The companies building the infrastructure for your AI agents are running out of runway, and Wall Street just noticed.
The Summary
- Goldman Sachs and Morgan Stanley are flagging increased scrutiny on AI sector debt, particularly for lower-rated borrowers, as credit conditions tighten and yields rise
- US data center power demand is forecast to grow 38%, creating infrastructure bottlenecks that favor established players with proven capacity
- Investors are shifting capital toward safer, proven data center projects while speculative AI infrastructure plays face funding challenges
- The credit party that funded the AI build-out is cooling just as energy demands are spiking
The Signal
Wall Street's two biggest banks are pumping the brakes. Goldman and Morgan Stanley are warning that AI companies' debt binge is creating financial instability, especially among lower-rated borrowers who fueled the data center boom. These aren't consumer startups burning venture capital. These are infrastructure plays that took on massive debt to build the physical backbone of Web4.
The timing is brutal. Goldman forecasts 38% growth in US data center power demand, which means the companies that survive this credit squeeze will print money. But survival requires something most AI infrastructure companies don't have: patient capital and proven cash flow. The result is a flight to quality that's reshaping who gets to build the agent economy's physical layer.
"Tighter credit conditions are forcing investors to pick winners before the race is over."
Here's what's actually happening on the ground:
- Data centers that can show signed capacity agreements are getting funded
- Speculative builds in secondary markets are stalling mid-construction
- Energy infrastructure partnerships are becoming mandatory for deal flow
- Lower-rated debt is repricing 200-300 basis points higher
The shift toward safer projects isn't just about risk tolerance. It's about physics. You can't run inference at scale without power, and power infrastructure takes years to build. The companies that locked in utility partnerships two years ago now have a moat that debt alone can't replicate.
This creates a strange dynamic. AI companies need compute. Compute needs data centers. Data centers need power. But the debt that was supposed to fund this vertical stack is drying up for everyone except the most creditworthy builders. Rising yields and tighter conditions mean the marginal data center project, the one that would have gotten funded six months ago, is now dead.
The Implication
The infrastructure layer of Web4 is consolidating before most people realize it exists. If you're building agents that need serious compute, your options for where to deploy them are narrowing. The hyperscalers will win by default if independent data center builders can't access capital.
Watch which AI companies are signing long-term capacity agreements versus which ones are still shopping around. The former are building for a future where compute is scarce and expensive. The latter are betting on a supply glut that this credit crunch makes unlikely. Energy partnerships and proven power access are now table stakes, not nice-to-haves.