This article originates from a Tom’s Hardware report that half of planned US data center builds have been delayed or canceled due to power shortages and Chinese component restrictions. an investor had Gemini conduct a Deep Research and obtained a report. This article is a verification analysis of that report, along with extended research into other overlooked beneficiaries.

Part 1: Original Gemini Deep Research Report

The full Gemini Deep Research report.

Scarcity Architecture: Deep Analysis of Strategic Beneficiary Assets Under US Data Center Infrastructure Bottlenecks

Executive Summary

The explosive growth of AI has triggered an unprecedented capital expenditure cycle, with massive capital pouring into hyperscale data center construction. However, the physical expansion of this digital infrastructure is now facing severe systemic hard constraints. The latest industry deep assessments and supply chain tracking data show that approximately half of planned US data center construction projects have been delayed or canceled due to power infrastructure shortages and trade restrictions on Chinese-made electrical components. Tech giants are spending tens of billions of dollars stockpiling advanced AI chips (such as Nvidia GPUs), but these chips are stuck in a situation where they cannot be powered on due to severe electrical equipment shortages and grid capacity limitations.

This report provides a detailed and in-depth analysis of the second- and third-order economic spillover effects behind this “delay and cancel” wave. When the industry bottleneck shifts from front-end semiconductor design and manufacturing to back-end physical infrastructure, pricing power within the industry chain also shifts dramatically. Companies that can provide critical power management, alternative supply chain networks, carbon-free baseload power, and ready-to-use available data center space are capturing unprecedented excess profits and valuation premiums in this historic structural scarcity.

Core Beneficiary Assets Matrix

1. Heavy Electrical & Power Equipment Manufacturers: Domestic and allied supply chain enterprises monopolizing large power transformers, switchgear, liquid cooling systems, and 800 VDC power distribution architectures.
2. “China+1” Strategy & Supply Chain Restructuring Beneficiaries: Mid-tier and precision component manufacturers with Southeast Asian and North American free trade zone production capacity.
3. Regulated Utility Giants & Nuclear Power Operators: Nuclear renaissance companies and grid integration giants that can bypass fragile traditional grids through dedicated PPAs for 24/7 uninterruptible baseload power.
4. Data Center REITs with Absolute Pricing Power: The underlying property holders who possess powered-on ready asset pools, enabling significant rent increases and ultra-long lease terms in a near-stalled new supply environment.

Macro Background & Structural Bottlenecks

Geopolitical Decoupling & Supply Chain Pain: The US data center industry has historically been deeply dependent on Chinese-made electrical components. Over the past four years, prices of key Chinese electrical components imported by the US have doubled. Large power transformer delivery cycles have extended from months to years.

Grid Capacity Saturation: Data centers accounted for 4.4% of US electricity demand in 2023, projected to rise to 6.7-12% by 2028. The core contradiction lies in the severe temporal mismatch: connecting new power plants to the grid takes more than twice as long as building a data center.

Key Beneficiary Assets

Eaton Corporation (ETN): Record Q1 2026 net sales of $7.45B (+17% YoY). Data center orders up ~240%. Total backlog ~$22.8B. Acquired Boyd Thermal for ~$9.55B. Technology written into Nvidia Rubin reference architecture.

Vertiv (VRT): Q4 2025 organic orders grew 252% YoY. Backlog reached $15B. 2026 full-year net sales guidance raised to $13.5-14B. Industry-leading services margins.

GE Vernova (GEV): Q1 2026 new orders of $18.3B (+71% organic). Total backlog $163B. Electrification segment captured ~$2.4B in data center equipment orders in one quarter. However, 90% of gas turbine capacity through 2030 is already locked.

Powell Industries (POWL): Record Q2 FY2026 new orders of $490M (+97% YoY). Backlog $1.8B. Landed a mega-order over $400M — largest in company history. $544.9M cash, zero short-term debt.

Hammond Power Solutions (HMDPF): Q1 2026 revenue $254.1M, significantly beating expectations. Key Mexico manufacturing expansion within USMCA framework.

Fabrinet (FN): World’s largest optical transceiver manufacturer. Thailand-based “China+1” haven. Deeply embedded with Coherent, Innolight, Eoptolink.

Constellation Energy (CEG): 20-year PPA with Microsoft to restart Three Mile Island Unit 1 (~835MW). $1B federal loan from DOE. FERC waiver granted June 1, 2026.

NextEra Energy (NEE) & Southern Company (SO): NEE’s $67B mega-merger with Dominion. SO raising capex plan to $81B, expanding Georgia generation capacity by 70% (95% for data centers).

Equinix (EQIX) & Digital Realty Trust (DLR): EQIX market cap ~$104B, YTD +42%. DLR market cap ~$65B, YTD +28%. Historic low vacancy rates (1-1.6%) driving unprecedented rental pricing power.

Macro Spillover: Inflation & Regulatory Restructuring

Dallas Fed model predicts data center buildout will add 0.04-0.13 percentage points annually to PCE inflation by 2030. In Georgia, residential electricity bills surged 43% from 2023-2025. Regulatory reforms expected to require hyperscalers to bring “behind-the-meter” generation capacity.

Part 2: StocKlaw’s Evaluation & Analysis

The above is the original Gemini Deep Research report. Below is my verification analysis:

Overall Assessment

The report’s skeleton is correct — the structural bottleneck in the AI boom has shifted from “not enough chips” to “not enough power and land,” and pricing power is transferring to infrastructure. However, the report has several notable issues.

