The AI capex wave in numbers

Estimates vary by methodology, but they converge on the same order of magnitude. The four largest US hyperscalers – Amazon, Google, Microsoft, and Meta – guided to roughly $630–725 billion in combined 2026 capital expenditure on their Q1 2026 earnings calls, up around 60–77% on 2025.

Goldman Sachs Research, using a broader definition that includes AI-related public companies beyond the hyperscalers, puts total global AI-related investment at approximately $1 trillion for 2026 alone, with cumulative AI investment reaching roughly $1.8 trillion by the end of the year. Looking further out, McKinsey's often-cited framework projects $6.7 trillion in cumulative global data centre capital expenditure by 2030, with around 70% of that attributable to AI workloads specifically.

Global data centre electricity consumption was approximately 415 TWh in 2024 – about 1.5% of all global electricity use. The IEA's base case projects this will more than double to around 945 TWh by 2030, representing just under 3% of projected global electricity demand, growing at roughly 15% per year – more than four times the growth rate of every other sector combined.

In the United States specifically, which hosts around a third of the world's data centre capacity, consumption is projected to rise from 200 TWh in 2022 to 260 TWh by 2026 – about 6% of all US electricity use.

This is genuinely contested, and both sides of the dispute are worth hearing. Industry-aligned estimates project permanent US data centre employment reaching around 650,000 positions by 2026, with the sector's broader economic footprint cited at 4.7 million jobs.

A separate Brookings analysis of counties receiving their first large data centre found real but modest gains: data-processing employment rose 56% and telecommunications rose 43% over the following decade – translating in practice to roughly 100–200 additional local jobs per county. The honest summary: construction employment is real but temporary; permanent operational employment is consistently small relative to the capital invested.

A significant and rapidly accelerating amount. According to Data Center Watch, local opposition blocked or delayed 75 projects worth $130 billion in just the first three months of 2026 – roughly matching the total blocked across the whole of 2025 in a single quarter.

In July 2026 alone, 142 protests were recorded across 42 US states. In Goodyear and Buckeye, Arizona, a $14 billion project from developer Tract was withdrawn entirely after local authorities blocked the rezoning needed to proceed, following sustained resident organising.

The polling is unusually consistent for a contested topic.

OpenAI's own CEO, Sam Altman, has publicly acknowledged the scale of the shift: "Clearly, people hate data centres – right now, at least."

A meaningful and growing amount, though it varies enormously by cooling technology and disclosure practice. Microsoft reported 6.4 million cubic metres of water consumption in its FY2022 filings, rising with its data centre expansion since. Across the hyperscalers as a group, disclosed water consumption rose an estimated 25–40% year-over-year through the 2024–2025 reporting cycle.

Newer liquid and immersion cooling systems can reduce direct water use by 70–90% compared with older evaporative designs – but they do not reduce the underlying electricity demand, which remains the larger environmental cost by most measures.

By most credible projections, yes, and the effect is already showing up as a driver of public opposition. Goldman Sachs Research projects rising data centre power demand could drive a 6% increase in US electricity bills nationally over the next year, sharpest closest to the facilities themselves.

In some regions the effect is already large: data centres already account for more than 20% of electricity demand in Ireland, and over a quarter of demand in the US state of Virginia.

Emphatically, and on a genuinely bipartisan basis ahead of the 2026 US midterms. More than 99% of political ad spending on data centres this year has been critical of them, spanning candidates from both major parties.

Pennsylvania's governor signed an executive order in August 2026 requiring developers to fund all new grid infrastructure their projects require and enter formal community benefit agreements. At the federal level, Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez introduced the AI Data Center Moratorium Act in March 2026, which would pause new large-scale construction pending legislation on AI safety, worker protections, and environmental standards.

It is a live and increasingly mainstream concern among investors, not a fringe position. Shares in several major hyperscalers sold off following their own 2026 capex guidance, as investors weighed the scale of committed spending against uncertain near-term returns.

Morgan Stanley and JPMorgan analysts have projected the technology sector may need to issue $1.5 trillion in new debt over the coming years specifically to finance AI infrastructure – a marked shift from a capex cycle funded largely from cash flow toward one increasingly funded by leverage. None of this means the infrastructure is unnecessary; it means the scale, financing structure, and pace of the buildout are legitimate questions for boards and investors, not just for the communities living next to the facilities.

The emerging best practice, visible in the regulatory responses already in motion, has a few consistent features: developers funding their own required grid upgrades rather than socialising the cost onto existing ratepayers; genuine, early community engagement rather than after-the-fact disclosure; transparent reporting of water and electricity use; and a realistic, evidence-based account of local job creation rather than the inflated broad-sector figures independent analysis has repeatedly had to correct.