Bitcoin Miners Are Becoming AI Landlords. But Which Leases Are a Positive Investment Signal?

Miners signed AI leases worth more than their own market caps, then the stocks cracked. A four-question screen for sorting bankable contracts from hopeful ones.

The best story in bitcoin mining hasn’t had anything to do with bitcoin in a long time. Years ago, this business (which long lived on the spread between bitcoin mining revenue and power bills) found a richer tenant next door. The AI industry needed power in gigawatts, permitted and ready to energize, and miners had spent multiple cycles acquiring exactly that. So they stopped selling only Bitcoin hashrate and started leasing megawatts to the AI labs training frontier models.

The receipts piled up fast. This month the U.S. bitcoin miner TeraWulf struck a 20-year deal with Anthropic that the company values around $19 billion over the term — more than TeraWulf’s entire market capitalization for a single campus in Hawesville, Kentucky.

Nevada-based bitcoin miner CleanSpark inked a $6.6 billion, 20-year lease in Georgia. Equity research firm Benchmark took its Hut 8 target from $85 to $165, then raised it again to $195 eight days later, and started calling it a “power-first data center REIT.” Bitcoin treasury company Empery Digital went further and sold nearly half its bitcoin stack to help fund a data center stake. Basically, you buy up miners while the market still values them on hash prices, and collect the rerate before Wall Street figures out they’re really infrastructure.

Logic Problems

Miners ETF WGMI, the exchange-traded fund of bitcoin mining stocks, more than doubled over the past year while bitcoin dropped by almost half. Then from its June 18 high to July 17, it fell 34% from $72.10 to $47.57. For most of that slide the selling was conventional: sentiment cooled across AI infrastructure and semiconductors broadly, the giant lease headlines invited profit-taking, and doubts crept in about whether the miners could actually capture AI’s demand for compute, even as analysts kept raising targets.

On July 16, the Chinese AI lab Moonshot released Kimi K3, a soon-to-be open-weight model that debuted at number one on LMArena’s Frontend Code Arena, a major leaderboard that ranks AI models on frontend coding, though it still trails the closed frontier, Fable 5 and GPT-5.6, on overall intelligence. 

On July 27, Moonshot says it will publish the model’s full weights, the trained parameters themselves, so that anyone can download the model and run it without paying a lab for access. Days later, Alibaba previewed its own frontier model, Qwen3.8-Max, which it says will also go open-weight, the second such escalation in a week. And the miner-to-AI trade, at its foundation, is a bet that compute stays scarce enough that model labs will keep signing decade-long leases to get it.

The stocks meant to be the purest expression of the AI capital cycle ended the month trading like the thing most exposed to it. Then, just as fast, they bounced, but not all of them, and not evenly.

The miner-to-AI-landlord trade is a leveraged bet that compute stays scarce. The market spent a month selling the whole complex, and is now spending this week sorting the bankable leases from the hopeful ones.

In this issue, subscribers get:

• • The scarcity test: why open-weight models are a genuine threat to the demand curve these 20-year leases are underwritten against, and the specific mechanism by which cheaper capability weakens a tenant’s reason to sign.
  • Contracted does not mean collected: the KBW detail buried in the selloff that tells you which of these leases are actually locked and which are more headline than substance
  • Where the screen breaks, and the second test a lease has to pass before the market will pay for it.
  • The structural shift that could eventually retire one of bitcoin’s most reliable sellers, even as it is bearish for the stocks.

Already a subscriber? Keep reading.