Ai Engineering 7 min read

Nasdaq 100 loses $2.3T in a day as voters reject data centers in 40 of 41 ballot measures

The Nasdaq 100 fell 6.3% on November 5 after a Democratic midterm sweep. Governors who ran on pausing construction now oversee about 41% of the planned US data center pipeline, though the selloff had causes beyond the election.

On November 5, two days after the US midterms, the Nasdaq 100 fell 6.3% and erased $2.3 trillion in market value, its worst day since April 2025. The Philadelphia semiconductor index fell 12.3%, its worst day since March 2020. The selling reached Asia on November 6, when South Korea’s exchange imposed a circuit breaker with the KOSPI down as much as 8.7%.

The trigger most often cited is politics. Democrats won Congress and several governorships in the states with the largest data center plans, and local voters rejected the buildout almost everywhere it was on a ballot. For anyone planning around AI compute, the more durable story is power and permits: the constraint on the AI buildout has moved from chips and capital toward local consent. The market move is the loud part, and it had other causes too.

What fell, and what fell less

The damage followed exposure to the physical buildout. According to NBC and CNBC:

GroupNovember 5 move
Memory and storageSanDisk -22%, Micron and Western Digital -17%, Seagate -16%
ChipsAMD -17%, Broadcom -10%, Nvidia about -9%
NeocloudsIren -17%, Nebius -15%, CoreWeave -10%
HyperscalersAlphabet, Amazon, Microsoft and Meta down 2-4%

Nvidia alone lost almost $450 billion of market value. Korea’s Samsung closed down 6.3% and SK Hynix fell more than 10%. CNBC reported that US tech was recovering by Friday morning, so read these as one day’s panic rather than a settled repricing.

The spread is the informative part. The companies that own the demand fell far less than the suppliers and the leveraged renters of compute. Barclays credit strategist Dominique Toublan gave the cleanest account of why: the pushback makes the buildout “pricier rather than stalling it,” and “scarce power favors the biggest balance sheets.” If that is right, friction helps incumbents with cash and hurts neoclouds that need debt-funded projects to energize on schedule.

The ballots were more lopsided than the stock move

Voters in 10 states decided 41 local data center measures on November 3, and 40 went toward limits, on unofficial results compiled from Ballotpedia and local election offices. All 13 questions that would have helped a project failed. The lone exception was Harwood, North Dakota, which rejected a review requirement 75% to 25%. Other trackers use wider definitions and report somewhat different totals, so the exact count may still move.

Ohio supplied 21 of the 41 measures. Sunbury’s charter ban on facilities of 25 MW or more targeted a possible Amazon hyperscale project. Trenton’s ban passed about 80% to 20%. Ashville voters repealed an EdgeConneX deal 82% to 18%. Ohio already had more than 125 local moratoriums before election day.

Two limits on these results matter. Some were advisory, and some cannot reach existing work: Trenton’s ban probably will not touch a four-building AWS campus already under construction, even though it aims to block an 18-building one.

Money did not change the outcome. Build American AI, linked to the Leading the Future super PAC, spent at least $2.54 million on Ohio TV ads promoting data center jobs and investment. An Intercept analysis found AI super PACs put only 5% of their direct spending into contested general election races, having concentrated earlier on primaries. NBC’s election day polling found that more than half of voters think AI will do more harm than good, and at least three-quarters oppose a data center in their own community.

158.6 GW under governors who campaigned on a pause

The number that should concern infrastructure planners came from Cleanview, reported by Bloomberg on November 6: 158.6 GW, about 41% of the planned US pipeline, sits in states whose incoming governors ran on stopping, pausing or extending pauses on construction. Goldman Sachs puts its own estimate at 45%.

StateExposed planned capacity
Texas88.4 GW
Ohio21.9 GW
Georgia20.2 GW
New Mexico11 GW
New York9.9 GW

Texas dominates. Governor-elect Gina Hinojosa called for a moratorium on all approvals until the legislature passes protections, and that would extend a temporary halt on environmental permitting and grid connections that outgoing Governor Greg Abbott imposed through about the end of the year. Wells Fargo estimates each 10% cut to planned Texas capacity would trim Big Tech’s AI infrastructure spending by 1.6%. In Ohio, Amy Acton proposed an immediate moratorium on taking office, and in Georgia, Keisha Lance Bottoms promised a temporary statewide pause.

Cleanview also says OpenAI’s entire planned US pipeline lies in states with a new governor who backed a pause. That leaves its build plans especially exposed, and it arrives while the Financial Times reports it has lowered its proposed valuation.

Keep the scale in proportion. These are plans. The same reporting puts operating US capacity at 62.7 GW, so the pipeline implied by Cleanview’s figures, about 387 GW, is roughly six times everything running today. Aterio counts 3,969 announced US data centers against about 800 under construction. Much of that pipeline was never going to be built on schedule. A pause mostly removes optionality before it removes concrete. It is also far from clear how much governors can do alone: permitting, agency appointments, utility regulation and tax incentives are in reach, while some actions need legislatures.

The election was only part of the selloff

Blaming the ballots alone would miss what was already wrong. Qualcomm issued a weak forecast on November 4, and Arm fell 13% after its own report. Qualcomm’s CEO said “We don’t have a demand issue; it’s a supply issue,” pointing at a memory shortage that is raising prices. Traders still put the odds of a December Fed rate hike around 60%. Meta, Alphabet and Amazon have turned free cash flow negative on AI capex.

Demand signals are softer as well. Ramp data published November 6 shows corporate token spending grew 1.5% across August and September, after 42% from June to July, before rebounding 8.9% in the month to October 23. Ramp’s economist attributes the slowdown to price cuts, competition and cost controls, not weak usage. The same TechCrunch report cites Bloomberg saying Anthropic’s potential IPO valuation has fallen from about $2 trillion to $1.2-1.3 trillion.

Ed Yardeni’s read was that this is “an AI bubble losing some air very, very rapidly”, which he argued lowers the odds of a damaging burst. Dan Ives, a persistent bull, now expects Big Tech capex growth to peak in 2027 rather than 2028 or 2029. Even after the drop, the Nasdaq Composite remains up more than 30% since the 2024 election.

What to watch

January moves in Austin, Columbus and Atlanta. Pledges become policy when the new governors take office. Whether they issue blanket pauses or conditions such as developers paying their own grid costs will decide how much of the 158.6 GW is actually delayed.

Federal bills, with low odds. A Sanders and Ocasio-Cortez moratorium bill could face a veto. Senator Martin Heinrich’s GRID Savings Act, which would stop grid costs being passed to communities, is the narrower proposal to track.

Deal terms becoming the template. Meta president Dina Powell McCormick said building “is going to be harder” and described the model she expects to survive: paying for your own electricity, local labor, and tax surpluses directed to teachers and first responders. Expect that to become the price of entry, and expect it to raise the cost per megawatt.

Where the buildout goes instead. University of Michigan’s Shobita Parthasarathy warned developers may shift to places with less organized opposition. For teams buying compute, the practical question over the next year is whether capacity that was promised for 2027 and 2028 arrives on time, and at what price.

Get Insanely Good at AI

Get Insanely Good at AI

The book for developers who want to understand how AI actually works. LLMs, prompt engineering, RAG, AI agents, and production systems.

Keep Reading