Tech news

Why AI data centres are showing up in electricity bills

AI data centres are now big enough to move power prices, and US lawmakers want them to pay their own way. Here is how a server farm ends up on a household bill, and where the story is more complicated.

By Shashi Kumar ·

A row of glowing server racks connected by a lightning-bolt power line to a small house with an electricity meter

When a tech giant announces a new data centre these days, the press release has a new line in it. Google's $17 billion Wyoming plan, unveiled on 5 November 2026, says it plainly: "We will pay for 100% of the power we use."

A company only promises that when people suspect the opposite. And a day later, Reuters reported that Democrats, who take control of the US Congress in January, plan to make protecting households from data-centre power costs a priority.

So how does a building full of AI chips end up on a family's electricity bill? And is AI a real cause or a convenient villain for bills that were rising anyway? The honest answer: sometimes, in some places, through specific mechanisms.

Why AI data centres need so much power

A data centre is computers that turn electricity into heat while they calculate, plus cooling to remove that heat. AI intensifies both.

The International Energy Agency's April 2026 report, Key Questions on Energy and AI, puts the numbers in perspective:

  • Data centres used about 485 TWh of electricity in 2025, slightly more than 1.5% of global demand. Their consumption grew 17% that year, and AI-focused data centres grew 50%.
  • The IEA's base case projects about 950 TWh by 2030, roughly 3% of global electricity.
  • A single AI server rack, the size of a large fridge, could by 2027 draw as much peak power as 65 households.
  • The largest AI campuses being built in the US, at 600 MW to 5 GW each, will use as much electricity as 600,000 to 5 million households once complete.

The real issue is concentration: gigawatts arriving in one region on a one-to-two-year build schedule, on a grid that takes many years to expand.

How a data centre ends up on your bill

There are three main routes, and they work differently.

1. Wholesale prices

When demand rises quickly, grids lean on older, pricier power plants that would otherwise sit idle. That can push wholesale prices up for everyone in the region, the IEA notes, especially where data centres cluster and congest transmission lines.

2. Capacity markets: paying for tomorrow's power today

This is the clearest example. PJM, the grid operator covering all or parts of 13 US states including Virginia's "Data Center Alley", runs an annual auction to pay power plants to be available years ahead. Higher forecast demand means higher prices, and those costs flow into customers' bills.

PJM's 2028/2029 auction, held in July 2026, cleared at the regulatory price cap of $325 per MW-day, totalling $16.4 billion, and still came up 6,831 MW short of its reliability target. PJM notes that $16.4 billion is not the same as the cost to consumers, since some load is hedged.

The independent market monitor was blunt. Its July 2026 analysis attributes $6.3 billion of that $16.4 billion (38.2%) to data-centre load, and $29.4 billion across the last four auctions. Its proposal: move data centres into a dedicated auction so they "pay for their own capacity and do not impose capacity costs on other customers."

3. Grid upgrades and utility rate cases

New substations and transmission lines are typically paid for through regulated rates, which utilities ask state regulators to approve in "rate cases". If those costs are spread across all customers, households help pay for upgrades built mainly for a data centre. If special large-load tariffs are used, the data centre pays more of its own way. The fight is largely about which of those happens.

That fight reached Washington. A bipartisan Ratepayer Protection Act, which would have required state regulators to consider making large users pay for grid upgrades, passed the House 417-3 but was rejected by the Senate on 30 September 2026, with Democrats calling it "toothless" for not making it mandatory.

What bills are actually doing

The US Energy Information Administration's October 2026 outlook expects the average US residential price to reach 18.2 cents per kWh in 2026, up from 17.3 cents in 2025, and 18.7 cents in 2027. It also notes commercial-sector electricity demand, which includes data centres, rose 5% year on year in the third quarter of 2026.

The counter-arguments (and they are real)

The evidence does not support blaming every bill increase on AI.

  • No simple link. The IEA found US states with greater load growth from 2019 to 2024 generally did not see higher price increases. Fuel costs, storms, taxes and network charges drive most bill changes.
  • More demand can lower unit costs. In grids with spare capacity, a big new customer spreads fixed costs over more units, which can push per-unit charges down.
  • Forecasts may be inflated. The IEA warns PJM forecasts may include duplicate or speculative requests from developers shopping several sites.
  • Efficiency is improving fast. Energy use per AI task has dropped by at least an order of magnitude a year recently, the IEA says. A simple text query now uses less electricity than running a television for the same time. The catch: video generation, reasoning and agents can use hundreds or thousands of times more.
  • Flexibility helps. The IEA estimates that if data centres could flex their demand in just 0.1-1% of the hours in a year, current systems could absorb all new data-centre capacity expected through 2035.

The scramble for generators

Because grid connections are slow, many developers want their own power. Reuters reported on 9 November 2026 that worldwide gas turbine orders exceed 100 GW against manufacturing capacity of 60 to 70 GW a year, with US delivery times averaging six years. Developers are buying second-hand turbines and even units based on jet engines.

Self-supply sounds like it protects households, but the IEA flags a twist: if data centres take power from existing plants or reduce their draw from the network, other customers can end up with higher per-unit infrastructure costs.

