AI trading’s “overlooked risk”: What if the massive capital expenditure cannot be spent?

```

The story of AI is evolving from “software eating the world” to “hardware getting stuck by the world.”

In the deeply divided American political environment, there are almost no issues that can unite the far-left Senator Bernie Sanders and the far-right Governor Ron DeSantis—except for “curbing data centers.”

This is not only a political spectacle in Washington, but also a cold “physical correction” faced by Wall Street. As Silicon Valley giants wield checkbooks more expensive than the Apollo moon landing project, attempting to sustain the AI boom through stacking computing power, they are hitting a high wall jointly built by the limits of politics and the physical power grid.

Lawmakers in New York State have already proposed a bill to impose at least a three-year moratorium on the construction and operational permits of new data centers. New York is at least the sixth state considering suspending new data center developments.

In summary, from community protests in Florida to sudden regulatory brakes on the Texas power grid, a market-ignored risk is rapidly escalating: If the physical power grid can’t connect and the political environment doesn’t permit, then the tens of billions of dollars that were calculated in valuation models may literally “not be spendable.”

When Sanders and DeSantis “collaborate”

Sanders and DeSantis are in fierce opposition on almost all issues, but in the surge of data centers, they have reached a rare consensus: the brakes must be applied.

This cross-party “united” hostility comes from Americans’ acute pain over the “side effects of AI.” Across the US, the 24-hour low-frequency noise of data centers disturbs nearby communities, huge cooling demands cause local water shortages, electricity bills for residents and small businesses are soaring, and the public protest is growing louder.

The rapid shift in attitude from Florida Governor DeSantis exemplifies this political trend. Just last June (2025), he signed a major tax relief bill, extending data center tax credits from 2027 to 2037. Yet, facing the rising tide of public protest, DeSantis quickly shifted stance.

“We do not want to subsidize technology that will replace human experiences,” DeSantis said at a recent roundtable. He called for the drafting of an “AI Bill of Rights” and supported legislation requiring data centers to pay their full water and electricity costs. He emphasized that local communities should not pay for the expansion of these “richest companies in human history”—“You should not pay one penny more because of this.”

This rhetoric is very similar to Sanders’. Sanders earlier released a report warning that if decisions are made only by profit-focused billionaires in boardrooms, technology will not improve workers’ lives. He explicitly called for Congress to pass a law to pause new data center constructions: “I believe this process must slow down.”

Politically astute lawmakers are following suit. In Arizona, Georgia, Virginia, etc., bills are being proposed to cancel tax incentives or prohibit public-confidential NDAs; in Georgia, Oklahoma, and Vermont, legislators have gone as far as directly proposing moratoriums on new projects, as Sanders suggested.

For tech giants, the era of “red carpet” investment attraction is over.

Can massive capital expenditures be spent?

If political resistance is a “soft constraint,” then the bottleneck of the physical power grid is a deadlier “hard wall.” Wall Street is facing a troubling logical paradox: Does the market really believe the expected $600 billion in capital expenditures projected for 2026 can actually be spent?

According to the latest data, just Microsoft, Meta, Amazon, and Google—the four big tech giants—plan to spend $670 billion this year on AI infrastructure.

In terms of proportion of US GDP, this amount has already surpassed the Apollo moon landing project in the 1960s and the interstate highway system of the 1970s, second only to the Louisiana Purchase in 1803. Amazon alone plans to increase capital expenditure by nearly 60% to $200 billion this year.

The majority of this massive capital will be used to build data centers, which require immense amounts of energy. According to BloombergNEF forecasts, by 2035, data center energy demand will triple, soaring from 34.7 GW in 2024 to 106 GW, equivalent to the power consumption of 80 million households.

The problem is that America’s current power grid simply cannot meet this demand.

This physical constraint has developed into a regulatory crisis in Texas. As the second largest concentration of data centers in the US, only behind Virginia, Texas grid operator ERCOT (Electric Reliability Council of Texas) is putting projects in unprecedented “emergency brake” mode.

ERCOT has proposed re-examining projects totaling about 8.2 GW of power consumption—equivalent to the output of eight conventional nuclear reactors. Notably, this includes many projects that had already been approved.

Currently, ERCOT has launched a review mechanism called “Batch Zero,” planning to review projects in batches to assess their overall impact on the power grid. Meta’s energy project manager, Katie Bell, frankly noted that some projects were submitted 18 months ago and still have not met the standards of “Batch Zero.”

This uncertainty is destroying tech giants’ expansion plans: if the power grid can’t connect, data centers can’t be built; if the data centers can’t be built, the $670 billion budget can’t be spent; if the money can't be spent, the expected growth and commercialization of AI computing power will become a mirage.

Wall Street’s most crowded trade faces a “physical correction”

When the risk of “money that can’t be spent” begins to be priced in, financial markets react sharply. Recently, US stocks experienced the fourth-largest single-day sell-off of “momentum stocks” in the past decade.

Notably, even independent power producers (IPPs) and nuclear power concept stocks, previously seen as “shovel sellers” in the AI boom, haven’t been spared. The previous market logic was “AI’s power shortage is good for power stocks,” but now the logic has turned to: If the power grid is blocked, new electricity demand cannot be realized.

UBS analysis indicates that, worried new loads cannot sign contracts with existing generating capacity, IPP giants like Constellation Energy have sharply dropped—year-to-date (YTD) declines have reached 27%. The market realizes that without physical grid expansion, pure generating capacity is meaningless.

This panic has given rise to “anti-AI trades.” Funds are flowing out of high-beta tech stocks and into defensive sectors like chemicals and regional banks. This is typical “deleveraging” driven by both quant funds and actively managed funds.

Looking at it now, the market must resolve this annoying paradox: either believe the grid can miraculously expand to accommodate that $600 billion in Capex; or admit we’ve hit a physical ceiling. If it’s the latter, it means no grid construction, no capital expenditure, no chip demand—in the end, the valuation bubble of the AI supercycle faces collapse.

At present, with the political alignment between Sanders and DeSantis and ERCOT shutting the door on the physical side, Wall Street seems to be forced to accept the second possibility.

Risk Warning and DisclaimerThe market involves risks; investments require caution. This article does not constitute individual investment advice, nor does it take into account any specific investment goals, financial situation, or needs of particular users. Users should consider whether any opinions, perspectives, or conclusions in this article are suitable for their particular circumstances. Investments based on this article are at your own risk. ```