The power story around artificial intelligence is not a mirage or a psyop; it is a concrete resource problem colliding with political economy, rate design, and community trust—exactly the kind of terrain where grand narratives flourish because the underlying math is both real and uncomfortable.
The Short Version
- AI-driven data center electricity demand is large, growing quickly, and documented by credible forecasts; the debate is anchored in physical constraints, not theater.
- Near-term grid impacts—capacity additions, transmission upgrades, and rate restructuring—are already arriving, with Virginia a front-line case.
- Public pushback reflects tangible local costs and risks (prices, land, water), not evidence of coordinated psychological operations.
- Engineering options exist to moderate peaks and integrate flexible compute, but they require deliberate incentives and market design to scale.
What’s actually happening: load growth you can count
Start with the facts that survive scrutiny. Independent system planners and analysts now track data center electricity use as a material and rising share of load. The International Energy Agency estimates global data centers consumed around 415 terawatt-hours in 2024, with a trajectory toward roughly 945 TWh by 2030 if current trends continue; AI workloads are a principal driver of that increment, both in training and inference cycles. RAND’s scenario work points in the same direction on capacity, estimating that AI data centers could require about 10 gigawatts of additional power in 2025 and on the order of 68 GW by 2027—figures that, even with uncertainty bands, explain why utilities and regulators suddenly treat “hyperscale” demand as a planning category of its own. These are not marginal deltas; they are utility-scale commitments arriving on compressed timelines.
Industry-facing research echoes the operational strain: power and grid capacity have reemerged as the binding constraints on data center growth. Executives surveyed in Capgemini’s work describe demand spikes as less predictable and more extreme, complicating the traditional cadence of resource plans, interconnection queues, and transmission build-outs. The analysis frames AI as a reshaper of load profiles rather than a mere increment—difficult for capacity planners, yet potentially manageable with better forecasting and new commercial structures. That gap between physical build speed and digital growth rates is the crux: steel and copper move on decadal time, model deployments on months.
Where the costs show up: Virginia as a live-fire example
When abstractions become bills, politics sharpen. Virginia—home to the densest concentration of data centers on the planet—offers a preview of the ledger. A 2024 analysis by the state’s Joint Legislative Audit and Review Commission projected that continued data-center buildout would drive an “immense” 180% increase in electricity demand and could add as much as $37 per month to typical residential bills by 2040, absent mitigating actions. You can debate the scenario assumptions; you can’t dismiss the mechanism. High-capex generation and network upgrades to serve concentrated, all-hours load have to be financed. If rate design leaves too much of that burden on general customers, backlash is rational rather than ideological.
Regulators are responding in the language they control: tariffs and contracts. The Virginia State Corporation Commission approved a new Dominion rate structure that carves out a GS-5 class for large data centers, with 14-year service agreements and demand charges designed to recover substantial shares of transmission, distribution, and generation costs from the beneficiaries themselves. That is a significant policy signal. It acknowledges data centers as a distinct cost driver and attempts to align payment with causation—less a culture war, more cost causality embedded in a tariff.
Is public opposition a “psyop,” or is it civic due diligence?
The more expansive claim—that public anxiety about AI infrastructure is orchestrated manipulation—finds little support in the best available record. What we observe instead is broad-based concern tracking visible, proximate harms and trade-offs. A Reuters/Ipsos poll reported that roughly three-quarters of Americans worry AI will make electricity more expensive; that sentiment cut across parties, which is consistent with price salience rather than message discipline. Interviews and analyses from academic and policy observers describe community objections centered on bills, water withdrawals for cooling, land-use conflicts, and skepticism about local benefits—classic siting politics when large, lightly staffed facilities seek tax abatements and priority hookups.
None of this negates the possibility that interested parties spin the story to their advantage. Utilities underscore reliability risks to justify rate cases and gas additions; developers highlight economic development and green procurement; activists stress externalities. But spin is not synonymous with psychological operations. The weight of sourced evidence documents a real, near-term infrastructure challenge—a condition that naturally generates contested narratives—not a covert campaign with identified operators, directives, and coordination artifacts. Absent named sources, documents, or testimony demonstrating orchestration, the parsimonious explanation is the right one: people respond to costs they can see.
Engineering the problem down: flexibility, siting, and market design
There is practical room to maneuver on the technical side, provided commercial instruments evolve with it. Compute is not monolithic. Training runs and many inference workloads can be shaped—capped, shifted, or queued—to respect power targets without collapsing user experience. Ayşe Coskun’s group has demonstrated prototypes that cap data center power, shift work in time, and provision facilities as grid-responsive reserves; the point is not that every workload is flexible, but that enough is flexible enough to matter if markets pay for it. Turning that into system-level relief requires three moves.
First, rate structures that reward demand elasticity explicitly—think contracted power envelopes with price-responsive overages, or credits for verifiable load reduction at system peaks. Second, interconnection and capacity markets that let data centers procure firm supply or flexibility portfolios (storage, dispatchable generation, demand response) rather than defaulting to socialized upgrades. Third, siting that co-locates large compute with low-marginal-cost, scalable power—renewables plus storage in strong-transmission regions, hydro or nuclear where available—so marginal expansion leans clean and reliable rather than straining weak nodes. None of this is exotic; all of it is design work.
This Anti-AI hoax is not organic & is well funded.
Here's a full breakdown of what the Lunatics are trying to do. https://t.co/1gnpVxhvH6— Fran Strajnar (@Techemist) September 15, 2026
What to watch next: proof, not posture
The next phase of this story will be measurable. On the physical side: interconnection queues, actual megawatt service agreements, and the build pace of transmission and firming resources relative to announced campuses. On the financial side: tariff innovations like Virginia’s GS-5, the degree to which demand charges and long-term contracts internalize costs, and whether residential bills trend with or diverge from data center expansion. On the governance side: whether public agencies release load letters, queue positions, and rate-case exhibits that allow independent verification of claimed impacts—because sunlight, not rhetoric, resolves most of these disputes.
Two judgments follow from the evidence. First, the AI power problem is real, quantifiable, and solvable with competent engineering and aligned incentives; treating it as theater is a disservice to customers and to the grid. Second, claims of coordinated psychological operations lack substantiation in the materials that do withstand scrutiny. Until named documents or testimony surface, the cleanest read is the obvious one: a fast-growing, capital-intensive industry collided with slow-moving infrastructure, and the resulting costs, risks, and trade-offs are being contested in public. That is politics, not psyops.
Sources:
pjmedia.com, arxiv.org, finance.yahoo.com, about.bnef.com, iea.org, mdpi.com
