Matthew Sigel, VanEck’s head of digital assets research and manager of the VanEck Onchain Economy ETF (NODE), said the current AI infrastructure buildout is not the bubble many fear. He also argued that the drag on crypto is coming less from macro conditions and more from institutional disappointment with major Layer 1 networks. Sigel added that NODE has outperformed Bitcoin by nearly 100 percentage points over the past 15 months, driven mainly by an early bet on Bitcoin miners pivoting toward AI data center exposure.
Sigel made the comments on The Rollup’s podcast AI Super Cycle, hosted by Rob, also identified as Robbie Klages, and Andy. The episode aired on Aug. 10, 2026. He also disclosed that his views may align with holdings inside VanEck-managed funds.
Market leadership flipped after June 1
Asked about recent market conditions, Sigel said the first five months of the year were defined by extreme concentration. The companies spending the most on capital expenditures posted the strongest stock performance. That relationship reversed after June 1.
“This year’s first five months, the companies that spent the most went up the most; but starting in June, the companies that spent the most were punished the hardest,” he said.
Sigel said Bitcoin and crypto tokens have effectively been grouped into the software bucket by the market. That matters because software has been under pressure relative to semiconductors. “Bitcoin is software. It happens to be open-source software. When software stocks as a group are under pressure, it is very hard for Bitcoin and crypto tokens to remain untouched,” he said.
He pointed to AI systems such as Claude and Codex as forces already affecting open-source software projects in a practical way. In his telling, the upgrade cycle in open-source software is not as top-down as it is in Web2, because users cannot simply be forced onto a new version. That weakness in software has fed into Bitcoin and the wider token market.
Sigel also said the market still carries a strong psychological attachment to the four-year halving cycle. He would prefer to see greater dispersion in returns so investors can find alpha in individual assets rather than in a single dominant factor trade. He said he does not think the relationship has to be framed as AI up and crypto down, or the reverse.
On his own positioning, Sigel said he still has AI infrastructure exposure through Bitcoin miners. He is constructive on a fourth-quarter bottom for Bitcoin, though he would rather wait for positive developments supported by trading volume than try to trade every move in a choppy tape.
Why NODE leaned into the miner-to-AI conversion
When asked about holdings such as MARA, Riot, APLD, and WULF, Sigel said NODE’s edge came from identifying that the market was severely undervaluing the megawatts controlled by Bitcoin miners relative to the valuation multiples assigned to a small number of data center REITs at the time.
He framed the old mining model as structurally weak. Miners used to issue stock, buy ASICs, and try to grow faster than competitors, while revenue was cut in half every four years. “ASICs are not the most valuable asset miners have. Power and land are,” he said. He also described the legacy mining model this way: “The early miner business model was to keep issuing shares to buy machines, while revenue halves every four years. That is a melting ice cube.”
Sigel said that financing conditions have changed. Not only hyperscalers, but also the leases around data center buildouts, have seen funding costs fall sharply. That means these companies can tap debt markets instead of repeatedly issuing stock and diluting shareholders. In his account, each new lease has tended to come with better economics than the last, and lower rates have added more value.
That backdrop shaped VanEck’s response when market leadership turned in June. Sigel said the style reversal hurt levered investors the most, while NODE itself did not use leverage and tried to stay underweight companies carrying heavy leverage. When Situational Awareness, a fund with significant overlap to NODE’s holdings, was liquidated by its prime broker, VanEck added risk instead of pulling back.
Sigel said that on the opening last Thursday, NODE executed its largest single-day trade since the ETF launched. The fund exited nearly 10% of its low-volatility, low-beta positions and doubled down on several miners it viewed most favorably.
He said that decision was grounded in fundamentals, not just price action. Based on earnings calls from companies such as Amazon, Sigel said returns on hyperscaler AI investment look better than early expectations. He gave one example: older GPUs that had been rented out at $2 per hour are now coming off contract, and customers want to renew at sharply higher rates.
At the lows, Sigel said some of his highest-conviction names were valued only on the basis of existing leases, with no credit being given to terminal value, platform value, or potential future leases. Even under a scenario where the 10-year yield rose another 100 basis points, he said that work had already been done, leaving what he viewed as a large margin of safety.
On whether a migration of computing resources from Bitcoin mining to AI could create a systemic problem for the Bitcoin network, Sigel said no. He argued the opposite: if hashrate falls, the miners that remain earn more. He added that companies such as Bit Deer and MARA still have valuable flexibility. They can continue mining or convert facilities to AI use.
That flexibility itself carries value, he said. In CleanSpark’s case, Sigel estimated Bitcoin would need to reach $360,000 before it made economic sense to tear up newly signed AI leases and return to a pure-mining model.
“CleanSpark would need Bitcoin at $360,000 to go back to a pure mining model, because only then would it be worth ripping up signed AI leases,” he said.
Why Sigel rejects the railroad-bubble analogy
Sigel said many critics respond to the AI buildout by comparing it with the 19th-century railroad bubble, arguing that both involve economy-changing capex cycles that destroy early capital. He disagrees with that analogy on both scale and funding structure.
On scale, he said the US devoted 3% of GDP to railroads for nearly 20 straight years. AI, by contrast, is only in about its fifth year if counted from the arrival of GPT, and this year is only the first in which spending has reached 3% of GDP.
