Was 2025 the Year of the AI Datacenter? The Infrastructure Boom and Its Discontents

Was 2025 the Year of the AI Datacenter?

TLDR: 2025 saw unprecedented AI infrastructure investment with over $400 billion in hyperscaler CapEx, but also $162 billion in stalled or cancelled projects. Community backlash is growing, ROI remains elusive, and the energy grid faces serious constraints as facilities scale to Gigawatt levels.

Summary:

This is a comprehensive and somewhat sobering analysis of the AI infrastructure buildout that defined 2025. The numbers are staggering: aggregate capital expenditure for top Big Tech hyperscalers exceeded $400 billion, primarily driven by AI infrastructure and data centers. AI captured close to 50% of all global funding in 2025, up from 34% in 2024, reaching $202.3 billion invested in the sector.

But the article raises uncomfortable questions that rarely make it into the hype cycle. According to datacenterwatch.org, $162 billion worth of data center projects in the US have been stalled or cancelled over the course of a year, with 60% in the latest 3 months. It's not just supply chain issues - local residents are increasingly protesting hyperscalers and their manipulation of real estate and local politicians.

The concentration of capital is remarkable. The "BigAI" cohort - OpenAI, xAI, Anthropic, and similar foundation model companies - received 66.6% of all AI funding. These are partially owned by Microsoft, Amazon, and Google, who are also their biggest customers. Oracle has taken on roughly $108 billion in debt to fund AI infrastructure, with $248 billion in lease commitments for data centers over the next 15-19 years. The question of whether this capital allocation makes sense increasingly depends on ROI that remains elusive - a December 2025 Reuters analysis noted that "the vast majority of businesses are struggling to realize a meaningful return on their AI investments."

The scale of planned facilities is genuinely hard to comprehend. OpenAI's Stargate Abilene data center will need enough electricity to serve the population of Seattle, more than 250x the computing power that trained GPT-4, and $32 billion in construction and IT equipment costs. And that's just one facility. By 2026, we'll see Gigawatt-scale data centers from Anthropic-Amazon, xAI's Colossus 2 with 1 million+ GPUs, Microsoft Fayetteville, Meta's Prometheus, and Stargate Abilene.

The environmental and community impact is becoming a political issue. Data centers significantly raise electricity bills, with some areas seeing costs jump up to 267% higher in places with heavy activity. National averages have increased 30%+ since 2020, with projections of 8-25% further increases by 2030. Even Republicans are distancing themselves - Florida's Ron DeSantis floated new limits as part of an "AI bill of rights," while Georgia's Marjorie Taylor Greene and Missouri's Josh Hawley have questioned Trump's state preemption drive.

Perhaps most interesting is the emerging race for orbital data centers. SpaceX is targeting a $1.5 trillion IPO (largest in history) partly on the strength of space-based AI data center plans. Blue Origin has reportedly had a team dedicated for over a year to developing orbital AI data center technology. The physics make sense: solar power in orbit is 8x stronger than on Earth, and data centers could soon consume 8% of global electricity.

Key takeaways:

$400B+ hyperscaler CapEx in 2025, but $162B in stalled/cancelled projects signals constraints
AI captured 50% of global funding; BigAI (OpenAI, xAI, Anthropic) received 66.6% of AI investment
Gigawatt-scale facilities coming in 2026: Stargate needs electricity for a Seattle-sized population
Community backlash growing; electricity bills up 30%+ with more increases projected

Tradeoffs:

Infrastructure first-mover advantage but questions about depreciation and long-term ROI
US capital dominance but less cost-efficient than China's open-weight model approach

Link: Was 2025 the Year of the AI Datacenter?

The summaries provided are based on newsletter content and represent interpretations of the original articles. Readers should consult the original sources for complete technical details and authoritative information.