Add preferred source
Shares of bitcoin miners that have shifted into AI data center infrastructure fell an average of 35% in October, driven by rising Treasury yields, hot inflation data and credit worries centered on GPU rental firms such as CoreWeave, whose debt yields approached 15%. VanEck analysts Griffin MacMaster and Matthew Sigel argue the selloff is a dislocation rather than a rejection of the business model, noting that no signed leases were canceled. They contend that power, not chips, is the binding constraint on AI compute, citing a projected 38-gigawatt US data center shortfall through 2028 and a grid queue in which only about 2% of capacity reached commercial operation in 2025. The firm says the miners trade at roughly half the 2028 EV/EBITDA multiple of traditional data center REITs, though it flags $600 million to $3.2 billion in potential equity needs in 2027-2028 and continuing permitting risk.
Key Elements
Bitcoin miners that have repositioned themselves as artificial intelligence data center landlords lost an average of 35% of their market value in October, a slide that VanEck contends has more to do with credit-market contagion than with the underlying economics of the business.
In a research note published November 4, VanEck analysts Griffin MacMaster and Matthew Sigel described the drop as a “dislocation” rather than a verdict on the AI pivot, arguing that investors have lumped two very different kinds of companies into a single trade. The firm tracks a group that includes Cipher Mining, Hut 8, TeraWulf, Core Scientific and Riot Platforms.
The immediate trigger was not customer demand
The immediate trigger was not customer demand but financing conditions. Treasury yields climbed, inflation readings came in hotter than expected in September, and credit markets grew uneasy about the debt being raised across the AI sector ahead of the November midterm elections. CoreWeave’s debt yields approached 15%, against bond coupons of roughly 10% or less, a spread that raised alarms about how the next wave of computing capacity would be funded.
That anxiety bled into miners-turned-landlords, which VanEck calls powered shell providers. These companies supply electricity connections, land and data center buildings under leases that typically run 10 to 20 years. GPU rental firms, by contrast, buy costly processors and rent out computing capacity on contracts of roughly two to five years. The report notes that not one signed lease among the publicly traded companies it examined was canceled during the selloff.
“The power these companies control appears to be the binding constraint on AI compute,” the analysts wrote, adding that at current levels the shares “appear significantly dislocated from fundamentals given their long-term contracted cash flows.”
The supply picture supports that view
The supply picture supports that view. Morgan Stanley projects a 38-gigawatt shortfall in US data center power through 2028, and only about 2% of the 2,290 gigawatts sitting in the US electricity interconnection queue reached commercial operation in 2025. Grid backlogs effectively put a scarcity premium on capacity that miners already have energized. TeraWulf, for one, has an agreement tied to AI infrastructure worth roughly $19 billion.
| VanEck’s comparison of two AI infrastructure models | Powered shell providers (miners) | GPU rental firms (neoclouds) |
|---|---|---|
| What they supply | Power connections, land, buildings | Processors and computing capacity |
| Typical contract length | 10-20 years | 2-5 years |
| Main cost pressure | Construction and project financing | Chip purchases and debt service |
| Key constraint cited | Grid interconnection and power supply | Cost of capital for hardware |
Note: Contract terms and cost pressures as described in the VanEck research note.
Valuation is where the argument sharpens. VanEck estimates the miners it tracks change hands at about half the enterprise-value-to-EBITDA multiple projected for traditional data center REITs in 2028, implying that investors are paying little for infrastructure already in the ground and cash flows already under contract.
The bull case does not erase the risks. VanEck estimates that additional AI projects could require $600 million to $3.2 billion of equity financing in 2027 and 2028, a range that could dilute existing holders. Local opposition to data center construction is another live threat; earlier restrictions on TeraWulf’s expansion showed how permitting can stall projects even when demand is strong. Companies that both own GPUs and lease out facilities may also carry heavier financing exposure than those that stick to powered shells.
For investors, the debate turns on whether October’s rout was a repricing of AI infrastructure risk broadly or a mispricing of one specific asset class: grid-connected power. The miners’ shares had earlier rallied hard on enthusiasm for the shift away from pure Bitcoin mining, and that enthusiasm has now reversed. What has not reversed, per the report, is the contracted revenue.
Bitcoin itself was trading near $80,310 at the time of the note.
