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

Translated by: Deep Tide TechFlow

Deep Tide Insight: As the generative artificial intelligence arms race enters deeper waters, Wall Street is experiencing a cognitive shock regarding capital expenditures for computing power. In October 2024, as new GPU cloud service providers face high debt issuance costs and a tech stock correction, a wave of sell-offs targeting the AI supply chain rapidly spread to the crypto mining sector.

However, global top asset management firm VanEck pointed out in its latest research report that the market confused the balance sheet pressures of computing power leasing companies with the underlying power infrastructure landlords, resulting in severe valuation mismatches. Wall Street overlooked a fundamental physical reality: the ultimate bottleneck limiting the global AI computing power ceiling is no longer the GPU chips themselves, but the extremely scarce grid electrical capacity and high-load power access points.

For publicly listed mining companies that already hold high-voltage substations, interconnection licenses, and energized shell assets, their long-term leases have not seen any cancellations, but rather have established an irreplaceable monopoly rent premium given the background of the national grid queue delay extending to 2030. The irrational correction of up to 35% in October has provided long-term capital with an extremely rare valuation recovery window.

October Plunge of 35%: Energized Shell Providers Suffer Indiscriminate Stampede

In the recently concluded month of October, the U.S. capital market showcased a striking asset mispricing event.

Publicly listed Bitcoin mining companies that actively transformed their high-load facilities into artificial intelligence data center infrastructure faced an average cliff-like plunge of up to 35% in stock prices within a single month.

From Hut 8, Core Scientific, to TeraWulf and Cipher Mining, these once pure crypto asset mining companies have sequentially completed their strategic transformation into "Powered Shell Providers (PSPs)" in recent years.

However, amidst the wave of overall pressure on U.S. tech stocks in October, Wall Street investors did not differentiate the underlying business logic of these companies, but treated the entire sector as a homogeneous trade, conducting indiscriminate stampede sell-offs.

VanEck's digital asset research team stated in its latest report that this sharp decline is completely detached from the core business fundamentals of the companies, and the current valuations of these mining companies have entered an extreme oversold zone.

VanEck emphasized that the current market stampede is essentially a typical "dislocation," and does not reflect a true denial from the capital market of the long-term business model of mining companies transforming into AI infrastructure.

Guilty by Association: Misconduction of Neocloud Debt Panic

The core trigger of this indiscriminate stampede lies in Wall Street's contagious panic regarding the debt crisis of new GPU cloud leasing companies (Neoclouds).

Taking CoreWeave as a representative, Neocloud computing power leasing companies have generally engaged in aggressive financing through highly leveraged asset-backed debt to rapidly procure hundreds of thousands of extremely expensive Nvidia GPU chips.

Accompanied by the persistent high interest rate environment of the Federal Reserve and the unexpected rebound of the U.S. CPI consumer price index in September, AI debt yields have skyrocketed, significantly increasing the debt repayment cash flow pressure on these computing power operators.

Moreover, added to the macro-political uncertainties ahead of the U.S. midterm elections in November, the secondary market began spreading pessimistic expectations of "default by computing power lessors."

Unfortunately, Wall Street categorized the "powered shell providers" that provide space and electricity to these computing power operators as "guilty by association."

In the panic sell-off by investors, there was a miscategorization that equated GPU operators, who bear the high chip depreciation risk and debt costs, with the infrastructure landlords who merely charge stable power and facility rent.

This risk transmission involves a substantial cognitive mismatch: mining companies providing electricity and facility shells do not bear the hardware depreciation risk of chip obsolescence; their commercial role is purely that of a utility infrastructure lessor.

Leasing Fundamentals Rock Solid: No Signed Agreement Canceled

In stark contrast to the pervasive panic in the secondary market, the actual business operations of publicly listed mining companies have demonstrated strong resilience.

VanEck disclosed a compelling key empirical data point in its report: despite the pressure from soaring debt service costs facing Neoclouds, none of the long-term infrastructure leasing agreements formally signed by major mining companies have been canceled or terminated.

