Bitcoin Miners Become Dominant AI Power Brokers, Reshaping Capital Flows

Twenty-seven gigawatts. That is the planned power capacity Bitcoin miners now collectively control, according to Bernstein Research [1]. Alongside that figure, Bernstein flagged nearly $90 billion in AI-related infrastructure deals these same miners have secured [2]. Jefferies has initiated buy ratings on five separate miners: Cipher, Terawulf, Hut 8, Riot, and Core Scientific [3]. Five simultaneous initiations from one institutional desk is not a momentum call. It is a structural reclassification.

This essay argues one thing: Bitcoin miners are no longer crypto operators. They are energy landlords sitting on the scarcest input in the AI economy. That repositioning has direct consequences for mining equity valuations, for structured finance desks at Tier 1 banks, and for the first wave of on-chain products backed by real-world power assets.

The Signal: What Bernstein Actually Said

Bernstein Research covers more than 1,000 stocks for institutional investors [4]. When their analysts publish a sector-level thesis, capital allocators pay attention. The thesis here is specific: Bitcoin miners have accumulated land, grid interconnects, and contracted power capacity over the past decade. That infrastructure was built to mine Bitcoin. It now has a higher-value buyer: the AI industry.

CryptoTimes reported five hours ago that Bernstein says miners are gaining leverage in the AI data center race as power shortages push cloud firms toward mining sites [1]. Crypto.news reported thirteen hours ago that Bernstein flagged nearly $90 billion in AI data center opportunity for Bitcoin miners [2]. Decrypt confirmed Bernstein remains bullish on specific names including IREN, Riot, and CleanSpark as they ride AI compute demand [5].

This is not an isolated data point. It lands on top of a pattern I have been tracking across three consecutive pieces. Leopold Aschenbrenner disclosed $13.67 billion in equity exposure through his Situational Awareness LP fund, with Bitcoin miners pivoting to AI hosting as his single largest sector bet. HIVE Digital bought a $58 million Toronto land plot and announced a 320 MW AI data center. Its stock jumped 45% on the news. Bernstein's research confirms what those individual moves suggested: the pivot is now broad and institutional, not a handful of early movers.

The Jefferies initiations matter for a separate reason. Institutional analysts do not initiate coverage on five companies in the same sector simultaneously without a structural thesis behind it. They are telling portfolio managers: the old model for valuing these stocks is broken. Build a new one.

Why Electricity Is the Real Constraint

The popular narrative around AI infrastructure focuses on chips. Nvidia's order books, TSMC's capacity, the geopolitics of semiconductor supply. That narrative is incomplete.

Chip supply has loosened. Capital is abundant. The actual bottleneck for hyperscaler expansion right now is electricity. Microsoft, Google, Amazon, and Meta need large blocks of reliable power delivered fast. Permitting a new grid connection in the United States takes years. Building a new substation takes years. Miners already did that work.

MinerWeekly reported six days ago that miners are increasingly shutting down Bitcoin mining fleets because AI infrastructure offers more stable long-duration cash flows, stronger financing conditions, and higher expected returns on power capacity [6]. That sentence contains the entire economic logic. A miner sitting on 500 megawatts of contracted grid capacity has two choices. Run Bitcoin mining hardware at thin margins in a competitive commodity market. Or lease that capacity to a hyperscaler at a fixed long-term rate with an investment-grade counterparty on the other side of the contract.

The second option wins on every financial metric: margin, duration, financing cost, and valuation multiple.

This is a classic infrastructure arbitrage. Assets built for one purpose turn out to be the critical input for a more valuable industry. The land, the grid connections, the cooling infrastructure, the power purchase agreements: none of that disappears when Bitcoin's price falls. It becomes available for redeployment. Miners who understood this early started signing AI hosting agreements. Miners who did not are watching their hardware margins compress while their land appreciates.

Canaan's Q1 2026 results illustrate the other side of this split. The world's leading maker of Bitcoin mining chips lost $88.7 million in Q1. Hardware demand is weak. But hardware is not the asset that matters anymore. Power capacity is. The divergence between mining hardware companies and mining infrastructure companies is widening. That split is the clearest signal the sector is bifurcating into two distinct businesses with two distinct valuation frameworks.

What This Does to Mining Equities

For most of Bitcoin's history, mining stocks moved with Bitcoin price. The logic was simple: higher BTC price means higher mining revenue means higher stock price. The correlation was tight enough that portfolio managers treated mining equities as leveraged BTC proxies.

That correlation is compressing. Miners with significant AI hosting revenue or signed power contracts are starting to trade more like data center REITs than like crypto assets. A REIT, for context, is a company that owns income-producing real estate and distributes most of its earnings to investors. REITs trade on yield, contract duration, and counterparty quality. Not on commodity price.

Investopedia confirmed that Bitcoin miners which entered AI have seen soaring stocks, with experts projecting further gains [3]. That outperformance is not random. It reflects a genuine change in the cash flow profile of these businesses. Long-term AI hosting contracts with hyperscaler counterparties produce predictable, bond-like revenue streams. That kind of cash flow commands a different multiple than cyclical Bitcoin mining income.

Portfolio managers running crypto-adjacent strategies need to revisit their factor models. The beta to BTC is no longer the right risk measure for this subset of miners. The right comparables are now Digital Realty, Equinix, and other data center infrastructure operators. If your model still prices Hut 8 or Core Scientific primarily as a function of BTC price, your model is stale.

