BlackRock is telling clients to stop looking at artificial intelligence as software and start treating it as energy.
In its 2026 Global Outlook, the BlackRock Investment Institute argued that the AI buildout is pushing against physical limits and highlighted electricity as the constraint investors are underpricing.
The report’s headline-grabber is its warning that AI-driven data centers could consume as much as 24% of US electricity by 2030, a scale that would reorder everything from utility capex to industrial siting.
That kind of forecast lands with an obvious follow-on question in crypto: if grid access becomes the scarce asset, what happens to the industry that built a business model around turning cheap, interruptible power into Bitcoin?
In 2025, narratives arrived around the potential synergy of crypto and AI due to the theory that AI agents will want to use crypto for payments over traditional finance. However, a power war may tarnish this relationship going forward.
For years, mining has lived in a political argument about energy waste. The industry’s counterargument has always been operational: miners can be the flexible load, switching off when the grid is stressed and soaking up surplus generation when prices collapse.
In Texas, the Electric Reliability Council of Texas (ERCOT) has explicitly designed programs for “large flexible customers, such as Bitcoin mining facilities,” encouraging curtailment during peak demand.
But AI data centers come with a different consumption profile, different contract terms, and a different level of political support. They don’t want to power down, ever. They want the baseload.
A power problem hiding inside a tech boom
BlackRock’s broader point is that the AI boom is unusually capital-intensive. The firm cites a $5 trillion to $8 trillion range of total capital spending intentions for the AI buildout through 2030, with heavy spending on compute, data centers, and energy infrastructure.
What began as a race for chips has quickly become a race for megawatts.
There’s wide agreement that data center electricity demand is rising fast, even if analysts debate the ceiling. A Department of Energy announcement tied to the Lawrence Berkeley National Laboratory’s data center report says data center load growth in the US has tripled over the past decade.
Moreover, it is projected to double or triple by 2028. EPRI modeling from 2024 cited by Utility Dive put US data centers at 4.6% to 9.1% of US generation by 2030, depending on AI uptake and efficiency gains.
A World Resources Institute explainer, citing a Berkeley Lab study, points to 6.7% to 12% of US electricity consumption by 2030. (wri.org)
BlackRock’s “up to 25%” framing sits at the aggressive end of that spectrum, and is meant to be provocative. Yet even the lower-end scenarios would be enough to tighten power markets and harden the grid politics around who gets to plug in first.
Reuters reported that utilities and grid operators are already adjusting rate structures and rules as hyperscalers and colocation firms scramble for capacity, especially in hotspots like Texas and Northern Virginia.
That’s the environment Bitcoin miners are walking into. They are large, mobile power users, and they’re first in line in regions with abundant generation or attractive pricing. Until now, those traits looked like advantages.
Miners built on flexibility. AI runs on certainty
Bitcoin mining is brutally simple at the physics layer. Specialized computers perform hashing to secure the network, and electricity is the dominant input cost. When power is cheap relative to Bitcoin’s price and network difficulty, miners print cash. When power is expensive, they shut down, relocate, or go bankrupt.
That operational flexibility has become the industry’s best talking point as public scrutiny has increased. The US Energy Information Administration estimated crypto mining likely represented about 0.6% to 2.3% of electricity consumption in the US in 2024, a small share in percentage terms but large enough to show up in local politics and grid planning.
Texas is the cleanest case study because the state’s competitive power market turns that flexibility into revenue. In a 2023 SEC filing, Riot Platforms said it curtailed power usage by more than 95% during periods of peak demand in August 2023, choosing to forego mining revenue to support ERCOT reliability.
CryptoSlate reported that ERCOT paid a miner $31.7 million in energy credits that month to power down during a heat wave, a detail that captures both the value of flexibility and why the politics can get ugly fast.
Now put that model next to AI. Training and serving large models need constant power and tight uptime. A hyperscaler signing a long-term lease wants predictable delivery, not voluntary curtailment.
If miners are the shock absorber, then AI is the shock creator.
