The same investors who funded the AI gold rush are now asking to see the blueprints.

The Summary

The Signal

The AI infrastructure boom is hitting its first serious speed bump in the bond market. Citigroup's analysts report that investors are becoming pickier about which data center projects get funded. This isn't a pullback, it's a filter. The money is still there, but it wants to see specifics: power contracts, tenant commitments, construction timelines that make sense.

The shift comes as real deals continue to close. Cipher Digital is raising $810 million in junk bonds for a data center project linked to Amazon. That "tied to Amazon" part matters. The bond market isn't saying no to data centers. It's saying show me the anchor tenant. Show me the power grid connection. Show me you're not just betting on vibes.

"Investors are scrutinizing AI infrastructure financing deals more carefully, according to Citigroup."

This is what market maturation looks like in real time. A year ago, anything with "AI" and "data center" in the pitch deck could get funded. Now the market wants to know:

  • Which hyperscaler signed the lease
  • How many megawatts you can actually deliver
  • Whether the local utility can handle the load
  • What happens if AI compute demand plateaus in 2027

The Cipher deal shows the playbook that still works. Amazon's involvement isn't just a nice-to-have, it's the entire credit story. Junk-rated bonds for infrastructure projects need that kind of counterparty. The investors buying these bonds aren't making a bet on the future of AI. They're making a bet that Amazon needs compute capacity and will pay for it.

The Implication

If you're building AI infrastructure, the era of easy money just got harder. Speculative data center projects without firm commitments from major tech companies will struggle to find bond financing. The winners will be companies that secured long-term contracts before trying to raise capital, not the other way around.

Watch for a widening gap in bond yields between data center projects with hyperscaler backing and those without. The market is learning to price the difference between "building for AI" and "building for a specific AI workload with a signed contract." That spread is where the real story lives.

Sources

Bloomberg Tech