ASU Study Finds Data Centers Heat Nearby Neighborhoods as US Power Demand Forecasts Climb

ASU Study Finds Data Centers Heat Nearby Neighborhoods as US Power Demand Forecasts Climb Arizona State research links data center waste heat to higher local temperatures and energy bills, sharpening scrutiny of a buildout already straining US grids. Data centers used about 4.6% of total US electricity in 2024, a share government estimates say could nearly triple by 2028.⁴ An Arizona State University study now adds a local cost to that national figure: the facilities can raise temperatures and energy burdens in the neighborhoods around them, putting a neighborhood-level price on infrastructure usually measured in gigawatts and capex.^3,4 That matters because the power-demand story behind the AI buildout has so far been told in aggregate. The IEA says data centers now account for more than 1% of global electricity use.² Within two years of ChatGPT's 2022 launch, about 40% of households in the US and UK reported using AI chatbots, according to the IEA.³ The ASU work points at where the strain concentrates, and it is not evenly spread. The emissions trail is already visible in the hyperscalers' own numbers. Google's emissions jumped nearly 50%, Amazon's rose 33%, Microsoft's more than 23%, and Meta's more than 60%, even as the same companies once promised to run on clean power.⁴ Google now calls its 2030 carbon-free goal a "moonshot."⁴ The retreat is telling. Firms say they must stay flexible as they race to build data centers that can consume more power than entire cities.⁴ What fills the gap is mostly fossil. Natural gas accounted for more than 40% of the electricity powering US data centers in 2024, while coal supplied 30% globally, the IEA said.⁴ So the marginal megawatt feeding an AI campus is, more often than not, a thermal one. That complicates the clean-energy narrative the sector spent years building. The demand math is the part traders care about. Some analysts expect nationwide US electricity use to rise as much as 20% over the next decade, with data centers a primary driver.⁴ The EIA, in its Annual Energy Outlook 2026, projects server electricity consumption growing across the commercial building stock, with standalone data centers adding more than all other data-center rooms combined by 2050.⁵ These are long-horizon curves, but they set the direction. Capital has already moved on the thesis. Fluence Energy shares closed at $24.16 on May 8 (2026-05-08), up 98.2% in a single week after the company disclosed master supply agreements with two hyperscalers and a record $5.6 billion backlog.¹ Quick Read Capital framed the rotation plainly: money is flowing into companies that can supply power for AI buildouts, with nuclear and renewable baseload pitched as the cleanest fixes to the constraint.¹ The Fluence move is worth a second look before chasing it. The stock is still down roughly 39% year to date, leaving a micro-cap in turnaround territory rather than a clean breakout.¹ Q1 2026 brought positive adjusted EBITDA of $2.0 million, a fourth straight quarter in the black, with non-GAAP gross margin widening to 52%.¹ CEO Arun Narayanan said the "operational discipline and margin profile we established in 2025 are proving durable."¹ A 98% week tells you about appetite, not about valuation. The bottleneck is physical, not financial. Battery storage firms are seeing surging interest from power-hungry AI data centers, but long interconnection queues and a supply chain heavily dependent on China are limiting how fast they can scale, Reuters reported.⁶ The same grid constraints that lift power-supplier equities also cap how quickly any of them can actually deliver. The IEA's framing of an industry getting greener per task while consuming more power in aggregate captures the tension.⁷ Not every signal points up. On the EnergyReader signal ledger, German baseload front-month carried bearish demand-side reads, a reminder that European power and US data-center load are different markets with different drivers.⁷ A US interconnection queue does not move a German curve. Traders extrapolating one buildout across every grid will get the regional dispersion wrong. The ASU study itself will not reprice a single contract. What it does is hand local regulators and ratepayer advocates a concrete grievance, in degrees and dollars, at exactly the moment utilities are negotiating the rate structures and interconnection terms that decide whether the AEO2026 demand curve gets built.^5,3 Watch the siting fights, not just the load forecasts. The next constraint on AI power may be a zoning board, not a turbine.