On November 3, polling stations in Ohio. In addition to governor and U.S. senator, the ballot also included 21 local referendums on data centers. Across the entire United States that day, there were 41 such referendums in total, and Ohio alone accounted for more than half. The result was 40 to 1, with almost all opposed.

On the same day, Democrat Hinojosa won election in Texas, and her platform stated: pause new data center construction.

Voters made their stance clear with their ballots. The next day, Wall Street made its stance clear with stock prices.

Microsoft rose steadily after the open and closed up 6.68% at $555.80, a record high. Amazon rose as much as 7.8% intraday and closed up 6.12%. Alphabet rose 3.18%, and Apple rose 5.31%. Several tech giants together added nearly $1 trillion in market value in a single day.

On the other side, emerging cloud providers that started out by renting out computing power (known in the industry as Neoclouds) and mining companies that pivoted to data centers were all in the red: CoreWeave fell 9.90%, IREN fell 11.24%, Nebius fell 6.67%, Applied Digital fell 6.74%, Hut 8 fell 23%, and Core Scientific fell 10%.

The broader market was up. The Dow rose 3.03%, its best trading day since April 9, 2025; the S&P 500 rose 1.58%, the Nasdaq rose 1.44%, and the VIX fell 20%. The 10-year U.S. Treasury yield fell 5.8 basis points to 4.894%.

Three layers of market explanation

The reasons given by the market roughly fall into three layers. The first two explain why the giants rose, and the third explains why the emerging cloud providers fell.

Higher barriers make what is already held more valuable

Needham analyst Laura Martin believes that higher regulatory barriers will entrench the position of existing players while raising the risk and return requirements for new entrants. She also mentioned a phenomenon: on the matter of "it is time for capex to come down," investors and politicians have stood on the same side for the first time, though for different reasons.

Baird's Michael Antonelli put it more bluntly to The Wall Street Journal: these companies "already have computing power, already have data centers. Once you pull up the drawbridge, you are essentially handing them a moat for free."

B. Riley's Art Hogan explained the same thing from an accounting perspective: the big companies already have data centers in hand, and since there will be no more overbuilding going forward, these assets are naturally more valuable.

So, data centers used to be built by whoever had the money, but now they increasingly resemble a license that can only be obtained through approval.

Forced to spend less, their cash flow actually looks better

Robert Pavlik of Dakota Wealth said the market's reaction suggests the AI infrastructure boom may be ending; but if that means cloud giants are no longer spending astronomical sums to expand computing power, their free cash flow could be expected to turn positive again. He also had a line repeatedly quoted by Chinese media: "What's putting the brakes on their spending is the government, not the companies themselves."

Jim Cramer, on CNBC's morning show, called these companies "the incumbents." He said the market believes this will "curb the astonishing spending of Amazon, Microsoft, Meta, and Alphabet," and that this "is seen as a good thing, because now there is spending discipline, even if that discipline is imposed by voters."

There is a set of numbers backing this up: the big companies' capital expenditure plans for this year exceed $740 billion, versus about $400 billion last year. Free cash flow at Amazon, Alphabet, and Meta has already turned negative, meaning they are spending more than they earn from operations, and the gap must be filled through debt issuance and financing. Only Microsoft remains an exception, still with $18.6 billion in free cash flow during the same period.

Emerging cloud providers expanding through borrowing will struggle once money tightens

Cramer warned everyone on X to "be careful with companies like Nebius and IREN," and said it even more bluntly on television: these companies are "the ones who can't get marginal funding." He used natural gas pipelines during Biden's term as an analogy: the government didn't ban you from building, it just made sure no one would be willing to lend you money to build anymore.

The evidence at this layer is the most abundant.

CoreWeave is the most typical representative of this round of "borrowing to build data centers." Part of Nebius's construction funding came from a $3.5 billion, 364-day loan from Citi; Citi cut its price target from $324 to $265 that same day. IREN's capital expenditure guidance for this fiscal year is $25 billion to $30 billion, and its share count expanded by 48% over the past year. Applied Digital still has 800 megawatts of capacity awaiting financing.

Valuations are the same. IREN's forward price-to-sales ratio on Monday was 3.1 times, versus a three-year average of 5.2 times; Nebius was 5.4 times, versus a five-year average of 11 times, and it reached as high as 32 times in October 2025.

Both are seeing capital expenditure decline, but the big players are saving cash while emerging cloud providers are facing an inability to borrow.

But new data centers won't disappear entirely

In fact, Texas had already started tightening. On August 3, Republican Governor Abbott ordered a pause on new data centers connecting to the state power grid and required an audit of all data centers waiting in ERCOT's interconnection queue. At the time, ERCOT had received interconnection applications totaling 474 gigawatts, roughly five times its historical peak load, with about 90% coming from data centers. BloombergNEF estimates the pause involves about 49.8 gigawatts of projects, roughly 20% of planned U.S. data center capacity, potentially affecting $8 billion in rental revenue by Q1 2027.

Hinojosa simply took it one step further. And her office later clarified that the pause targets new permits, with projects already under construction unaffected.

At the federal level, major action is unlikely in the short term: the new Congress doesn't take office until January, and any bill restricting data centers would still need the president's signature or a two-thirds majority in both chambers to override a veto.

Second, projects under construction are largely unaffected.

According to Aterio, 86% of capacity slated for delivery in 2027 is already under construction; for 2028, that figure is 34%. U.S. capacity under construction exceeds 80 gigawatts, with Texas alone at around 20 gigawatts, roughly matching the "20.5 gigawatts" circulating online.