What is happening in India

India is on the same curve, a few years behind. JLL's mid-2026 report puts capacity at about 1.6 GW and projects it to triple to 6 GW by 2029, led by Mumbai and Chennai, with hyperscalers committing more than $50 billion.

The pressure points are local grids and water. On 14 October 2026, NITI Aayog released a data-centre evaluation framework urging states to check both before approving projects, Outlook reported. It recommends joint power-readiness plans with discoms, approving extra load only once infrastructure is commissioned, automated demand response for sites above 20 MW, and 20-30-year water balance studies, with treated or desalinated water considered in stressed areas. It is guidance; states decide how to apply it.

Household tariffs in India are set by state electricity regulators, and commercial users typically already pay more than homes, so the US capacity-market route does not translate directly. The more likely local effects are strain on specific substations, more backup generation, and competition for water in already stressed areas.

How it unfolded

Data centres and power bills: 2026 so far

  1. PJM auction clears at the price cap The 2028/2029 capacity auction clears at $325/MW-day, $16.4 billion in total, and 6,831 MW short of target. Source
  2. Market monitor blames data-centre load Monitoring Analytics attributes $6.3 billion of the auction's charges to data centres. Source
  3. Senate rejects ratepayer bill The Ratepayer Protection Act fails 57-43 after Democrats call it toothless. Source
  4. NITI Aayog framework India's policy think tank asks states to check grid readiness and water stress before clearing data centres. Source
  5. Google's $17B Wyoming plan Google pledges to pay for 100% of the power its Cheyenne site uses, backed by a 702 MW fuel-cell project. Source
  6. Democrats line up AI bills Reuters reports data-centre power costs, independent AI testing and liability are on the agenda for January. Source
  7. Turbine shortage deepens Reuters reports gas turbine orders far exceed manufacturing capacity, with six-year US waits. Source

What could change

Reuters reported on 6 November 2026 that Chuck Schumer, expected to become Senate majority leader, named protecting consumers from data-centre energy costs as a priority, while Representative Lori Trahan plans bills on independent testing of AI models and on company liability for AI harms. President Trump has repeatedly dismissed calls for mandatory AI rules, so anything binding faces a hard road.

Other levers to watch: large-load tariffs, "use it or lose it" rules for reserved grid capacity, faster connections for sites that flex demand, and separate capacity auctions for data centres.

What we know vs what we don't

  • We know: data-centre demand is growing fast; in PJM, data-centre load is a major driver of capacity costs; companies are under pressure to pay their own way.
  • We don't know: how much of the forecast load will actually get built, how much bills will rise because of it rather than fuel and weather, and whether new US rules will pass.

Frequently asked questions

Do AI data centres really raise electricity bills?

In some places, yes. In the PJM region, the market monitor attributes $6.3 billion of the 2028/2029 capacity auction's $16.4 billion to data-centre load. But the IEA finds no simple nationwide link between demand growth and prices.

Why do AI data centres use so much electricity?

AI chips are packed far more densely than ordinary servers, and they need intensive cooling. The IEA says one AI rack could draw as much peak power as 65 households by 2027.

How much electricity do data centres use worldwide?

About 485 TWh in 2025, slightly over 1.5% of global electricity, according to the IEA. Its base case projects about 950 TWh by 2030, roughly 3% of global demand.

Will data centres affect electricity bills in India?

Not in the same way as in the US, because state regulators set tariffs and there is no PJM-style capacity market. The bigger near-term concerns are local grid strain and water use, which NITI Aayog's October 2026 framework asks states to check.

Is AI getting more energy efficient?

Yes. The IEA says energy per AI task has fallen at least tenfold a year recently, but heavier uses such as video generation and agents use far more per request, so total demand still rises.

The bottom line

AI data centres are not the sole reason power bills are rising, but in a few fast-growing regions they have become a measurable part of it. The real question is not whether to build them, but who pays for the wires and power plants they need.

If you run AI features in your own product, the same logic applies at a smaller scale: compute is not free. Our guide to controlling LLM feature costs and our look at free cloud tiers and their limits are good next reads, and for the jobs side of the AI boom, see what the 2026 evidence says about AI and work.

Sources and further reading

  1. Key Questions on Energy and AI · International Energy Agency · 2026-04-16
  2. 2028/2029 Base Residual Auction Report · PJM Interconnection · 2026-07-14
  3. Market Monitor Report, IMM MC, July 28, 2026 · Monitoring Analytics · 2026-07-28
  4. Short-Term Energy Outlook: Electricity, coal and renewables · US Energy Information Administration · 2026-10-06
  5. US Senate rejects bill targeting AI data centre electricity costs · Al Jazeera · 2026-09-30
  6. Democrats planning flood of AI bills when they take control of Congress · Reuters via The Star · 2026-11-06
  7. Investing in Wyoming's infrastructure, energy and workforce · Google · 2026-11-05
  8. AI power rush sends data centres hunting for used turbines and jet engines · Reuters via BOE Report · 2026-11-09
  9. India data centre 2026 mid-year report · JLL · 2026-08-31
  10. Why NITI Aayog Wants States to Check Power, Water Before Clearing Data Centres · Outlook India · 2026-10-15

Contact Acquiretech Software Solution: info@acquiretechsoftwaresolution.in · WhatsApp +91 80786 33912 · Send an enquiry