“The US poured 3% of GDP into railroads for 20 straight years, while AI has only just reached that level for one year so far,” he said.
Sigel argued that current market valuations for AI companies do not reflect a scenario in which this buildout continues for two decades. He said most analysts still think the cycle peaks by 2030, which he sees as a mismatch.
He put more emphasis on the funding side. In his telling, the railroad bubble was government-led. Congress passed the railroad legislation in 1862, granting hundreds of millions of acres of federal land to railroad companies, but ownership would not transfer until the network was completed. The Treasury also issued construction bonds that were junior to private capital.
“The railroad bubble was government-led: Congress passed legislation in 1862, land was promised first and ownership transferred only after the network was built, and Treasury construction bonds were subordinate to private capital,” he said.
Sigel said major railroad companies then sold bonds abroad as safe assets, even though the bonds depended on land sales and the companies did not yet hold title to that land. That, in his view, was bubble financing.
AI factories are different, he said, because they become productive as soon as power, fiber, and chips are in place. They can begin training models and running inference immediately. Railroads, by contrast, had limited utility until routes connected across the country.
“An AI factory can train models and run inference as soon as it is connected to the grid, fiber is installed, and chips are in place. A railroad had to be connected coast to coast before it was truly useful,” he said.
Sigel’s final distinction centered on backlog. He said the four largest cloud providers now hold more than $2 trillion in signed contract backlog, with Microsoft and Oracle making up roughly half. Many of those contracts include customer prepayments, customer-supplied GPUs, and terms longer than five years.
“The four largest cloud providers have more than $2 trillion of contract backlog, and many of those deals include customer prepayments, customer-supplied GPUs, and terms longer than five years. No one in 1870 was buying train tickets for 1885,” he said.
That means current data center financing is supported by real purchase commitments rather than by government subsidies or land speculation, according to Sigel.
VanEck cut L1 exposure after the election and shifted attention to enterprise chains
Sigel said VanEck reduced exposure to mainstream L1s after the election, even though many tokens doubled. The reason, he said, was that adoption did not accelerate in a meaningful way, no truly breakout application appeared, and no application pulled in global capital at the scale investors had hoped for.
“A lot of tokens doubled after the election, but there was no real breakout app and no application that pulled global capital into the system,” he said.
What followed, in his view, was the rise of enterprise chains. He cited Circle, Stripe, and Robinhood as examples of semi-permissioned chains that let public companies customize user experience and capture part of the economics. He also said Wells Fargo is building its own custom chain.
“The winners right now are enterprise chains: Circle has one, Stripe has one, Robinhood has one, and even Wells Fargo is building its own custom chain,” he said.
Sigel argued that this may not satisfy open-source purists, but the institutions that want to use these networks at scale need predictable fee flows. He mentioned ETH specifically as an example where fee volatility has been too high for many large institutional participants. Solana’s fee volatility has been less severe, he said, but until recently investors in New York could not buy USDC on Solana.
His broader point was that banks and other regulated entities do not want to place significant value directly on open networks. Even if they do engage, they often have to support three to five chains at the same time, which dilutes the winner-take-all case for any single L1.
“Banks and other regulated institutions do not want to put real money on open chains. Even if they participate, they have to support three to five chains at once, which dilutes the winner-take-all nature of any single L1,” he said.
That is why VanEck has remained very underweight these tokens, he said. Market share is leaking away, and the winner-take-all profile is weakening.
CLARITY Act could trigger a relief rally, but Sigel remains cautious
Sigel said the CLARITY Act remains an important swing factor. If the bill passes, he expects some tokens to stage a large relief rally. Until then, he said, VanEck will remain cautious.
“If the CLARITY Act passes and creates a real disclosure regime, some tokens could see a huge relief rally. Until then, we stay cautious,” he said.
He said the value of the bill would be in creating an information-disclosure framework. That would let investors identify the true beneficial owners behind tokens, see how much is held by labs and how much by foundations, and avoid a market where key opinion leaders promote assets without disclosing positions. Sigel said the lack of disclosure has been one reason many institutions have chosen to ignore the sector.
He added, however, that the probability of the bill passing this year has fallen to its lowest level of the year, which is why it has not changed his current stance.
Lower token inflation is constructive, but second-order effects matter
On proposals by ETH, Solana, and Near to reduce validator inflation, Sigel said the direction makes sense. A few quarters ago, he said, VanEck studied the relationship between average L1 inflation rates, user growth, and fee revenue across major networks. The conclusion was that inflation is a real issue in the sector.
He drew another comparison with software companies. This year, semiconductors have materially outperformed software, which has forced many software firms to rethink how quickly they issue new shares. Some have already reduced dilution meaningfully.
That, Sigel said, is why it is reasonable for L1s to reconsider inflation six years after launch. But he also warned that execution details will determine outcomes, and that there could be second-order consequences for companies that rely on staking income.
He cited Bitmine as one example, saying the company often highlights staking yield in its messaging. If lower inflation undermines that income stream, part of that business logic could break down.
Overall, Sigel described the move to revisit inflation policy as a sensible step, but not one that can be judged only at the token level.