The smart money is tapping the brakes while everyone else floors it into the data center gold rush.
The Summary
- Carlyle's Mark Jenkins warns investors not to rush AI data center deals, citing a massive pipeline of new capacity that lets capital be selective
- Growing concentration risk and uncertain AI business models make speculative developers vulnerable while well-capitalized hyperscalers benefit
- Communities are demanding more from data center projects, raising the bar for who gets to build
The Signal
Carlyle Group manages $435 billion in assets. When their infrastructure head tells you to slow down, you listen. Mark Jenkins isn't saying the AI data center boom is fake. He's saying the boom is so real that you don't need to chase bad deals anymore.
The math is simple: a long pipeline of new capacity means supply is coming whether you fund it or not. The leverage shifts to capital. Investors can wait for better terms, better locations, better partners. The desperate money that funded anything with "AI-ready" in the pitch deck six months ago? That money is about to learn expensive lessons.
"A huge pipeline of new capacity means they can afford to be selective."
Jenkins flags three specific risks that separate winners from wrecks:
- Concentration risk: Too many data centers in the same grid creates power bottlenecks and regulatory pushback
- Business model uncertainty: Nobody knows which AI use cases actually print money at scale yet
- Community demands: Local governments now have enough proposals to pick winners based on job quality, environmental impact, and tax deals
The gap between hyperscalers and speculative developers is widening into a chasm. Amazon, Microsoft, Google, and Meta have balance sheets that let them negotiate with utilities, wait out permitting battles, and absorb cost overruns. They're building for their own compute needs with 10-year horizons.
Speculative developers building on spec? They're pricing in 18-month lease-up timelines and investment-grade anchor tenants that may never materialize. When those models break, the well-capitalized players will buy the assets at a discount or just build around them.
The Implication
If you're allocating capital to AI infrastructure, Jenkins just handed you permission to be patient. The best deals in 2027 will likely come from distressed 2026 projects that overestimated demand timing or underestimated community resistance.
For hyperscalers, this validates the build-it-yourself strategy. For speculative builders banking on the Field of Dreams model, it's time to lock in anchor tenants or find partners with deeper pockets. The window where "AI data center" was enough of a story to raise money is closing. Now you need power contracts, tenants, and community buy-in before anyone writes a check.