Data Center Boom Accelerates: 3 Top AI Stocks Averaging 296% Forward EPS Growth

PwC projects global data-center capital expenditure will reach $31.6T through 2050, with a growing share allocated to AI infrastructure. U.S. data-center construction spending has surged 57% year over year to record levels, supporting the view that investors may be underestimating the duration and scale of the AI infrastructure buildout despite concerns over spending durability.

Analysis

The investable implication is not broad "AI capex" exposure but ownership of the physical bottlenecks that convert announced campuses into operating capacity. Vertiv (VRT), Eaton (ETN), GE Vernova (GEV), Quanta Services (PWR), and nVent (NVT) have pricing power where lead times, certification requirements, and field-installation capacity constrain substitution; their revenue should lag GPU orders by roughly 2-6 quarters but carry less risk of compute-price deflation than NVIDIA-linked hardware. The second-order beneficiary is firm power: Constellation (CEG), Vistra (VST), and select regulated utilities with data-center-heavy service territories can monetize rising load, although regulated returns arrive more slowly than merchant-power upside.

The consensus error is extrapolating aggregate construction outlays directly into durable earnings for every infrastructure supplier. Hyperscalers can defer shell construction or redesign server architectures, while grid interconnection delays can shift revenue recognition rather than eliminate demand; this creates a likely pattern of volatile quarterly orders alongside intact 12-24 month backlog conversion. The long-duration projection has little valuation utility today, and premium multiples in VRT, ETN, GEV, CEG, and VST leave downside if quarterly bookings, backlog, or large-project margins miss even while end-demand remains healthy.

Near term, monitor hyperscaler capex guidance and the order/backlog commentary of VRT, ETN, and PWR for evidence that spending is moving from land and buildings into electrical and thermal equipment. Over 1-3 months, a construction slowdown would favor suppliers with replacement/aftermarket exposure over pure project contractors; over 6-18 months, transmission approvals, nuclear recommissioning, and long-term power purchase agreements are the more important catalysts. The thesis is falsified by two consecutive quarters of declining data-center equipment orders, materially shorter lead times, or customer cancellations that reduce backlog rather than merely extend delivery schedules.

Key Decisions for Investors

• Build a 6-12 month basket long VRT, ETN, and GEV, sized modestly given elevated expectations; favor ETN as the lower-volatility core holding and VRT as the higher-beta expression. Add on post-earnings order volatility only if backlog remains growing and management reiterates large-project margin assumptions.
• Pair long ETN / short SOXX over the next 3-6 months as a relative-value AI infrastructure trade: ETN captures electrical-distribution scarcity while SOXX has greater exposure to semiconductor inventory and accelerator pricing normalization. Exit if semiconductor demand broadens beyond AI while ETN book-to-bill falls below 1x for two quarters.
• Use PWR as a 12-18 month grid-buildout exposure, but treat utility interconnection timing as the key risk. Initiate only after confirming transmission backlog growth and avoid adding if labor utilization or project-margin guidance deteriorates.
• Keep CEG and VST on a watch list rather than chase: initiate on power-price or regulatory-driven pullbacks if incremental contracted data-center load is disclosed. The risk/reward turns unfavorable if new capacity additions or regulatory intervention cap merchant-power economics.
• Do not use the long-range aggregate capex forecast as a standalone catalyst. Set alerts for hyperscaler capex-guide cuts, VRT/ETN/PWR backlog revisions, and data-center power-contract announcements; these are the actionable confirmations or disconfirmations over the next two earnings cycles.