World GDP 2026F3.0%World CPI 2026F4.7%Brent$88/bblFed funds3.50-3.75%ECB deposit2.25%Data centre load485 TWhReadings as at 1 Aug 2026
QuantiaQuantitative economics
Dashboard QD-03 · Quarterly

Compute and Power Monitor

The compute buildout is usually covered as a technology story. It is now large enough to be a macroeconomic one. This monitor follows the money from capital commitment through to grid load, household electricity bills and the share of measured output growth it accounts for.

Vintage 1 August 2026Coverage 11 seriesBasis Published sources, all attributed
Reporting interface on a laptop display
$725bn
Combined 2026 capital spending guided by four hyperscalers
Up about 77% on 2025
485 TWh
Global data centre electricity consumption in 2025
Projected to roughly double by 2030
+32%
US household electricity prices against their 2019 level
Larger in data centre regions
-0.1pp
Estimated net drag on US GDP growth from higher power prices
2026 and 2027 combined
Scope note. Capital expenditure figures are total company capex, not data centre capex. The overwhelming majority of the increase is AI infrastructure, but the series is not a clean measure of it. Where an estimate is a bank or consultancy projection rather than an official statistic, the source column says whose it is.
Exhibit 1Hyperscaler capital expenditure, 2025 outturn against 2026 guidanceBillions of US dollars, calendar year. 2026 figures are company guidance or the midpoint of a guided range.

Sources: Hyperscaler 2026 capex guidance, February 2026 for the February 2026 guidance round; Financial Times capex tally, via Tom's Hardware, April 2026 for the revised calendar-year figures compiled after first quarter results. Microsoft's calendar 2026 figure was set at 190 billion dollars, well above the 152 billion average analyst estimate, with about 25 billion attributed by the company to higher memory and component costs. Not all of this spending is data centres.

Exhibit 2Global data centre electricity consumptionTerawatt hours per year. 2030 is the IEA base case.

Sources: IEA, Energy and AI: energy demand from AI; IEA, Key Questions on Energy and AI, executive summary. The 2030 base case is about 3 percent of projected global electricity consumption. Consumption in accelerated servers, the AI-driven component, is projected to grow around 30 percent a year against 9 percent for conventional servers.

Exhibit 4Composition of AI compute demandShare of artificial intelligence compute accounted for by inference against training, percent.

Source: Deloitte estimates as summarised in PJM capacity market analysis, January 2026. The shift matters commercially because inference load is proportional to usage rather than to model development, which changes the shape of the demand curve utilities have to plan around.

Exhibit 3Where the additional load lands, 2024 to 2030Increase in annual data centre electricity consumption, terawatt hours, IEA base case.

Source: IEA, Energy and AI: energy demand from AI. The United States and China together account for close to 80 percent of global growth in data centre electricity consumption to 2030.

Where it shows up in the rest of the economy

Transmission channels into the wider economy
ChannelEstimated magnitudeNoteSource link
Consumer spending-0.2ppDrag on real consumer spending growth in 2026 and 2027 from higher electricity prices, larger for lower-income householdsgspublishing.com
GDP growth, net-0.1ppConsumption drag partly offset by higher utility capital spendinggspublishing.com
US household electricity prices+32%Level relative to 2019, with larger increases in data centre intensive regionscepr.org
PJM capacity market$9.3bnPrice increase in the 2025-26 auction attributed to data centre load, implying roughly 18 dollars a month in western Marylandavidsolutionsinc.com
Share of US GDP~5%AI-related capital formation as a share of output, a level last seen in the late 1990s technology boombetafinch.com

Magnitudes are the estimating institution's own, not Quantia's. They are not additive: the first two rows describe the same mechanism at different points in the accounts. Sources: Goldman Sachs, macroeconomic spillovers from AI electricity demand, February 2026; CEPR, powering the digital economy; PJM capacity market analysis, January 2026; AI capital formation and Q1 2026 US GDP.

A note on timing. Capital spending enters GDP immediately and revenue arrives later, so an investment cycle of this size flatters growth on the way up and subtracts on the way down. The turn does not require the spending to fall. It only requires it to stop rising, because it is the change in the level that enters the growth rate.