Bernstein: Bitcoin miners control 27 GW of planned power as AI power bottleneck tightens
The note tallies $90B+ in announced AI agreements covering 3.7 GW as post-halving economics push miners into hosting.
Bernstein estimates publicly traded Bitcoin miners control more than 27 gigawatts of planned power capacity and have announced more than $90 billion in AI-related agreements covering 3.7 GW. The research frames grid-connected electricity access as the scarce input for scaling AI data centers, a setup that could reshape how miner equities trade post-halving.
Key Takeaways
- Publicly traded Bitcoin miners control more than 27 GW of planned power capacity, based on Bernstein’s compilation.
- Announced AI-related agreements total more than $90 billion and cover 3.7 GW across hyperscalers, neocloud providers, and chipmakers.
- Bernstein pegs grid interconnection as the binding constraint for AI data centers, citing a median ~50-month wait to secure 1 GW of power across states.
- The 2024 Bitcoin halving is pressuring mining margins and accelerating miners’ push into AI hosting and high-performance computing.
Bernstein Quantifies Miners’ Power Footprint in the AI Buildout
Bernstein’s latest note puts a hard number on what has mostly traded as a narrative: power. The firm estimates publicly traded Bitcoin miners control more than 27 gigawatts (GW) of planned power capacity, alongside data center real estate that can be repurposed for high-density compute.
On the demand side, Bernstein counts more than $90 billion in announced AI-related agreements tied to miners, covering 3.7 GW with hyperscalers, neocloud providers, and chipmakers. The gap between the 27+ GW pipeline and the 3.7 GW already spoken for is the point. It implies the AI pivot is real but still early-stage, leaving room for incremental contracting or, if execution slips, a long runway of under-monetized capacity.
Bernstein also cites a RAND research brief dated April 29 projecting the US will add approximately 82 GW of additional net available power capacity by 2030. In that context, miners’ planned portfolio is large enough to matter at the margin, especially if interconnection and permitting remain slow.
Electricity, Not GPUs: The Interconnect Queue as the New Constraint
Bernstein’s core claim is that electricity access, not chips, is now the primary bottleneck for scaling AI data centers. The mechanism is the interconnect queue, the approval and engineering process required to connect a large load to the grid.
The analysts wrote: “The median waiting time to secure a GW of power is nothing less than ~50 months across states, and even in politically friendly states such as Texas, the utility is following a batch review process to navigate the interconnect queue and resource load,” framing the backlog as structural rather than cyclical.
For traders, this is the re-rating pathway. If interconnection takes years, existing grid-connected sites and credible planned power pipelines start to look like scarce infrastructure assets. That can pull miner equities away from pure BTC beta and toward an “infrastructure optionality” multiple, but only if contracts convert into delivered megawatts and recurring hosting revenue.
Post-Halving Economics Push Miners Toward Hosting and HPC
Bernstein ties the diversification push directly to post-2024 halving economics. The block subsidy cut reduced mining rewards and tightened margins, increasing the incentive to monetize power and facilities through AI data centers and high-performance computing (HPC).
Soluna Holdings is cited as an example of the revenue mix shift. The company reported a 58% increase in first-quarter revenue, driven primarily by data center hosting, with crypto mining contributing a smaller share of total sales. That is the cleanest proof point in the note that hosting can become the primary driver for some operators, not just a sidecar business.
Deal Quality, Timelines, and Power Additions Traders Should Track Next
The market is being asked to price a pipeline without full transparency on conversion. The immediate catalyst set is disclosure quality: whether miners break down the more than $90 billion in announced AI agreements by counterparty, contract structure, and start dates, and whether contracted capacity expands beyond the cited 3.7 GW.
Interconnection updates matter as much as deal headlines. Bernstein’s median ~50-month timeline makes approvals and queue progress in key markets, including Texas, a gating variable for when revenue can show up in reported numbers.
Quarterly revenue mix is the scoreboard. Soluna’s hosting-led growth sets a benchmark for miner-hosting hybrids, and future prints will show whether hosting continues to outgrow mining through the cycle.
Bernstein also flags IREN as a potential bellwether, arguing its multibillion-dollar agreements with Microsoft could fundamentally change its business model. Follow-through milestones like capacity brought online and any guidance changes tied to AI infrastructure will determine whether that thesis is trading reality or just positioning.
Trading the Miner-to-AI Narrative Without Overpaying for Announcements
I’m treating Bernstein’s framing as a market-structure story, not a tech story. If the median wait to secure 1 GW is ~50 months, then grid-connected power and credible expansion rights become the scarce input, and miners with real sites can earn an optionality premium.
The threshold that matters is conversion from “announced” to “energized.” The scale gap between 27+ GW planned and 3.7 GW contracted says the pivot is early, which cuts both ways. If contract disclosures tighten and capacity actually comes online, the setup starts to look structural rather than narrative-driven, and miner equities can decouple from pure BTC sensitivity in a way that survives the next drawdown.
