ThinkPatternGet the app
Perspective
BUSINESS · SEP 17, 2026

Washington Kept AI Off the Electric Bill. The Cost Moved to the Price of Money.

Washington has shut the public out of AI's costs in every venue it regulates, and the private debt that filled the gap is quietly bringing the bill back to households.

The doctrine that has actually governed the AI buildout has its purest statement in a Wisconsin rate case. In May, the state's utility commission decided what Microsoft and the data-center developer Vantage would pay for their projects there, ruling that they cover the full cost of the generators and transmission lines serving them [1]. The reasoning fit in one sentence.

Existing Wisconsin customers should not pay a single cent to subsidize the service of data centers. — Public Service Commission of Wisconsin

Wisconsin was early, not unusual. The same refusal has since been recorded in nearly every venue that touches the machines' costs: the White House ruling out federal loan backstops last November [2], the president demanding in January that the big technology companies building data centers pay their own way [3], and PJM — the grid operator for the mid-Atlantic — extending a wholesale price cap its filings describe as designed to shift grid-upgrade costs from homes to large technology loads [4]. May brought a wave of states, Oklahoma to Florida, each finding its own wording for the same rule [1][5]. June brought the Federal Energy Regulatory Commission, ordering six grid operators to connect data centers faster on the condition that the data centers pay the full cost of the upgrades [6]. And the White House pledge — under which Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI promise to finance the power infrastructure for their own projects — is being extended to governors and third-party operators [7]. For a genre not known for follow-through, the execution has been remarkable. Each order did what its text said, and the doctrine carries assent running from Oklahoma conservatives to the NAACP [3]. The system is now essentially complete. What it produced is arriving this month, in the credit market. Start with the refusal that gets the least attention, because it settled where the money would come from. In November, OpenAI's chief financial officer, Sarah Friar, floated federal backstops and loan guarantees to bring her industry's borrowing costs down [2]. The answer came from the president's AI adviser, David Sacks, and has governed ever since.

We believe that governments should not pick winners or losers, and that taxpayers should not bail out companies that make bad business decisions or otherwise lose in the market. — Sam Altman

Sam Altman fell in line, agreeing taxpayers should not rescue companies from their own bad bets, and OpenAI redirected its lobbying toward tax credits for data centers [2]. The refusals were strategy, not thrift. As the White House expanded the ratepayer pledge this summer, its stated aim was to accelerate the buildout while avoiding backlash from voters worried about rising energy costs [7]. FERC's June orders — show-cause proceedings demanding the six major grid operators appear and justify themselves — put the prevention of cost-shifting to ratepayers among their five reform pillars [8]. To the investors financing the same buildout, the orders made a different promise.

I know that Americans across the country are concerned about affordability, and so are we. — Federal Energy Regulatory Commission

Put the two promises side by side and the doctrine is whole: the household shielded, the investor reassured, the machines fed. What the orders never say is what the companies will pay with. Obliged to fund their own plants, substations and transmission lines while buying the computers at the same time, they have turned to the one pool of money large enough for both: the bond market. The scale shows up in one company's books, published the same week as its newest project. Meta announced a one-gigawatt data center in El Paso this week, a partnership with BlackRock, on top of a construction program that has nearly consumed the cash its operations generate [9].

$784M Meta's free cash flow last quarter, against $60.8 billion in quarterly revenue — 2026 capital-spending guidance: $130–145 billion [9]

Meta is the biggest spender but no longer the outlier: most of the hyperscalers — Amazon, Google, Microsoft, Meta and the other continent-spanning cloud operators — now spend more than their operations bring in, and the industry's capital spending is approaching a trillion dollars this year [10]. A company that spends more than it earns has one place left to go. The borrowing that followed is remaking the market it entered. Bank of America projects the hyperscalers will account for 9 percent of gross investment-grade bond supply this year — investment-grade being the safest tier of corporate debt — up from 7 percent last year and just 1 to 3 percent from 2020 to 2024 [11]. For the first half of the decade the hyperscalers were a rounding error in it. Chase's market letter now treats their paper as competing with Treasuries themselves for investors' money [10]. That paper does not sit in its own corner of the market. It competes with every other borrower of similar quality for the same savings. Josh Farberow of BNP Paribas put the mechanism in one line in August.