For instance, Core Scientific's 12-year lease agreement worth up to 500 megawatts signed with CoreWeave is still being executed as planned.

The expansion project at Lake Mariner under TeraWulf is also being delivered in an orderly fashion, continuously supplying stable clean power and space to external AI clients.

Meanwhile, Hut 8 is advancing digital infrastructure platform construction in collaboration with the well-known investment firm Coatue, and Cipher Mining is also on track to secure ultra-large power capacity as planned.

Capital expenditures by tech giants and hyperscalers in generative AI model training and inference have not slowed down; on the contrary, they are showing a continuous upward trend.

Moreover, even if individual computing power lessees encounter extreme liquidity crises in the future, there is virtually no risk of underutilization concerning the underlying facilities and electrical capacities.

In the current market landscape with extreme demand for electricity, tech giants like Microsoft, Amazon, Google, or Oracle are always ready to fully take over any premium energized assets.

The Ultimate Hard Constraint of Computing Power: AI Lacks Electricity, Not Chips

VanEck astutely pointed out the core truth of the current competition in computing power: the electrical capacity that these mining companies firmly control is the ultimate hard constraint determining the total global AI computing power.

In the early stages of the generative AI explosion, the focus of global tech giants was on Nvidia's high-performance GPU chip production capacity.

However, as the bottlenecks in the chip supply chain gradually ease, a harsher physical reality presents itself: there is no excess immediate electricity available for distribution across the U.S. grid.

Currently, the interconnection queue times for major regional transmission organizations (such as PJM, ERCOT, and MISO) across the U.S. have been extended to 4 to 7 years, with many new projects pushed back to after 2030.

Furthermore, the waiting period from ordering to delivery for core electrical equipment like large high-voltage transformers has already reached 3 to 4 years, with severe supply chain shortages.

Industry estimates indicate that by 2028, the power demand gap for U.S. data centers could reach several tens of gigawatts (GW).

In this extremely challenging landscape of electrical grid infrastructure shortages, the energized assets that Bitcoin mining companies have built over the past few years are endowed with an unprecedented "scarcity premium."

These companies already possess compliant interconnection access permits, dedicated independent substations, and real-time available energized capacities up to hundreds of megawatts.

Tech giants cannot buy time with ample U.S. capital in a short period, while the ready-to-use energized sites held by mining companies are the only shortcut across the lengthy grid interconnection queue.

Serious Valuation Inversion: Forward EBITDA Valuation Welcomes Historic Value Gap

Given the irreplaceable barrier of electric assets and robust leasing cash flow, VanEck believes that the current valuations of mining company stocks have shown serious structural inversion.

In the traditional data center sector of U.S. stocks, mature real estate investment trusts (REITs, such as Equinix and Digital Realty) boast their stable rental returns, with their enterprise value multiples (EV/EBITDA) typically stable in the high range of 20 to 25 times.

More importantly, most of these traditional data center giants currently have their mature parks operating at full load, generally lacking the capability to support the oversized new energized sites for the next generation of large model training.

In contrast, publicly listed Bitcoin mining companies holding gigawatt-level undeveloped power reserves, after experiencing a plunge of up to 35% in October, have their forward EV/EBITDA valuations roaming only between 8 to 12 times, a valuation level that's merely half of traditional data centers.

This significant valuation discount evidently fails to properly account for the long-term fixed-income cash flow locked in by the contracts of 10 to 20 years that mining companies have secured.

As the energized shells of mining companies are gradually delivered and begin contributing highly certain recurring earnings before interest, taxes, depreciation, and amortization (EBITDA), the capital market will undoubtedly initiate a value reevaluation of these companies.

VanEck concludes that the sharp decline triggered by market panic in October does not mark the collapse of the AI transformation narrative, but instead provides a historically attractive buying window for visionary long-term investors.