The Jefferies initiations signal that institutional research is already making this shift [3]. Buy-side models will follow. When they do, the repricing of miners with substantial power assets will accelerate. The window to position ahead of consensus is narrowing.

The Structured Finance Angle: RWA Collateral

Here is where this story connects directly to the tokenization thesis.

Contracted power capacity, land, and grid interconnects are real-world assets. That term, RWA, means physical or financial assets that exist outside a blockchain but can be represented on one through tokenization. These particular assets have characteristics that structured finance desks find attractive: long-duration cash flows, investment-grade counterparties, and hard collateral in the form of land and infrastructure.

Bernstein called 2026 the start of a tokenization supercycle in January of this year [7]. Stablecoin supply is projected to grow 56% year-over-year to $420 billion, driven by capital markets integration and major fintech adoption [7]. The infrastructure for on-chain structured products exists today. What the market has lacked is a new category of high-quality RWA collateral at scale.

Miner power assets fill that gap. A 500 MW site with a ten-year AI hosting agreement signed with a hyperscaler is a financeable asset. The cash flows are predictable. The counterparty is creditworthy. The collateral is real. Goldman, JPMorgan, and BlackRock now have a legitimate entry point into miner balance sheets through power infrastructure debt. This does not require any crypto exposure. It is infrastructure lending with a familiar project finance structure.

The next step is packaging those cash flows into on-chain structured products. A structured product, in plain terms, is a financial instrument that takes cash flows from underlying assets and sells them to investors in different risk layers. Senior investors get paid first and accept lower yield. Junior investors take more risk and receive higher yield. Miner power assets are strong candidates for this structure because the contracts are long-term and the counterparties are large technology companies.

The tokenization infrastructure to execute this exists today. The legal frameworks for RWA tokenization are advancing in multiple jurisdictions. The question is not whether this product gets built. The question is which desk moves first and which miner signs the first on-chain facility.

The Bear Case and Why It Does Not Hold

Skeptics argue that the AI hosting pivot is overstated. Their case has three parts. First, hyperscalers have their own land acquisition programs and do not need miners. Second, miners lack the technical expertise to operate enterprise-grade data centers at the reliability standards AI workloads require. Third, if AI capex slows due to a demand correction or regulatory pressure, miners who converted capacity will be stranded with assets that no longer suit Bitcoin mining.

These are legitimate concerns. They are not fatal ones. The hyperscaler land acquisition programs exist but are constrained by permitting timelines and grid interconnect queues that take years to clear. Miners already cleared that queue. On technical expertise, the market has already answered: Core Scientific, Hut 8, and IREN have signed and are executing AI hosting contracts with real revenue showing up in earnings. On demand risk, the counterparties signing these contracts are Microsoft, Google, and Amazon. Long-term contracted capacity with investment-grade counterparties is not the same risk profile as speculative mining hardware. Bernstein's $90 billion figure reflects signed and near-signed deals [2], not projections.

Who Should Care and What They Should Do

If you are a portfolio manager: Your correlation assumptions between mining equities and Bitcoin price are likely stale. The miners with substantial AI hosting revenue or signed power contracts are repricing toward data center infrastructure multiples. Review your factor exposure before the repricing becomes consensus. The Jefferies initiations [3] are the leading indicator. Buy-side model updates follow initiations, not the other way around.

If you are a capital markets professional or structured finance desk: The first project finance facility explicitly tied to AI-linked power capacity, not Bitcoin mining, is the product to build or watch for. The collateral quality is there. The counterparty quality is there. The legal structure is familiar. What is missing is the first mover willing to label it infrastructure debt rather than crypto exposure. That label change is worth hundreds of basis points in financing cost for the miner and opens a new asset class for your LP base.

If you are a family office allocator: This is a new access point to the AI infrastructure buildout that does not require a venture fund or a direct hyperscaler relationship. The entry is through infrastructure debt on miner balance sheets, or through listed mining equities that are repricing toward infrastructure multiples. The risk profile is closer to a data center REIT than to a crypto fund. Size it accordingly.

What to Watch Next

A Tier 1 bank project finance announcement. Watch for Goldman, JPMorgan, or a large infrastructure lender announcing a project finance facility for a miner's power capacity that is explicitly described as AI infrastructure, not Bitcoin mining. That announcement marks the moment institutional capital formally reclassifies this sector. The language in the press release matters as much as the dollar amount.

The first on-chain structured product backed by contracted power capacity. The tokenization infrastructure exists. Bernstein called 2026 the tokenization supercycle year [7]. Miner power assets are the most compelling new RWA category to emerge this cycle. Watch which tokenization platform files first and which miner signs. Ondo Finance and similar RWA platforms are the logical execution layer.

Canaan's next earnings and the hardware-versus-infrastructure divergence. Canaan lost $88.7 million in Q1 2026 as hardware demand contracted. If power asset valuations continue rising while mining hardware margins stay compressed, the split between mining hardware companies and mining infrastructure companies will widen into a permanent sector bifurcation. That divergence, tracked across two or three more earnings cycles, will be the clearest confirmation that the sector has permanently split into two different businesses.

The question I am sitting with: when the first Tier 1 bank files a project finance facility for miner power capacity labeled as AI infrastructure, will it price tighter than comparable data center debt, or will the crypto origin of the assets still carry a spread premium?