All high-quality credit competes with hyperscalers for capital. — Josh Farberow

The consequence, in his telling, is a convergence that could reprice the entire investment-grade market [12]. The repricing already shows in places with no data center in sight. Credit default swaps — insurance-like contracts that pay out if a borrower defaults, and the cleanest gauge the market has of perceived risk — have widened more than 10 percent since the end of 2025 for LVMH, the luxury group; Sanofi, the drugmaker; and BAE Systems, the defense contractor. None of them builds anything for the machines, and all of them are now priced as if they share a market with the companies that do [12]. One complication deserves its own line. Amazon sold $53.8 billion of bonds in March, the largest corporate sale on record, and investors bid more than double that [13]. JPMorgan's John Servidea drew the boundary the sale makes visible.

Notwithstanding just how volatile markets are right now, for the right companies – the truly great companies and credit stories – there is still a tremendous amount of capital available. — John Servidea

Capital stays abundant for the strongest credits. The bite lands on everyone sharing a market with them. The deepest reach is the one that touches every household: the long end of the Treasury market, trader's shorthand for the slow, long-maturing debt that anchors mortgages and corporate borrowing alike. The 30-year yield hit about 5.28 percent in late July, its highest level since 2007, with the Federal Reserve holding steady [14]. The market analyst Casey Sprake ties part of the move to the top five AI firms' borrowing [14].

the bond yield curve is actually doing the the work for the Fed in that — Casey Sprake

The rout has made headlines all summer. The mood beneath it, in her account, is something other than alarm.

the market is being sort of punch drunk around anything AI CapEx spend. — Casey Sprake

The bluntest warning has come from outside the country. The European Central Bank said this month that big tech's AI debt could push up borrowing costs for all sectors, with potential spillover into sovereign bonds, and that the high ratings behind the paper may rest on growth assumptions that do not stand the test of time [15]. Frankfurt has no stake in Oklahoma's rate cases. Its warning is about the arithmetic of supply. Every venue Washington can order, it has now ordered. The one that remains takes no orders — the bond market only prices, and there is no show-cause process for a yield. So the household that was spared the substation bill is meeting the machines anyway, at the mortgage desk and in the spreads on borrowers with no connection to them. The aim, all along, was a fast buildout and a quiet electric bill [7]. The pass-through is an effect analysts and central banks describe, not an outcome anyone ordered. And the next time the 30-year Treasury climbs, the explanations will run to the Fed, the deficit, the usual nerves. One contributor will be quieter: a gigawatt of compute going up in El Paso, financed without a cent of public credit, and priced into the rates everyone else pays to borrow [9].


Sources
  1. 1. US States Implement New Power Tariffs for Data Centers
  2. 2. Trump Administration Rejects OpenAI Requests for Federal AI Bailouts
  3. 3. Trump Pressures Tech Firms to Offset AI Data Center Costs
  4. 4. PJM Extends Electricity Price Cap to Save Customers $45 Billion
  5. 5. Utilities Face Legal Challenges Over Data Center Power Costs
  6. 6. US Officials Move to Prevent Data Centers from Raising Utility Rates
  7. 7. Trump Administration Expands Ratepayer Protection Pledge for AI Power
  8. 8. FERC Orders Six Grid Operators to Reform Large Load Access
  9. 9. Meta Announces Texas Data Center to Expand AI Compute Business
  10. 10. Hyperscalers Issue Massive Debt to Fund AI Infrastructure
  11. 11. Hyperscalers to Reach 9% of Investment-Grade Bond Supply
  12. 12. Big Tech AI Debt Sales Drive Up Global Credit Risk
  13. 13. Amazon.com Inc. Raises Record $53.8 Billion for AI Infrastructure
  14. 14. AI Capital Spending Drives US 30-Year Bond Yields Higher
  15. 15. European Central Bank Warns US Tech AI Debt Risks

Keep reading in the app

The full perspective, free in the app.

Download on the App StoreComing soonGoogle Play