On 6 July 2026, TeraWulf — a company that began as a pure bitcoin miner — signed a 20-year lease with Anthropic to host AI computing at a Kentucky data center. The deal is expected to generate roughly $19 billion in contracted revenue, and The Block reported the same figures. TeraWulf shares jumped as much as 19% on the news.
The striking detail: $19 billion in future revenue is larger than TeraWulf's entire market value of about $12 billion. This article explains what the deal actually involves, why bitcoin miners are converting into AI landlords, and what the trend signals about the economics of both industries.
What the deal actually is
The lease is a real-estate and power arrangement, not a mining or crypto transaction. The concrete terms:
- Facility: TeraWulf's Justified Data campus in Hawesville, Kentucky.
- Capacity: about 401 megawatts of computing power, built out in phases.
- Timeline: first power expected in the second half of 2027, full site running by early 2028.
- Term: 20 years, backed by what TeraWulf described as investment-grade credit.
- Revenue: roughly $19 billion over the initial term.
Anthropic is the tenant. TeraWulf is the landlord, providing power, cooling, and physical infrastructure for AI workloads. The AI company runs its own compute; TeraWulf supplies the building and the megawatts.
Alongside the lease, TeraWulf announced it will sell its entire 50.1% stake in the Abernathy data-center joint venture in Texas to a group led by its partner Fluidstack for about $530 million — monetizing roughly $450 million of invested capital and freeing cash to expand facilities it owns outright.
Why miners are becoming AI landlords
TeraWulf's pivot is not idiosyncratic. It is the clearest recent example of an industry-wide rotation, and the economics explain why.
Mining margins tightened after the halving. Bitcoin mining earns newly issued coins by running warehouses of specialized computers. Roughly every four years, a halving cuts the block reward in half. Last year's halving cut mining revenue per block by 50%, squeezing margins across the sector.
AI hosting offers steadier income. A single creditworthy tenant on a 20-year lease provides predictable cash flow. Mining revenue swings with bitcoin's price and network difficulty; a lease payment does not.
Miners already own the scarce input: power. The hardest part of building an AI data center is securing large, reliable electricity capacity and the sites to house it. Bitcoin miners spent years acquiring exactly that. Repointing 401 megawatts from mining rigs to AI servers reuses the asset that took longest to assemble.
The scarce resource in the AI buildout is power and the sites that carry it. Bitcoin miners spent a decade accumulating both for a different purpose.
Think of a company that built a fleet of trucks to haul one commodity, then discovered a second commodity paying far more per mile on long-term contracts. The trucks and depots — the hard part — are already there. Only the cargo changes.
The scale signal
The single most revealing number is that the lease is worth more than the company.
| Metric | Value |
|---|---|
| Contracted lease revenue (20 years) | ~$19 billion |
| TeraWulf market value | ~$12 billion |
| Share price move on announcement | up to +19%, settled ~+4% |
| Stock performance year-to-date | up more than 80% |
When a single contract exceeds the entire market capitalization of the company signing it, the market is repricing what the business fundamentally is. TeraWulf still runs a bitcoin operation, but the Anthropic lease and its wider pipeline may now define its value. It is being valued less as a miner and more as an AI-infrastructure landlord.
The broader rotation backs this up. As of March 2026, bitcoin miners had sold more than 15,000 coins from peak holdings and signed over $70 billion in AI computing contracts — moving capital from volatile mining toward the steadier margins of the AI trade.
The counter-signal worth noting
The same day told a two-sided story. TeraWulf's stock jumped on the AI lease, but it "stood out against a soft day for bitcoin itself." The token slipped toward $61,900 after Strategy disclosed selling 3,588 bitcoin for about $216 million — a sharp escalation from the 32 coins it had sold weeks earlier.
The juxtaposition is the real signal. Capital is rotating out of crypto exposure and into AI infrastructure, and companies straddling both are being rewarded for the AI side. That is good for miners with power and sites to lease — and a reminder that the enthusiasm is for AI compute demand, not for mining itself.
What this means for builders
Three implications for teams working across crypto and AI infrastructure.
AI compute demand is now a macro force in crypto markets. Miner valuations, power markets, and even bitcoin's price action are increasingly entangled with AI data-center demand. If you model crypto-mining economics, AI hosting contracts are now a variable you cannot ignore.
Long-dated, single-tenant leases change the risk profile. A 20-year lease to one AI tenant trades mining's price volatility for concentration risk — the counterparty's health and the durability of AI compute demand now matter more than hash rate. Evaluate these companies as landlords with tenant risk, not as miners.
Power and interconnection are the real moat. The lesson beneath the deal is that megawatts and grid connections are the scarce, slow-to-build asset. Whoever controls large blocks of reliable power holds leverage in the AI buildout, regardless of what those sites were originally built for.
Conclusion
TeraWulf's $19 billion Anthropic lease is a clean illustration of a broader shift: bitcoin miners, squeezed by post-halving economics, are converting the power capacity and sites they built for mining into AI data-center infrastructure with steadier, contracted returns.
When a single 20-year lease is worth more than the entire company, the market has already decided the story. The scarce resource of the AI era is power, and the companies that spent the last decade accumulating it for bitcoin are now leasing it to the labs training the next generation of models.