The Anti-Data Center Wave: Can the AI Trade Survive Without GPU Rental Prices?
TREE NEWS reports: Political headwinds against data centers are reshaping the investment logic behind the AI boom, but the fate of the trade ultimately hinges on an overlooked price signal: GPU rental prices. Following a sweeping Democratic victory in the midterm elections, the market has already reacted sharply, with last week’s selloff underscoring fears that compute supply could tighten. The core question is whether rising GPU rental prices can offset the capital expenditure lost if data center construction is blocked. Early price signals suggest the math may work, potentially driving a structural rotation within the AI trade—away from AI labs, neoclouds, and data center builders toward hyperscalers with large existing compute footprints.
Election Shock: The Political Roots of Tightening Compute Supply
The Democratic landslide has put compute resources at risk. In several districts, elections effectively became referendums on AI, with the largest vote swings occurring in states hosting the densest data center clusters. Yet campaign rhetoric often diverges from governing reality. Data centers are vital to the economy, and how aggressively Democrats will restrict project approvals remains uncertain. The market, however, is not waiting. After last week’s sharp decline, stabilization has been tentative, complicated by a spy scandal engulfing the world’s second-largest company and volatility-control funds still needing to reduce exposure. Whether the calm holds is unclear.
GPU Rental Prices: The Key Gauge for the AI Trade
Macro strategist Simon White of Bloomberg Intelligence notes that once sentiment settles, investors must closely track GPU rental prices—the critical indicator of whether compute supply is contracting relative to demand and thus of how the AI trade will evolve. So far, the post-election uptrend in GPU rental prices shows no clear change. But if new data center projects are halted or approvals delayed, rental prices could accelerate. This would fundamentally alter the economics of the AI trade: favoring hyperscalers with massive existing compute capacity, while hurting AI labs and neoclouds that are effectively short compute, as well as data center builders, equipment suppliers, and power providers—including Nvidia, Eaton, and Vertiv. Since the election, hyperscaler stocks have rebounded, while neoclouds, power suppliers, and semiconductor stocks have generally fallen, confirming this logic.
How Big Is the Supply Shock? Estimating Data Center Disruption
The potential scale of construction delays is significant. Energy research firm Modo Energy estimates that about 40% of announced data center projects could be affected by moratoriums, representing up to 256 gigawatts of compute capacity. However, the firm also believes that by 2035, net compute load growth will only decline modestly—some projects will be delayed but proceed, while others relocate to friendlier jurisdictions, whether other U.S. states or abroad. Modo’s calculations show a net reduction of about 4 GW by 2035, against a planned total exceeding 600 GW; the gap in Texas (ERCOT grid) is expected to be largely offset elsewhere. Of course, forecasts are uncertain, and even if the overall impact is limited, localized compute bottlenecks could emerge in the short term, pushing GPU rental prices higher.
Capex vs. Rental Prices: Can They Offset?
The key question for investors: if data center restrictions shrink capex, can higher compute prices sufficiently compensate? Rough math suggests it is feasible. Epoch AI data indicates total capex per gigawatt of data center capacity is about $37 billion, or roughly $18.5 billion annually over a two-year build. CoreWeave currently generates about $40 billion in annualized compute rental revenue per gigawatt. If GPU rental prices rise 30% (H100 rental prices have already climbed over 50% in the past year), that would add about $12 billion in annual revenue. Thus, a 30% price increase on 1.5 GW of existing capacity would cover the capex loss from 1 GW of new capacity. Given the U.S. has about 30 GW of installed compute, if 10-20% can be quickly repriced, the economics shift from ‘building new compute’ to ‘monetizing existing compute.’ The reverse signal is clear: if capex and GPU rental prices both fall, the thesis is wrong. Otherwise, the worst outcome for the AI trade may be an internal rotation. Even if Anthropic and OpenAI delay IPOs due to compute shortages, the market can likely absorb it, as some economic benefits have already accrued to hyperscalers.
Market Stabilization: Hedge Fund Deleveraging Provides a Buffer
In the near term, two reasons support stabilization after recent volatility. Hedge funds—the main sellers last week—have already deleveraged significantly: overall positions fell nearly 9 percentage points in the past week, and beta-adjusted risk exposure dropped nearly 16 percentage points. Morgan Stanley notes that hedge fund net risk exposure has fallen to a one-year low. Highly leveraged, homogeneous positioning makes hedge funds prone to localized stampedes, but retail and institutional investors—less constrained and slower to decide—have not yet followed en masse. Leveraged ETF flows show retail selling is not yet evident; institutions are unlikely to exit fully without investment committee review, and their tech allocations remain high. Second, recession signals are not pronounced. Macro data shows that deep stock market declines (over 20%) typically occur during recessions, and no such signs are present.
Rotation, Not Termination: The Evolution of the AI Trade
A massive cancellation of data center projects would deal a heavy blow to the economy, but that is not the base case. As analyzed, many projects can relocate to friendlier policy environments. More likely is a structural rotation within the AI trade: hyperscalers benefit from a scarcity premium on existing compute, while AI labs, neoclouds, and the data center construction chain come under pressure. This rotation would narrow the breadth and vitality of the AI trade, but it could also prolong the trade in a narrower, more fragile form.
Key Takeaways for Investors
- Watch GPU rental prices closely: They are the clearest signal of whether compute supply is tightening relative to demand and will dictate the next phase of the AI trade.
- Favor hyperscalers over neoclouds and builders: Companies with large existing compute capacity stand to gain from scarcity premiums, while those short compute or dependent on new construction face headwinds.
- Monitor capex trends: A simultaneous decline in capex and GPU rental prices would invalidate the offset thesis and signal a broader downturn in AI investment.
- Short-term stability likely, but risks remain: Hedge fund deleveraging and absent recession signals support stabilization, yet volatility could persist amid political uncertainty and idiosyncratic shocks.
- Rotation, not termination: The AI trade may continue in a more concentrated form, with hyperscalers as the core beneficiaries.