And BlackRock’s yearly outlook effectively says that the shock is coming and there’s no stopping it.
Grid constraints make cheap power a moving target
In the mining playbook, “cheap power” means stranded hydro, surplus wind at night, or a friendly industrial tariff. But as data centers scale, cheap power becomes a moving target, because grid access itself becomes the bottleneck.
Interconnection queues and transmission delays are the new friction. Even when a region has generation, it may not have the wires, the transformers, or the permitting pathway to deliver it to a new 500-megawatt campus.
NERC has warned about reliability threats from rapid load growth tied to AI, data centers, EVs, and electrification colliding with generator retirements and slow buildouts. (Financial Times)
That matters for miners because their advantage is speed.
They can drop containers on a site, energize, and start hashing faster than a conventional industrial plant can ramp. But if the gating item becomes substation capacity and interconnection approval, then that speed turns into a regulatory contest.
The political optics are shifting, too
When power markets tighten, lawmakers start looking for villains. Mining has often been convenient because it feels optional, even to people who understand nothing about it. In contrast, AI is now being both to the public and to lawmakers as national competitiveness.
That asymmetry is what will shape policy. It’s easier to impose reporting requirements or additional tariffs on miners than on the data centers the local chamber of commerce is courting. It’s also easier to frame mining as a speculative luxury and frame AI as the backbone of defense, productivity, and medicine.
If BlackRock is right that AI’s energy footprint will become a macro risk, the political coalition supporting grid investment may widen, but so may the pressure to prioritize “productive” loads.
Miners might respond by leaning harder into the flexibility story. A Duke University report cited by Utility Dive argues the existing US grid can handle significant new load if it can be curtailed during stress events, and mining can do that. Many AI workloads, especially inference for consumer products, generally can’t.
That creates a potential wedge: miners as a controllable load that helps integrate renewables, versus data centers as an inflexible load. This argument is already brewing in policy circles and utility commission hearings.
However, whether it wins will depend on local economics and lobbying, not internet debates.
The hedge: turning mining sites into AI sites
There’s another adaptation path already underway: pivoting from hashing to hosting.
The logic is straightforward. If you already own land, power rights, and a substation, you have what AI developers need most. And if your legacy business is volatile, the prospect of contracted cash flows from compute hosting is tempting.
CryptoSlate reported in October that some firms originally focused on Bitcoin mining have been pivoting toward AI infrastructure, with deals tied to cloud and AI workloads, precisely because power access in places like Texas has become valuable. The article’s message is not that every miner will become an AI landlord, but that the industry’s prime asset is shifting from machines to megawatts.
This pivot is harder than it sounds. AI data centers require different cooling, different network architecture, and different uptime guarantees. Mining can tolerate interruptions, but many AI customers won’t.
The cost of retrofitting can be enormous, and the competition includes specialist data center operators with deep relationships and financing advantages.
Yet the direction of travel is clear. When power becomes scarce, the highest-value use of a megawatt tends to win.
Where Bitcoin mining lands
BlackRock’s forecast isn’t about Bitcoin specifically, but about the end of cheap abundance. If AI pushes the US toward a world where electricity demand grows fast, and transmission remains slow, any business built on marginal power economics gets squeezed.
Of course, miners won’t disappear. Bitcoin’s incentive structure is designed to keep hash power online somewhere, and the industry’s mobility means it can chase new energy pockets. But the center of gravity could shift.
Regions with surplus generation and friendly policy will likely see miners as a stabilizing industrial load, especially if they can credibly offer curtailment. Regions courting hyperscalers will surely treat miners as a second priority.
The likely outcome is a barbell.
On one side: miners that integrate with grids, sign structured demand-response agreements, and become part of utility planning.
On the other: miners that turn their energy positions into broader compute infrastructure, essentially arbitraging their early arrival in power markets into a new line of business.
Either way, the easy era is ending. BlackRock’s warning that AI data centers could swell to an enormous share of US power demand is a reminder that the next phase of digital infrastructure won’t be constrained by code, but by the messy physical world of wires, permits, turbines, and heat.
