Audited load disclosure.
Publish maximum firm load, expected average MW, annual MWh, ramp schedule, expansion rights, and the identity of the ultimate compute customer.
Every chatbot query requires a heavy industrial footprint. The industry is currently scrambling to secure gigawatt-scale power allocations, grid upgrades, and natural gas access, leaving communities to absorb the cost. This advocacy toolkit converts raw technical infrastructure data into direct, actionable consumer protection rhetoric.
AI is not a cloud; it is a demand for power, water, land, subsidies, and public risk. If a data-center plan cannot name its load, power source, emissions, bill impact, expansion rights, and exit plan, it is not progress. It is a black box plugged into the public wallet.
That is the rhetoric shift: stop debating whether computers are useful and start asking who gets the upside, who pays for the wires, who breathes the exhaust, who loses the water, and who is left holding the shell if the boom goes bust.
The old mental model says “software company.” The disclosed infrastructure model says “industrial electricity customer.” Below: a transparent OSINT estimate of average frontier AI loads (including Meta and CoreWeave) through 2026 Q2, followed by the announced capacity ladder that communities are being asked to absorb.
Read carefully: these bars are not automatically additive. They mix data-centre capacity, accelerator-system capacity, cloud commitments, and on-site generation. That is precisely the point: public agencies need binding load, emissions, water, and ratepayer disclosures before approvals.
Model stance: the historical line is a guesstimate of average electrical load from public signals. The capacity ladder is a claims inventory. Use the graph to force disclosure, not to pretend private labs have published audited power bills.
Quick number: the legacy sourcebook’s 2026 Q2 base case totals — GW across OpenAI, Anthropic, xAI, Meta, and CoreWeave.
Regulatory translation: one gigawatt of continuous load consumes about 8.76 TWh per year before data-centre overhead, cooling, and grid losses are debated.
A peer-reviewed 2026 study projects data-center and crypto growth will raise U.S. electricity costs 6% to 29% on average — and up to 57% in hard-hit regions like Virginia — by 2030.
Environmental Research Letters 2026The Federal Reserve Bank of Dallas finds existing data centers have already raised wholesale electricity prices 2% to 6%, with high-utilization buildout capable of pushing them ~50% higher by 2028.
Dallas Fed · Kay et al.PJM's independent market monitor blamed data centers for $6.5B — about 40% — of one record $16.4B capacity auction, and $21.3B across three. Those costs flow to roughly 67 million ratepayers.
Monitoring Analytics · PJMIn March 2026 Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI signed a White House pledge to cover all power-delivery upgrades and build, bring, or buy new generation so costs aren't passed to households.
White House 2026Gartner predicts 40% of existing AI data centers will be operationally constrained by power availability by 2027 — which is exactly why developers reach for on-site fossil generation.
GartnerTo skip multi-year grid queues, developers install behind-the-meter reciprocating gas engines and turbines that bypass clean-grid planning and threaten local air quality.
Enverus EIREntergy will build more than 7 GW of new natural gas across 10 plants for Meta's Louisiana "Hyperion" campus — over a 30% jump in the state's grid — alongside up to 6.6 GW of nuclear by 2035.
Entergy · MetaThe U.S. EIA forecasts residential electricity prices up about 5% in 2026 to 18.2¢/kWh, with the largest increases on the East Coast — where data-center load is concentrated.
US EIA · May 2026Ohio now makes big data centers pay for 85% of reserved power for 12 years; Virginia created a dedicated data-center rate class; Georgia shifts upstream costs onto >100 MW users — all to stop the cost-shift onto you.
PUCO · VA SCC · GA PSCThe scientific warning is simple: warming follows cumulative CO₂, and infrastructure choices can lock emissions in for decades. A data centre that creates new fossil demand is not just a warehouse. It is a claim on the atmosphere, the grid, local air, public water, and public money.
Every tonne matters. IPCC’s TCRE framework treats warming as roughly proportional to cumulative carbon dioxide, which is why “temporary” dirty power still counts.
IPCCLarge, always-on loads can shape transmission, capacity markets, generation queues, and utility resource plans. That is why AI power demand belongs in public planning, not private side letters.
DOE / LBNLDiesel engines, gas turbines, fuel cells, and “temporary” generators can become a real pollution source when grid service is constrained or deadlines outrun permits.
Air permitsWater use, cooling choices, thermal discharge, and peak electricity demand all get worse during heat waves—the same moments when communities are most vulnerable.
Public healthBut the public experiences speed as skipped hearings, missing load numbers, emergency generators, rushed substations, and “trust us” power deals.
