What AI lacks is power: Bitcoin miners with power assets were indiscriminately sold off 35% in October

Authors: Matthew Sigel, Griffin MacMaster (VanEck Digital Assets Research Team)

Translated by TechFlow

TechFlow Editor's Note:As the generative AI arms race enters deeper waters, Wall Street is experiencing a cognitive shock over compute capital expenditure. In October 2024, as new GPU cloud providers faced high debt-issuance costs and a pullback in tech stocks, a sell-off across the AI supply chain quickly spread to the crypto mining sector.

However, global top asset manager VanEck stated clearly in its latest research report that the market conflated the balance-sheet pressure of compute lessors with that of underlying power-infrastructure landowners, creating a severe valuation mismatch. Wall Street overlooked a fundamental physical reality: the ultimate bottleneck determining the global ceiling for AI compute is no longer GPU chips themselves, but extremely scarce grid energization capacity and high-load power connection points.

For listed miners that already hold high-voltage substations, grid interconnection permits and powered-shell assets, their long-term leases have not been cancelled; instead, amid U.S. grid queues stretching into 2030, they have established irreplaceable monopoly rent premiums. The irrational 35% drawdown in October has given long-term capital an exceptionally rare valuation-repair window.

35% October plunge: powered-shell providers suffer indiscriminate stampede

In the just-ended October, the U.S. equity market saw a striking indiscriminate sell-off.

Publicly listed Bitcoin miners actively converting their high-load sites into AI data-center infrastructure suffered cliff-like declines averaging 35% in a single month.

From Hut 8 and Core Scientific to TeraWulf and Cipher Mining, these former pure crypto-mining companies have in recent years completed a strategic pivot to powered shell providers (PSPs).

However, amid the broader pressure on U.S. tech stocks in October, Wall Street investors did not differentiate the underlying business logic of these companies; instead, they treated the entire sector as a single homogeneous trade and sold it off indiscriminately.

The digital assets research team at globally recognized asset manager VanEck said bluntly in its latest report that this sharp decline was completely disconnected from the companies' core business fundamentals, and that valuations for these miners have now entered an extremely oversold zone.

VanEck emphasized that the current market stampede is essentially a classic pricing dislocation, not a genuine rejection by capital markets of miners' long-term business model of pivoting to AI infrastructure.

Guilt by Association: Misdirected Spillover of Neocloud Debt Panic

The core trigger of this indiscriminate sell-off was Wall Street's contagion fear over the debt crisis of new GPU cloud lessors (Neoclouds).

Neocloud compute-leasing companies such as CoreWeave have generally raised aggressive financing through highly leveraged asset-backed debt to secure hundreds of thousands of extremely expensive Nvidia GPU chips in a short period.

As the Federal Reserve's high-rate environment persisted and September U.S. CPI rebounded above expectations, AI debt yields surged sharply, significantly increasing debt-service cash-flow pressure for these compute operators.

Coupled with macro-political uncertainty ahead of the November U.S. midterm elections, pessimistic expectations of compute lessors defaulting began spreading in the secondary market.

Unfortunately, Wall Street treated powered-shell providers supplying these compute operators with space and power as 'guilty by association'.

Amid panic selling by investors, people wrongly conflated GPU operators bearing high chip depreciation risk and debt costs with infrastructure landowners that simply collect stable power and data-center rental income.

This risk contagion involves a major logical misperception: miners providing power and data-center shells do not bear hardware impairment risk from chip obsolescence; their business role is purely that of a utilities infrastructure lessor.

Leasing fundamentals remain rock solid: no signed agreements have been cancelled

In stark contrast to the panic pervading the secondary market, the actual business operations of listed miners have shown exceptional resilience.

VanEck disclosed a highly compelling key empirical data point in its research report: despite Neoclouds facing sharply rising debt service costs, none of the long-term infrastructure lease agreements formally signed by major mining companies has been canceled or terminated.

Taking Core Scientific as an example, its massive 12-year, 500 MW colocation lease agreement with CoreWeave continues to be performed steadily as planned.

The expansion at TeraWulf's Lake Mariner campus is also being delivered in an orderly manner, continuously supplying stable clean power and data center space to external AI clients.

Meanwhile, Hut 8 is advancing a digital infrastructure platform buildout together with prominent investment firm Coatue, while Cipher Mining and IREN are locking in hyperscale power capacity as planned.

Capital expenditure by technology giants and hyperscalers on generative AI model training and inference has not slowed; instead, it continues to rise.

Even if individual compute lessees face extreme liquidity crises in the future, the underlying data centers and power capacity carry virtually no risk of sitting idle.

In the current market environment of acute power demand, top technology giants such as Microsoft, Amazon, Google, or Oracle stand ready to fully take over any high-quality energized assets at any time.

Ultimate hard constraint on compute: AI is short of electricity, not chips.

VanEck sharply identified the core truth of the current compute competition in its research report: the power capacity firmly controlled by these mining companies is the ultimate binding constraint (Binding Constraint) determining total global AI compute capacity.

In the early stages of the generative AI boom, the focus of competition among global technology giants was NVIDIA's high-performance GPU chip capacity.

However, as chip supply chain bottlenecks gradually ease, a harsher physical barrier confronts everyone: the U.S. grid has no surplus immediately available power for allocation.

Currently, grid interconnection queue times at major U.S. regional transmission organizations, such as PJM, ERCOT, and MISO, have stretched to 4 to 7 years, and many new projects have even been delayed beyond 2030.

Moreover, lead times for core power equipment such as large high-voltage transformers from order placement to delivery have lengthened to 3 to 4 years, reflecting extremely severe supply chain shortages.

Industry estimates show that by 2028, the U.S. data center power demand shortfall will reach tens of gigawatts (GW).

Precisely under this severely strained grid infrastructure shortage, energized assets built by bitcoin miners over the past several years have been assigned an unprecedented 'Scarcity Premium'.

These companies already hold compliant grid interconnection permits, dedicated independent substations, and hundreds of megawatts of immediately available energized capacity.

Technology giants cannot buy time in a short period with abundant dollar capital, while the ready-to-use energized sites held by miners are the only shortcut to bypass lengthy grid queues.

Severe valuation inversion: forward EBITDA valuations present a historic value trough.

Based on irreplaceable power asset barriers and stable lease cash flows, VanEck believes that current valuations of mining company stocks have experienced a severe structural inversion.

In the U.S. traditional data center sector, mature real estate investment trusts (REITs, such as Equinix and Digital Realty), benefiting from stable rental returns, have maintained enterprise value multiples (EV/EBITDA) persistently in a high range of 20x to 25x.

More importantly, most mature campuses of these traditional data center giants are currently running at full capacity and generally lack the massive incremental energized sites needed to support next-generation large model training.

By contrast, publicly listed bitcoin miners holding gigawatt-scale undeveloped power reserves, after an October drawdown of up to 35%, are trading at forward EV/EBITDA valuations of only 8x to 12x, roughly half the level of traditional data centers.

This significant valuation discount clearly does not properly incorporate the fixed-income-like long-term contracted cash flows already locked in by mining companies over 10- to 20-year terms.

As mining companies gradually deliver their energized shells and begin contributing highly predictable recurring EBITDA, capital markets will inevitably initiate a revaluation of these companies.

VanEck concluded that the sharp October selloff driven by market panic is by no means the demise of the AI transformation narrative; instead, it offers a highly attractive historic buying window for forward-looking long-term investors.

Source: TechFlow