AI's Debt Comes Due Before Its Power Does
The industry can build a data center in eighteen months, but the grid to power it takes five years — and the debt raised to fund the buildout comes due in 2027, before the power ever arrives.
A data center can be built in eighteen months. The transmission lines and generation to power it take five years or more. The debt raised to build it comes due in 2027 and 2028. Those three clocks do not line up, and the gap between them is where the AI buildout's risk now lives. The scale of what is being asked of the grid is not subtle. American Electric Power reports that customers are seeking 190 gigawatts of additional load — five times the utility's current system size [1].
Beyond the 24 gigawatts, customers are also actively seeking to connect approximately 190 gigawatts of additional load to our system. This is five times our current system size of 37 gigawatts. — William J. Fehrman
The constraint is already binding, not hypothetical. Data center demand has driven up electricity bills for 67 million Americans across 13 states, with 64% of the increase in PJM Interconnection's capacity payments attributable to data centers [2]. New York has enacted the first statewide moratorium on certain AI data center developments [3], and communities have blocked or delayed more than $130 billion in projects nationwide [4]. Ohio and Virginia have shifted the risk onto the builders themselves — Ohio's tariff cut project inquiries from 30 gigawatts to 5.6, and Virginia now requires 14-year commitments [5]. The Ohio Manufacturers' Association put the objection plainly.
customers are being asked to pay for a future that may never arrive. — Ryan Augsburger
The financial problem is that the buildout was financed on a schedule the grid cannot meet. The industry has spent roughly $717 billion over three years while end-user license revenue grew from $1 billion in 2023 to $4 billion in 2024 [6]. Most hyperscalers are now free cash flow negative [7]. OpenAI's CFO has warned the company may struggle to fund $600 billion in committed compute contracts without accelerated growth [8].
In hindsight, that caution looks less like discipline and more like underestimating how fast demand would arrive. — OpenAI
None of this is because the revenue is fake. Cloud revenue is genuinely surging — Google Cloud up 82%, Azure up 43% past $100 billion [9]. The trouble is the other side of the ledger is growing faster: Nvidia projects hyperscaler capex rising from $800 billion this year to $1.3 trillion next [10]. The gap is widening, not closing. That gap becomes a debt problem on a specific date. OpenAI signed roughly $1.2 trillion in compute commitments between June and December 2025, and the reset wall arrives in 2027-2028, when take-or-pay contracts shift from booked backlogs to actual billing [11]. The infrastructure those contracts pay for cannot be powered fast enough to generate the revenue that would service them — the grid simply cannot be expanded in time. Financial authorities have begun to say the quiet part. The Bank for International Settlements compared the boom to the 1830s canal mania and warned that high leverage in debt-heavy nonbank structures could produce cascading defaults if optimism fades [12].
These episodes ended with an eventual reversal in investment, inducing economy-wide recessions. — Bank for International Settlements
Treasury career analysts drafted a report warning the AI market is forming a bubble with systemic reach [13].
AI and Big Tech companies are increasingly reliant on shadowy forms of debt and balance sheet magic to fund their multi-trillion dollar AI buildouts. — Elizabeth Warren
The Federal Reserve is split on what to do about it. Kansas City Fed president Jeff Schmid asked whether the industry is becoming too big to fail.
We have to correlate what's happening in AI, just from a pure scale standpoint, to some of the other experiences we've had that could, in fact, create a systemic problem. — Jeffrey Schmid
New York Fed president John Williams pushed back.
I’m not as worried about the financial stability from the leverage right now. — John Williams
That division is the unresolved question the buildout now rests on. The people who could act cannot agree on whether this is a bubble or a buildout — and the debt wall arrives in 2027 regardless of who is right.
- 1. AI Data Center Growth Strains U.S. Electrical Grid
- 2. AI Data Centers Drive Electricity Costs for 67 Million Americans
- 3. New York Bans AI Data Centers as Michigan Project Begins
- 4. US Federal Regulators Order Faster Grid Connections for AI
- 5. Ohio and Virginia Implement Utility Tariffs on AI Data Centers
- 6. Roger McNamee Warns AI Infrastructure Spending Is Unsustainable
- 7. Investors Question AI Spending as Tech Giants Face Cash Flow Pressure
- 8. OpenAI Growth Misses Spark AI Sector Sell-Off
- 9. Big Tech Cloud Revenue Surges on AI Demand
- 10. Nvidia Reports 106% Revenue Growth Amid AI Supply Constraints
- 11. AI Credit Cycle Risks Compare to 2008 Subprime Crisis
- 12. BIS Warns AI Investment Bubble Could Trigger Global Recession
- 13. Treasury Draft Report Warns of Systemic AI Market Bubble