But if the plan requires new gas, higher bills, water stress, and tax abatements, the innovation is being financed by everyone who did not get equity.
But reliability for one private compute customer cannot mean fragility for homes, hospitals, schools, small businesses, and ratepayers.
The company will arrive with renderings, jobs numbers, and a promise that the grid will be fine. The community response should be procedural and relentless: no approval until every promise becomes a permit condition, utility filing, emissions limit, curtailment right, clawback, or bond.
Publish maximum firm load, expected average MW, annual MWh, ramp schedule, expansion rights, and the identity of the ultimate compute customer.
Require new, deliverable, hourly clean-energy matching on the same grid—not annual paper offsets that can coexist with local fossil buildout.
Lock in who pays for substations, transmission, capacity-market exposure, emergency upgrades, stranded assets, and reliability costs.
List every diesel engine, gas turbine, fuel cell, battery, and generator. Cap runtime, publish logs, and require real-time air-emissions reporting.
Show withdrawals, consumption, discharge temperature, chemical use, drought operations, and heat-wave operations before any rezoning vote.
During grid emergencies, heat waves, winter peaks, droughts, or air-quality episodes, residents should come before speculative compute.
Publish abatements, job guarantees, school-district impacts, clawbacks, and the permanent jobs-per-megawatt ratio.
Ban silent phase-two load creep. Any new MW, generators, cooling towers, or substations must trigger a new public hearing.
If the AI customer leaves, the operator changes, or the bubble pops, the public should not inherit a stranded shell and unpaid cleanup obligations.
Annual offsets are not enough. Demand hourly, local, additional, deliverable clean power with penalties for failure.
Ratepayers should not finance speculative AI load through transmission upgrades, capacity prices, or stranded fossil assets.
Diesel engines, gas turbines, and fuel cells need public operating limits, emissions caps, and real-time reporting.
Ask how many permanent local jobs remain after construction per MW of load, per gallon of water, and per dollar of tax abatement.
Choose your state/context to automatically output an authoritative, facts-anchored letter for local commissioners or city councils demanding compliance with the 2026 Ratepayer Protection Pledge.
“I am not asking this board to decide whether AI is good or bad. I am asking you to refuse a blank check. Before any approval, the applicant must publish the project’s maximum load, annual electricity use, water use, backup-power emissions, ratepayer impacts, tax abatements, expansion rights, and enforceable clean-power plan. If those numbers are confidential, the project is not ready for public approval.”
“Please identify every grid upgrade required for this load, the expected cost, the party responsible for paying, and the mechanism that prevents residential and small-business customers from subsidizing speculative AI demand.”
“How many backup or on-site generation units are proposed, how many hours can they run, what pollutants are capped, and where will the public see operating logs in real time?”
“Will you commit to delaying approval until all utility studies, water correspondence, environmental review documents, tax-abatement models, and side agreements are posted publicly?”
“A data-center project near us may need power at power-plant scale. We are asking for basic public terms before approval: no secret utility deal, no ratepayer subsidy, no fossil loophole, no water surprises, and no expansion without another vote.”
“This is being described as a local land-use issue, but the unanswered story is electricity: who pays for the grid upgrades, whether new fossil generation is involved, and whether the company’s clean-energy claims are enforceable hour by hour.”
The playbook is not mysterious: show up early, demand documents, translate technical load into household consequences, and force every elected official to vote on the real project—not the sales deck.
Ask for a moratorium, continuance, or full environmental review before rezoning. A delay is often the first win.
Request utility studies, interconnection applications, water correspondence, tax-abatement models, NDAs, and meeting calendars.
Farmers, ratepayer advocates, climate groups, respiratory-health advocates, fiscal conservatives, labor, and nearby homeowners often share the same demand: prove it.
Translate load into household bills, substations, gas units, diesel hours, water withdrawals, truck traffic, noise, and lost tax base.
Promises are press releases. Conditions are enforceable: 24/7 clean power, no ratepayer subsidy, no unpermitted generation, public monitoring, curtailment rights.
“Data centre” is often a shell. Ask whether the beneficiary is Microsoft, OpenAI, Anthropic, xAI, Meta, Amazon, Google, Oracle, CoreWeave, or another cloud customer.
When companies announce 5, 6, 10, or 30 GW of capacity, demand a reconciliation table: what is duplicate, what is incremental, and what hits this grid?
Archive every deck, hearing, promise, press release, and permit. Build a public timeline that journalists and regulators can use.
Use this as a starter packet. It mixes primary sources, official announcements, campaign pages, legal filings, and reputable reporting. Search it, copy it, and attach it to comments.