TrendForce takes a more conservative view: most projects scheduled for completion in 2027 have already broken ground, and the ones truly at risk of delay are those slated for 2028 and beyond.

BMO's energy infrastructure team put it plainly in its report that day: "Existing orders and projects should still proceed as normal, because underlying demand is still growing." They also cautioned that Digital Realty's development risk would rise, but it would be the biggest winner on pricing.

Bernstein ran the numbers on Monday: of planned U.S. data center capacity, 244 gigawatts, or about 56%, is exposed to political risk, with Texas accounting for about a quarter of that. But they believe the risk is overstated, because developers can absorb more of the power and grid upgrade costs themselves, and those costs remain manageable.

In fact, new cloud providers also have ready capacity on hand.

IREN has three campuses in Texas totaling 2.75 gigawatts, half of its roughly 5 gigawatts of power. It has already signed $4 billion in annualized revenue contracts this fiscal year, with the corresponding capacity coming online this year. The contracted revenue falls in fiscal 2027, with the real risk in fiscal 2028.

Applied Digital signed leases spanning five campuses, with revenue of approximately $36 billion over the initial lease term. CEO Wes Cummins said at the October earnings call: "Any new restrictions introduced elsewhere, in our view, will only make what we already have harder to replicate."

Looked at another way, he is effectively acknowledging that the value of his company's assets depends on how hard they are for others to replicate.

Existing data centers are like annuities, future ones like options

Putting the above together, we can understand this from a clearer angle.

Data centers already built and generating revenue are like an annuity, steadily bringing in income every year. Those to be built in the future are like an option—there are returns only if they are completed, and nothing if they are not.

In the valuations of the big tech companies, the annuity portion is the bulk, including existing cloud businesses, advertising and subscriptions; the option portion is the smaller part, namely the growth brought by future additional computing capacity. Over the past year, the market has actually not been buying into this option portion: for every extra dollar of capital expenditure by the big companies, there is more pressure on their stock prices, because investors do not believe this money can be earned back. Now the option has become harder to exercise, but the money that would originally have been spent on it has also been saved. The annuity is unchanged, and with one subtraction and one addition, the net effect is actually positive. This is the reason Microsoft, Amazon and Alphabet rose at the same time.

Emerging cloud vendors are exactly the opposite: their valuations are almost entirely options. Their annuity portion is very thin, and some are even negative: interest is already being paid, while revenue has not yet ramped up. With the option blocked, the foundation of their valuation moves as well. CoreWeave fell nearly 10% in a day, and IREN fell 11%, not because their data centers suddenly became worthless, but because their stock prices had priced in too much expectation of "continuously opening new campuses next year, the year after, and the year after that."

This also explains a phenomenon that seems strange at first glance: IREN genuinely has 2.75 GW in Texas, while Nebius has not a single site in Texas—its U.S. footprint is in New Jersey, Missouri and Pennsylvania—yet the two still fell together. What the market sold off was less Texas exposure than the expansion story.

Demand is still there, only the payer has changed

On November 4, some companies still signed large orders as usual.

Riot leased its entire Corsicana campus in Texas to an unnamed frontier AI lab: 910 megawatts, a 15-year lease term, an initial contract value of approximately $28 billion, rising 3% annually, and with two five-year renewal options, a total value of approximately $54 billion. Riot rose 9.63% that day.

The structure of this deal is worth a closer look. Riot is responsible for the land and power infrastructure, delivering electricity all the way to the interface of the compute modules; total development investment is approximately $5.5 billion. The compute modules are purchased, owned, and operated by the tenants themselves, and electricity costs, property taxes, and insurance are all borne by the tenants. On a converted basis, Riot invested only about $6 million per megawatt, whereas the contract IREN signed with Microsoft requires over $40 million per megawatt.

In other words, whoever puts up the money to build the compute capacity captures the bulk of the rent; those who only provide land and power earn only the land and power money, but take on far less risk.

On the same day, TeraWulf disclosed that Google signed three credit support agreements for Anthropic at the Justified campus in Hawesville, Kentucky: three buildings, approximately 401 megawatts of IT load, a 20-year lease term, an initial contract value of approximately $19 billion, with Fluor as the general contractor. Note that what Google is guaranteeing is the rent, not the construction — the construction money and risk remain with TeraWulf. This is probably also why TeraWulf, despite landing a major deal, closed down 6.62% that day.

Also on that day, Alphabet announced an additional $15 billion in Malaysia to expand AI data centers and cloud operations.

What to Watch Next

Over the next six months, there are five time points that can test this logic.

November 5, after U.S. market close (early morning Beijing time on the 6th), IREN releases its fiscal 2027 first-quarter report. Focus on the share count as of September 30, the data center financing arrangements, and how management talks about Texas.

Next week, CoreWeave releases its third-quarter report. Citi has already cut its target price from $159 to $128 but maintained its Buy rating. Focus on whether there are new contracts signed and on financing costs.

November 12, Nebius releases its third-quarter report. Its annualized revenue at the end of June was $3 billion, with a target of $7 billion to $9 billion by year-end. If that growth rate cannot be sustained, what it faces will be more than just a Texas problem.

By the end of this year, the Texas audit conclusions come out. The audit ordered by Abbott is expected to be completed within the year, and it will determine whether the $8 billion in rent risk in the first quarter of 2027 becomes reality.

From January to May 2027, the Texas legislature meets. Whether the moratorium gets written into law or is merely the starting point for negotiations will be clear by then.

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia