Markets spent this week pricing the political turn against data centers as a ceiling on artificial intelligence. Less building, less electricity demand, less capital spending, and therefore a smaller and cheaper AI cycle. That reading is wrong, and the price action that followed the midterms has the sign backwards.
A data center moratorium does not subtract load from the American grid. It subtracts the cheapest places to put that load. The compute still gets built, because the people buying it are not price sensitive at the margin, and what changes is the cost and the timeline of every megawatt it consumes. The correct way to read the anti data center vote is as a supply side cost shock to electricity and to compute, arriving at the precise moment the Federal Reserve has the least ability to do anything about it.
What the market actually did, and what it said it was doing
The week produced two completely unrelated selloffs filed under a single label. On Wednesday 4 November, the day after Democrats took both chambers of Congress, Reuters had GE Vernova down 11.5 percent, Constellation Energy down 9.2 percent and Vistra down 15.4 percent. The hyperscalers rose on the same news, on the logic that a slower buildout raises the value of compute capacity that already exists and relieves pressure on capital budgets.
That is not a risk off move. It is a relative value trade, and it embeds a specific forecast: the build shrinks.
Thursday was the broad event. The S&P 500 fell 334.89 points, or 4.4 percent, to 7,353.73, and the Nasdaq Composite had its steepest single session fall in more than a year and a half. Friday reversed part of it, with the Nasdaq up 1.1 percent and the S&P up 0.3 percent, leaving the index down 2.7 percent on the week and still its worst week since April 2025.
Meanwhile the worst performing sector of the week was energy, at roughly 12 percent, close to double the second worst. That had almost nothing to do with artificial intelligence. Crude fell about 8 percent on Wednesday alone as the Gulf war premium unwound, with West Texas Intermediate ending near 90.55 dollars a barrel and down about 13 percent on the week, and Brent down about 16 percent for its worst week since 2020. Two mechanisms, two timeframes, one headline.
The binding constraint moved from Washington to the statehouse
The explanation most widely offered for Thursday was federal AI regulation. That is the one channel which is close to shut. The new Congress does not convene until January, and any bill restricting data centers would need a signature from a Republican president or a two thirds override.
The channel that is wide open is the state one, and it is where the money actually repriced. Texas elected Gina Hinojosa, its first Democratic governor since 1990, on a pledge of an immediate moratorium on data centers until the legislature puts real protections in place for Texans. Wells Fargo puts Texas at about 27 percent of planned American data center gigawatts. Jefferies estimates the proposed moratorium could place 13.3 gigawatts of announced Texas projects at risk.
Goldman Sachs expects Democratic governors in Texas, Georgia and Michigan to impose conditional moratoria pending restrictive legislation. In Pennsylvania, Governor Josh Shapiro has already signed an executive order requiring data center developers to build, bring or buy their own electric capacity.
Here is the detail that should puncture the partisan framing. Texas had already begun slowing data center permitting under its outgoing Republican administration, with the state environmental regulator instructed to stop issuing new permits before anyone voted. The Texas legislature also remains in Republican hands, which limits what a governor can do alone. This is not a story about one party discovering a grievance. It is a story about a grid running out of headroom, and about the price of electricity becoming the most legible cost in American politics.
Why a moratorium makes the build more expensive rather than smaller
The demand side of this market is not rate sensitive or sentiment sensitive in any way that a permitting delay can reach. The four largest hyperscalers have guided to roughly 732.5 billion dollars of capital spending in 2026. Consensus for 2027 is near 934.5 billion dollars, about 28 percent higher, and Goldman Sachs has the figure approaching 1.14 trillion dollars. Alphabet, whose 2026 range was raised to 195 to 205 billion dollars, has told investors capital spending will increase significantly in 2027 without putting a number on it.
Firms spending at that scale are not making a return on capital calculation at the margin. They are making a competitive one. Block a connection in Texas and the load goes into another state's interconnection queue, or the developer builds its own generation on site. Both outcomes cost more and take longer than connecting to an existing system with spare capacity.
That is the mechanism, and it is worth being precise about it. Build, bring or buy is a mandate to add cost, not a mandate to subtract load. It converts a shared grid investment, which spreads fixed costs across every user, into a set of bespoke private generation projects at smaller scale. It fragments the buildout across more jurisdictions, each with its own permitting clock. None of that reduces the number of gigawatts the AI industry intends to consume by 2030. All of it raises the price per gigawatt and pushes delivery to the right.
Voters asked for lower bills and will be sent a cost
The political logic behind the moratorium wave is that data centers are raising household electricity bills, and the evidence for that is real. PJM's independent market monitor attributed 63 percent of the price increase in the 2025 to 2026 capacity auction to data centers, roughly 9.3 billion dollars passed through to customers. IEEFA put that at about 16 dollars a month on the average residential bill in Ohio and 18 dollars in western Maryland. Ohio residential bills rose 53 percent between June 2021 and June 2026. ICF estimates residential rates across PJM could climb 30 to 60 percent by 2030.
The problem is that a moratorium does not retire the load already approved and under construction. It constrains new supply while that load arrives. Restricting generation and transmission additions on the same grid where demand is already contracted is the textbook recipe for higher clearing prices, not lower ones.
The national data already shows it. In the September 2026 CPI, electricity rose 4.4 percent over the year against headline inflation of 3.7 percent, and in the Midwest the electricity index was up 8.8 percent, with the Northeast up 7.1 percent. Electricity is running hot in exactly the regions where the data center politics is hottest.
This is the inflation component monetary policy cannot reach
The Fed is already leaning the other way. The target range has been 3.75 to 4.00 percent since the October 28 meeting, which held rates with two governors dissenting in favor of a hike, and CME pricing on Friday had the odds of a December increase near 73 percent. Friday's October payroll report did nothing to soften that: 165,000 jobs against a consensus near 59,000, unemployment down to 4.1 percent, and 84,000 added to the prior two months in revisions.
And yet the long end has been falling. The thirty year Treasury ended the week at 5.26 percent, down five basis points and down for a fifth consecutive week. The ten year settled at 4.858 percent, about seven basis points lower on the week. The bond market is treating the data center backlash as demand destruction, a disinflationary shock delivered by politics.
I think that is the mistake. A higher policy rate does not commission a transformer, shorten an interconnection queue or persuade a county to approve a substation. If the political settlement on AI is that the industry may build as long as it brings its own power and waits longer for permission, the result is a persistent, supply driven increase in the cost of electricity and of compute, embedded in a CPI component the Fed watches closely and can barely influence. That is the opposite of a fifth consecutive weekly rally in long bonds.
What would change my mind
This is a falsifiable claim, so here is what would break it. If 2027 hyperscaler capital spending guidance comes down in absolute dollars rather than simply relocating, the demand side is more elastic than I have assumed and the deflationary read wins. If the Republican held Texas legislature blocks the moratorium and permitting resumes, the most important state constraint never binds. If electricity CPI decelerates below headline over the next two quarters while contracted load connects, the pass through mechanism is weaker than the PJM numbers imply.
The strongest counter is that moratoria arrive paired with fast tracked generation. If states trade a pause on new connections for accelerated approval of co located gas and nuclear capacity, then build, bring or buy becomes capacity additive and my cost argument collapses into a timing argument. Watch for that pairing in the Texas legislative session. Until it appears, the political turn against data centers is a price increase wearing the costume of a spending cut.
Discussion
Log in to join the discussion