Why the AI data centre boom is failing — and what comes next.
We were told AI would change everything.
Nobody mentioned it would first need to rewire the entire electricity grid.
Here is where we actually are in June 2026. Against a 16 gigawatt announced pipeline of new AI data centres in the United States, only 5 gigawatts is actually under construction. Between 30% and 50% of projects planned for this year are delayed or cancelled. The bottleneck is no longer money. It is no longer chips.
The binding constraint has shifted to the physical electrical layer — electricity, transformers, switchgear, and battery systems.
Let that sit for a moment.
The most expensive construction boom in history — stalling
Tech companies are collectively spending $400 billion this year on AI infrastructure. More than any group of firms has ever spent to do just about anything. The Apollo programme allocated roughly $300 billion in inflation-adjusted dollars to get America to the moon — over an entire decade. The AI buildout requires a new Apollo programme every ten months.
And half of it is stalling in planning committees.
The infrastructure gap — May 2026 data
- 16 GW announced in the US pipeline for 2026
- Only 5 GW actually under construction
- 30–50% of projects delayed or cancelled
- 5-year transformer lead times reported by US ITC
- $650B in capex pointed at facilities that cannot energise on schedule
Capital is no longer the gating factor. Power infrastructure is.
The market is pricing in 114 gigawatts of power capacity. Only 15.2 gigawatts is actually under construction. That is a 750% gap between hype and reality.
Communities are saying no
This is the part of the story that the earnings calls do not mention.
A Gallup poll from March 2026 found that 70% of respondents oppose new AI data centres being built in their neighbourhood. In 2025 alone, local opposition led to the delay or cancellation of projects totalling $156 billion. Community pushback has become — in the words of industry analysts — a ‘material driver of attrition.’
It is not hard to understand why. A data centre that arrives in a rural county brings noise, draws down the water table, strains the local grid, and employs very few people. The benefits flow to shareholders in San Francisco. The costs fall on residents in Virginia, Texas, and Indiana.
| What communities are losing | What tech companies are getting |
|---|---|
| Quiet agricultural land and green space | Cheap land at scale |
| Stable, affordable electricity bills | Below-market power contracts |
| Local water supply and groundwater | Free cooling infrastructure |
| Planning certainty and community character | Local tax incentives |
| Any say in what gets built near them | Speed to market on AI infrastructure |
142 activist groups in 24 US states have now successfully stalled or stopped $64 billion worth of data centre projects. Many US states have introduced temporary building freezes. The backlash is organised, cross-partisan, and growing.
The economics do not work yet
Here is the number nobody wants to say out loud at a conference:
95% of enterprises report zero measurable return on their generative AI investments. Despite $30 to $40 billion deployed into enterprise AI.
OpenAI missed its own internal revenue targets for 2025. Microsoft is losing over $20 per user on some AI subscriptions. A National Bureau of Economic Research study from February 2026 found that 90% of firms report no impact of AI on workplace productivity — even as executives keep projecting gains.
The circular economy of AI investment
- Nvidia invests in OpenAI. OpenAI buys Nvidia chips.
- Oracle borrows $43 billion to build data centres for OpenAI.
- OpenAI commits to $1.4 trillion in infrastructure spending over 8 years.
- OpenAI generates $20 billion in revenue. Projects losses through at least 2028.
- Microsoft loses $20+ per user on AI subscriptions.
- Investors fund the next round anyway.
This is not a business model. It is a belief system backed by debt.
Only 21% of S&P 500 companies reported at least one measurable AI benefit as of Q1 2026, according to Morgan Stanley. That is adoption without returns — which is the definition of a commercialisation challenge.
What happens next — a realistic timeline
- 2026The correction
Infrastructure investment continues but slows. Weaker AI companies lose funding. The gap between announced projects and actual construction becomes impossible to hide. Power and community constraints force site consolidation.
- 2027The shake-out
Smaller AI startups running on negative margins fold. Foundational model providers consolidate. Enterprises stop running pilots and demand proof of ROI before any further AI spend is approved. The 95% with zero return begin to question every contract.
- 2028The reset
Surviving firms emerge leaner, more focused, and with real unit economics. AI stops being a story about infrastructure and starts being a story about outcomes. The technology is still transformative — but only for organisations that built the foundation right.
- 2030+The maturity
AI becomes utility infrastructure — like electricity or broadband. Essential, mostly invisible, and delivered by a small number of dominant providers. The race is over. The winners have already been decided.
The verdict
This is not a collapse. It is a correction. And corrections are healthy.
The technology works. The hype was the problem. We built the story of AI around ambition — around the scale of the investment, the audacity of the vision, the speed of the build. We forgot to build the story around the problem being solved and the community bearing the cost.
The firms that survive the next three years will not be the ones that built the most infrastructure. They will be the ones that asked the harder question first:
What problem are we actually solving — and who actually benefits?
The infrastructure imbalance is not bursting. It is being stress-tested by reality.
And reality, as always, wins.
Sources
- Sightline Climate / Bloomberg (May 2026)
- National Bureau of Economic Research (February 2026)
- Morgan Stanley, Q1 2026
- MIT NANDA research group (August 2025)
- Gallup (March 2026)
- IBM, Cost of a Data Breach 2025
- Deutsche Bank
- Oliver Wyman
- Man Group
- Video reference: ColdFusion, ‘Why Building AI Data Centres Isn't Working Anymore’ (8 June 2026)
First published on LinkedIn on 12 June 2026. Read the original and join the discussion
