How Far Can AI Trading Go Amid the Anti-Data Center Wave? It Depends on GPU Leasing Prices

The political headwinds against data centers are reshaping the logic of AI investment, but the fate of this trading ultimately hinges on a price signal that has received little attention—GPU leasing prices.

According to Bloomberg macro strategist Simon White, after the U.S. midterm elections, the supply of computing power faces tightening risks, and the market has already reacted. Although the market stabilized somewhat after a sharp decline last week, volatility has not yet calmed down. The core question is: If data center construction is hindered, can the rise in GPU leasing prices sufficiently compensate for the losses caused by shrinking capital expenditures?

From the existing price signals in the market, the answer appears to be feasible. The establishment of this logic will drive a structural rotation within AI trading—from reliance on continuously expanding AI laboratories, emerging cloud service providers, and data center construction chains, to hyperscalers that possess a large amount of existing computing power.

Election Impact: The Political Roots of Tightening Computing Supply

The Democratic Party achieved a landslide victory in the midterm elections, which may lead to tighter computing resources. Elections in several districts have essentially turned into referendums on AI, with the states that saw the largest increases in Democratic votes being precisely those with the densest data center layouts.

However, there has historically been a gap between campaign slogans and governing realities. Given the critical importance of data centers to the economy, it remains to be seen how much the Democratic Party will tighten the approval of related projects.

The market did not wait for answers. After last week’s sharp drop, although there was some stabilization, the global second-largest company is embroiled in a spy scandal, and volatility control funds still need to reduce risk exposure to cope with higher volatility, making the sustainability of this stabilization uncertain.

GPU Leasing Prices: The Core Indicator for Judging AI Trading Quality

Bloomberg macro strategist Simon White points out that after the market sentiment settles, investors need to closely track the trend of GPU leasing prices—this is a key indicator for determining whether the supply of computing power is contracting relative to demand, and thus measuring the direction of AI trading evolution.

Currently, there is no significant change in the upward trend of GPU leasing prices post-election. However, if new data center projects are halted or approvals delayed, leasing prices may accelerate their rise.

This will fundamentally change the economic logic of AI trading: It will benefit hyperscalers that already possess large-scale existing computing power, while being detrimental to AI laboratories and emerging cloud service providers that are essentially in a "short" position on computing power, as well as to data center builders, equipment suppliers, and power providers—including Nvidia, Eaton, and Vertiv. Since the election, hyperscaler stock prices have rebounded, while emerging cloud service providers, power suppliers, and semiconductor stocks have generally declined, confirming the aforementioned logic.

How Significant is the Supply Shock? Estimating the Risk of Data Center Interruptions

The potential scale of obstacles to data center construction should not be underestimated. Energy research firm Modo Energy estimates in its post-midterm election assessment report that about 40% of announced data center projects may be impacted by suspension orders, involving a computing capacity of up to 256 gigawatts.

However, the firm also believes that by 2035, the growth of computing loads will only moderately decline—some projects will still advance after delays, while others will move to more policy-friendly regions, whether in other U.S. states or overseas markets. Modo Energy's estimates show that by 2035, the net reduction in computing capacity will be about 4 gigawatts, while its estimated total planned capacity exceeds 600 gigawatts; the gap in Texas (ERCOT grid) is expected to be largely filled by other regions.

Of course, predictions themselves carry errors, and even if the overall impact is limited, short-term local computing bottlenecks cannot be ruled out, which would push GPU leasing prices higher.

Capital Expenditures and Leasing Prices: Can They Form a Hedge?

The core question facing investors is: If data center controls lead to a contraction in capital expenditures, can the rise in computing prices sufficiently compensate for the losses?

Rough calculations indicate that this logic is mathematically feasible. Epoch AI data shows that the total capital expenditure for each gigawatt of data center is about $37 billion, which translates to an average annual investment of about $18.5 billion over a two-year construction cycle. CoreWeave's current annualized revenue from computing leasing is about $40 billion per gigawatt. If GPU leasing prices rise by 30% (the leasing price of Nvidia's H100 chip has increased by more than 50% over the past year), it could bring in an additional $12 billion in revenue annually.

Accordingly, a 30% price increase on 1.5 gigawatts of existing computing power could cover the capital expenditure losses of 1 gigawatt of new computing power. Considering that the U.S. currently has about 30 gigawatts of installed computing capacity, if 10% to 20% of that can be quickly repriced, the economic logic will shift from "building new computing power" to "monetizing existing computing power."

The reverse signal to verify whether this logic holds is clear: if capital expenditures and GPU leasing prices decline simultaneously, the conclusion is incorrect. Otherwise, the worst outcome for AI trading may only be a structural rotation within. Even if Anthropic and OpenAI delay their IPOs due to insufficient computing power, the market may still be able to absorb it, as some economic benefits have already concentrated in hyperscalers.

Recent Market Stabilization: Hedge Fund Deleveraging Provides Buffer

From a short-term perspective, there are two reasons supporting the market's tendency to stabilize after recent volatility.

According to Goldman Sachs data, hedge funds, which were the main sellers last week, have completed a considerable degree of deleveraging: over the past week, the overall position size has decreased by nearly 9 percentage points, and the beta-adjusted risk exposure has decreased by nearly 16 percentage points. Morgan Stanley pointed out that hedge funds' net risk exposure has dropped to a one-year low. The high leverage and high homogeneity of their holdings make hedge funds prone to triggering local sell-offs, but retail and institutional investors, who are less constrained and have slower decision-making rhythms, have not yet followed suit on a large scale—leverage ETF fund flows indicate that retail selling pressure is not yet significant; on the institutional side, the possibility of a complete exit before investment committee review is also low, and institutions still hold a relatively high proportion of tech stocks.

Secondly, there are no obvious signals of an economic recession at present. Macro data shows that significant stock market declines (over 20%) typically occur during economic recessions, and there are currently no such signs.

Rotation Rather Than Termination: The Evolution Path of AI Trading

A large-scale cancellation of data center projects would deal a heavy blow to the economy, but this is not the baseline scenario. As analyzed above, many projects can move to regions with more favorable policy environments.

The more likely outcome is a structural rotation within AI trading: hyperscalers benefit from the scarcity premium of existing computing power, while AI laboratories, emerging cloud service providers, and the data center construction chain face pressure. This rotation may weaken the breadth and vitality of AI trading, but it could also continue this trading in a narrower and more fragile form.