The AI trade involves a complex capital structure where revenues circulate within a system powered by non-traditional credit. This setup works well as long as cash flow covers expenses. However, issues arise when the cash flow diminishes, and the costs have to be addressed through term sheets.

Oracle’s large commitment illustrates this shift from a cash-powered structure to debt-driven operations, resulting in changes in the capital cycle dynamics. The shift from equity support to creditor involvement signals a change in priorities. The reliance on private credit suggests a potentially unstable situation if credit conditions change, affecting AI capital expenditure (capex) significantly.

Challenges extend beyond funding to infrastructure, particularly in power requirements. Building new data centres requires substantial electricity input, which cannot be expedited through financial means alone. Special tariffs impact overall costs, and delays in power supply exacerbate these problems. This reflects a broader economic cycle where perceived self-funding productivity becomes questioned against macroeconomic factors.

The potential AI bubble depends on the rising cost of marginal expansion in both finances and power. Financial market responses will likely fluctuate between periods of fear and recovery. The key indicator of issues will not be market sentiment, but rather tightening funding and slippages in capex projections with utilities facing increasing challenges to meet energy demands.

The AI capital spending cycle, where revenues just loop back into the system, is starting to show its first real cracks. We’re now watching the funding shift from easy cash flow to debt, and the cost of that debt is getting noticed. For instance, recent Q3 2025 filings from major private credit funds revealed that borrowing costs for new data center projects have ticked up by at least 75 basis points since the summer.

This nervousness is visible in the options market, which is now pricing in more risk for the tech sector than it has since the brief market panic of 2024. Implied volatility on long-dated puts for semiconductor ETFs has climbed steadily over the past quarter, suggesting traders are buying protection against a slowdown. The pure upside momentum we rode for the last two years is clearly losing conviction.

Beyond the cost of money, we are hitting the hard limit of electricity. Major utilities are now openly discussing the strain, with reports from grid operators showing that interconnection queues for new power-hungry projects now extend well into 2029. The latest Energy Information Administration forecast, released just last month, hiked its estimate for data center power consumption, a demand that the current grid infrastructure simply cannot meet on schedule.

Therefore, our focus is shifting from pure growth plays to relative value trades that reflect these new constraints. We are initiating pairs trades, buying call options on transmission and electrical equipment manufacturers while simultaneously buying puts on select data center REITs in power-constrained regions. This strategy is designed to profit from the physical bottleneck that financial engineering can’t solve.

The key signals to watch are no longer product demos, but the fine print in quarterly reports and the tone of utility company earnings calls. We are listening for any mention of “capex rationalization” from the big cloud providers in their upcoming guidance. That will be the first clear sign the funding is tightening and the era of growth at any cost is hitting a wall.