The AI infrastructure boom has been the financial story of the year, driving huge debt issuances from Big Tech and emerging AI players to fund the massive global data center build-out. But according to new research from Goldman Sachs this week, that funding frenzy is starting to face a cold, hard reality check: investors are getting nervous about the debt, and performance is lagging the broader credit market.

This isn't just routine market anxiety; it reflects a deep skepticism that the AI bubble, fueled by debt, might collapse.

The "Untested" Execution Risk

The key issue causing investor worry is simple: execution risk. Investors are being asked to fund multi-billion dollar construction projects (data centers and power grids) with unproven revenue models.

As one fixed income portfolio manager put it, until investors see data centers delivered on time and on budget, and providing the computing power they are intended to, the entire market is "untested". His team refuses to invest in recent AI-linked bonds because the level of risk requires compensation "like an equity... not debt".

The underperformance of AI-linked bonds signals a divide in the market:

  • Investment-Grade: Concerns are focused on issuer-specific problems (Can Microsoft deliver on its 1GW commitment to Anthropic?).
  • High-Yield (Junk Bonds): Anxiety is a broader, sector-wide concern—investors fear the whole data center model is unsustainable or overhyped.

Regulatory Warnings and Financial Stability

The market jitters coincide with regulators sounding the alarm. The Bank of England issued a warning this week that the growing role of debt in the AI infrastructure boom could heighten financial stability risks if valuations were to correct.

The enormous capital expenditures required for AI infrastructure—which we've seen with billions poured into power sources like nuclear energy and construction projects for hyperscalers—are reaching a critical mass where the promise of future AI revenue is no longer enough to secure cheap capital.

I was talking to my friend who manages a hedge fund, and he said the entire AI boom has been based on projected growth, but now the fixed income market—the boring world of credit and debt—is demanding proof that the revenue models supporting the debt are real, not just hype.

The "Haves and Have Nots" Divide

Despite the caution, not everyone is pulling back. Some fund managers see the volatility as an opportunity to buy into high-quality debt from the "haves"—the most structurally sound companies. This confirms the trend of power consolidation we've seen across the industry.

The market is dividing into companies that can reliably execute their infrastructure build-out, and those that can't. Security selection is going to be the name of the game.

My Take

The bond market is often the first to call out a financial bubble, and this report is a loud warning shot. We've been so focused on who has the best model or the fastest chip that we ignored the simple question: can they afford to build the infrastructure to run it?

The AI hype cycle is encountering the reality of construction logistics, cost overruns, and slow regulatory approval for power projects. This isn't just about software; it's about rebar and electricity. If the market continues to punish AI-linked debt, it could force a slowdown in the build-out, giving smaller companies a reprieve but potentially stabilizing the entire sector's valuation at a more realistic level. I think the reality check is healthy, even if it causes a few sleepless nights for tech CFOs.