Strengths

1. Clear structural scarcity chain — from transformer lead times → Chinese component cutoff → grid saturation → community backlash → rent spikes. More depth than most “AI narrative” reports.
2. Dallas Fed PCE inflation model — the most easily overlooked macro spillover effect. If this chain holds, it impacts the entire asset allocation landscape.
3. Behind-the-meter trend — correctly identifies that data center operators bringing their own gas turbines/battery packs explains the surge in GEV orders.

Problems

⚠️ Valuation Extremely Stretched

The report reads like a “fundamental perfection narrative.” Current market prices have already priced in massive optimism:

⚠️ “The Shovels Have Already Been Priced”

The report talks extensively about “selling shovels” but fails to mention these shovels have already doubled or tripled in price. The market has been trading this narrative since 2023 — the best part may be over.

⚠️ Demand-Side Risk Ignored

The report assumes AI compute demand only goes up. What if AI ROI disappoints and CapEx gets slashed? What if inference cost declines (DeepSeek effect) mean less hardware needed for same compute? Even the original Tom’s Hardware article was about “China lagging 5-10 years in chips,” not “US data center shutdown” — the report may have magnified the negativity.

⚠️ Investment Bank Narrative Traces

“Valuation should converge toward Vertiv (52x)” — classic IB speak. 52x VRT might itself be a bubble; why should ETN chase that?

⚠️ Timing Mismatch

The report discusses rent spikes at EQIX and DLR but ignores that high rates structurally suppress REIT valuations. If the Fed maintains high rates due to data center-driven PCE inflation, these REITs’ financing costs increase.

My Ranking by Certainty

Tier 1: Reasonable Valuation + True Unreplaceable Moat

1. GEV (28x) — Gas turbine capacity locked through 2030. High certainty. Limited upside surprise.
2. CEG (26x, PB 2.9x) — TMI restart has FERC waiver, $1B federal loan, 20-year Microsoft PPA. Nuclear is the most reliable carbon-free baseload for AI.
3. NEE/SO (22-24x) — Normal utility valuation. Data center electricity demand is real incremental growth.

Tier 2: Right Logic, Valuation Needs Digestion 4. ETN (41x) — Best data center power stock. Wait for pullback. 5. EQIX (76x) — Interconnection density is an unassailable moat, but this valuation discounts 10 years of future moat value.

Tier 3: Cyclical, Low Error Tolerance 6. POWL (59x, $11B) — Single orders move the stock. Good for small positions, not long-term holds. 7. FN (61x) — Contract manufacturing. Customer concentration. Low barrier to competition.

Avoid:

• VRT (83x) — 252% order growth ceiling approaching. Post-earnings 12% drop was not an accident. Zero room for error at this valuation.

Summary

This is an excellent industry education piece, not a direct investment recommendation. Its value lies in understanding “power and infrastructure are AI’s new bottleneck.” Current market prices already embed massive optimism — chasing highs requires caution.

My advice: Wait for pullbacks to build GEV and CEG positions. Use ETN for medium-term allocation. VRT and POWL need valuations to return to 50x or below before consideration.

Part 3: Extended Stock Picks Beyond the Report

The report focuses on large-cap leaders. Here are additional beneficiaries discovered through the same logical framework, covering five angles the report missed.

① Engineering & Construction Contractors — “Whoever Builds, I Get Paid”

The report discusses equipment shortages but doesn’t cover who installs this equipment. Half the projects stopping means the remaining half’s installation prices surge.

EME is the most surprising find — PE of just 28x, $38B market cap, good liquidity, data center construction orders booming. Half the valuation of ETN/VRT. Potentially the most undervalued data center infrastructure stock.

② Backup Generators — Every Data Center’s Necessity

The report covers transformers and power distribution but misses that every data center needs dozens of diesel generators for emergency backup. With the behind-the-meter trend, generator demand is shifting from backup to primary.

CMI at 36x PE is reasonable for a century-old industrial. Data centers are incremental, not the whole story.

③ Liquid Cooling — Beyond VRT and ETN/Boyd

NVT at 58x isn’t cheap but has accelerating momentum.

④ Grid Intelligence — The Overlooked “Soft” Infrastructure

When transformers and switchgear are scarce, grid management software becomes the bottleneck.

13x PE for an AI infrastructure stock? Either a value trap or severely undervalued. The question is how much of Itron’s revenue comes from the data center theme. Worth a deep dive.

⑤ Copper — The Most Basic Infrastructure Material

Every data center needs hundreds of tons of copper for wiring and grounding. Transformer windings are copper.

The copper story isn’t just data centers — it’s also EVs and grid modernization. But 37x PE for a cyclical commodity stock isn’t cheap.

Final Ranking (Risk-Adjusted)

Tier 1: True Undervaluation

1. EME (EMCOR) — PE 28x, direct data center construction beneficiary, cheapest in the space
2. HUBB (Hubbell) — PE 29x, industrial electrical equipment with dividend, steady
3. CMI (Cummins) — PE 36x, century-old giant + data center generator growth driver

Tier 2: Small & Interesting 4. ITRI (Itron) — PE 13x, needs thesis verification on data center revenue share

Tier 3: Right Logic, Needs Better Entry 5. NVT (nVent) — PE 58x but accelerating. Wait for pullback to ~45x

Tier 4: Fully Priced In Report’s VRT(83x), PWR(98x), STRL(84x), BE(13712x) — thesis is correct, but the market has run too far ahead

— StocKlaw, June 3, 2026

Data sources: Longbridge Securities, Company filings, Tom’s Hardware, Gemini Deep Research