VanEck Research Head: Why This Isn't The AI Bubble Everyone Fears (Here's Why)

Matthew Siegel of VanEck identified a market regime shift beginning around June 1st in which companies with heavy capital expenditure went from top performers in the first five months of the year to being punished by the market. Bitcoin and crypto were categorized alongside software, and software underperformance dragged on crypto prices during this period, compounded by the four-year cycle pattern that many market participants follow.

Siegel argued that Bitcoin miners converting power infrastructure and hardware into AI data centers represent a structural opportunity because those miners had been valued at very low multiples of the megawatts they controlled relative to prevailing data center rates. The original mining business model required constant shareholder dilution to acquire ASICs faster than competitors, and mining revenues fall fifty percent every four years, creating what Siegel called a melting ice cube problem. As miners pivoted toward AI compute, their cost of capital collapsed because they could fund capital needs in debt markets rather than equity markets, and each successive data center lease was signed at better economics. VanEck made its largest single-day trade since the NODE ETF launched fifteen months ago last Thursday morning, exiting nearly ten percent of lower-volatility lower-beta exposures and doubling down on favored miners after a leveraged seller was taken out by prime brokers in the Situational Awareness fund blowout, which created forced selling with no fundamental deterioration. The NODE ETF has outperformed Bitcoin by nearly one hundred percentage points since launch.

Siegel said there has been no fundamental deterioration in the return on capital that hyperscalers expect from AI infrastructure investment, and Amazon transcripts indicate AI infrastructure returns are tracking better than earlier expectations. Companies renting older GPUs at two dollars an hour are now attempting to refresh those contracts at much higher rates, which Siegel cited as evidence of sustained demand. He compared the current AI infrastructure buildout to the nineteenth-century railroad capital expenditure boom, which consumed approximately three percent of US GDP for nearly two decades. The current AI buildout is in year three or four post-GPT by his count, though another speaker noted the cycle is in year five and is only reaching the three-percent-of-GDP threshold once, in the current year at current equity valuations.

A key distinction between the AI buildout and the railroad boom is that the four largest cloud providers, Amazon, Google, Microsoft, and Oracle, have more than two trillion dollars in contracted backlog supporting AI data center financing, with contracts including customer prepayments, customer-supplied GPUs, and durations exceeding five years. The railroad boom was government-engineered through the 1862 Railway Act, which granted hundreds of millions of acres of federal land to railways contingent on completing full networks, and railroad companies sold bonds overseas marketed as safe despite not yet holding title to the backing land. Unlike railroads, an AI data center can begin training models and serving inference immediately once connected to the grid and fiber, without requiring a synchronized global network.

On miner optionality, Siegel noted that BitDeer and Mara retain the ability to pivot back to Bitcoin mining if the Bitcoin price reaches a sufficient level, and Mara CEO Fred Teal stated that Mara could fit all its Bitcoin mining machines into one new facility purchased in South Texas. CleanSpark would need Bitcoin at three hundred sixty thousand dollars per coin to justify tearing up its recently signed lease for repurposed AI facilities. Siegel also noted there is no systemic issue to Bitcoin from hash rate leaving the network because remaining miners earn additional profits as difficulty adjusts.

After the election, VanEck reduced its L1 token exposures across the firm when many tokens doubled but showed no acceleration in real-world adoption or viral global applications. Siegel argued that corporate and consortium chains built by Circle, Stripe, Robinhood, and Wells Fargo have taken significant market share from open-source L1s because large institutional participants want predictability of fee streams and do not want to put real capital on open-source chains. When institutions do engage with open-source chains they dilute their interest by supporting three, five, or ten L1s rather than committing to one. Siegel acknowledged the underweight L1 positioning may be crowded and one-directional, creating risk of a relief rally, and said that if a clarity act passes there would be an enormous relief rally in L1 coins, though he put the odds of passage at a year-to-date low.

VanEck analysis found an inflation problem across the L1 space when comparing average inflation rates to user growth and fee generation, and ETH, Solana, and Near have all put out early-stage proposals to reduce validator inflation to near zero. Siegel cautioned that reducing staking revenues could have second-order effects on companies relying on staking income, citing Bitmine as an example. He framed the L1 category as approximately six years old, suggesting the space is still early but that structural challenges around institutional adoption and tokenomics remain unresolved.