Invest1 publisher3 min readPublished
Bitcoin miners trade at 13 times sales on AI capacity that is 14% switched on
Public miners have announced $70 billion to $100 billion of AI and HPC contracts. Spread across 12-to-15-year terms that is roughly $6 billion a year, and 550 of about 4,000 contracted megawatts are earning it today.
The Investor · Invest desk
What happened
- Public Bitcoin miners have announced AI and high-performance computing contracts worth an estimated $70 billion to $100 billion, according to cryptobriefing.com.
- Of roughly 4 gigawatts contracted across these companies, about 550 megawatts is generating revenue, at an estimated $1.1 billion to $1.5 billion annualized.
- An estimated 35 exahashes per second of mining capacity is expected to move away from publicly traded miners as power goes to AI tenants.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- constraint The scarce input is the grid connection. US interconnection can take years, so a hyperscaler that wants power this year has to rent someone else's substation.
- decision Every megawatt a board moves to an AI tenant is a megawatt that stops earning bitcoin, so the choice is contracted rent at three times the revenue against keeping commodity upside on the block reward.
- exposure At 12 to 13 times forward sales, shareholders are paying today for about $5.2 billion a year of revenue that sits behind construction and energisation schedules the companies have not disclosed quarter by quarter.
- contradiction The revenue already flowing implies $2.0 million to $2.7 million per megawatt, well above the $1.5 million benchmark used to size the pipeline, so either the mix is richer than leasing or the two figures are measuring different things.
Core Scientific's deal is the one with all three numbers in it: $10.2 billion, 12 years, 590 megawatts of capacity leased to CoreWeave [6]. Divide it out and you get about $850 million a year, or roughly $1.44 million per megawatt per year [1]. That is within rounding of the $1.5 million per megawatt that cryptobriefing.com gives as the annual take from an AI tenant [2].
About 4 gigawatts are contracted across these companies [11]. At the benchmark rate that is around $6 billion a year [2], and $6 billion over a 12-year term is $72 billion [3] - the bottom of the $70 billion to $100 billion range of announced contracts [1]. The headline figure is a term total.
The realized side does not sit neatly against that benchmark. At $1.5 million per megawatt, the roughly 550 megawatts currently producing revenue [11] would earn about $825 million a year. But cryptobriefing.com puts annualized revenue from leased capacity at $1.1 billion to $1.5 billion, implying $2.0 million to $2.7 million per megawatt [4]. The source does not break the 550 megawatts down by type. Product mix explains some of it. IREN's Microsoft agreement is for GPU cloud services, roughly $9.7 billion over five years [8], which is $1.94 billion a year on its own, more than the current run-rate for the whole group [7]. Selling GPU hours and renting a powered shell are different businesses.
What the miners give up for this is bitcoin production. Mining yields about $500,000 per megawatt against the AI tenant's $1.5 million [2][3], a gap the 2024 halving widened when block rewards were cut in half [5]. The conversion is already visible: an estimated 35 exahashes per second is expected to move away from public miners [14].
The tenant is buying the interconnection. Connecting a new facility to the US grid can take years, and the miners spent those years securing power purchase agreements, building substations and negotiating interconnection rights [10][16]. A hyperscaler cannot reproduce that paperwork on its own timetable. That is why a 12-year lease exists at all.
AI-focused miners trade at about 12 to 13 times next-12-month sales against 3.7 to 6 times for miners still pointed at bitcoin [13]. At midpoints that is a spread of about 2.6 times [6], stacked on top of the 3x revenue premium per megawatt [4]. The equity is paying for the 3,450 contracted megawatts not yet earning, about $5.2 billion a year of revenue at the benchmark rate [5].
I would expect that multiple to compress toward the bitcoin-miner range if the in-service figure does not move well past 550 megawatts over the next several quarters. The 12-to-13x is a claim on deployment. If realized revenue per megawatt settles nearer $2.7 million than $1.5 million, the same 4 gigawatts is worth about $10.8 billion a year [10]. And Hut 8's leases run 15 years at $7 billion to $9.8 billion each, or $467 million to $653 million a year apiece [9][8]. Over a term that long, a slow ramp costs a shareholder less than a missed quarter suggests. Analysts, per cryptobriefing.com, are now judging these companies on energy capacity and tenant quality instead of hashrate or bitcoin produced [15].
What to watch
- Whether megawatts in service climb materially past 550 in the next few quarterly reports, and on what disclosed schedule.
- Whether reported HPC revenue per megawatt settles near $1.5 million or nearer the $2.0-$2.7 million the current run-rate implies.
- Whether the 35 exahashes per second actually leaves public miners' fleets, or gets retained on cheaper sites.