AI Capex Crosses $1 Trillion Next Year. Here Are 2 Industrial Stocks That Will Benefit

The article argues AI-driven data center buildouts will lift power and cooling spending, citing U.S. hyperscaler AI spend that could reach $1T by 2027. Vertiv’s revenue rose from $5.0B (2021) to $10.2B (2025) and adjusted EBITDA from $698M to $2.2B, with backlog more than doubling to $15B at end-2025, and analysts expecting 28% revenue and 38% adjusted EBITDA CAGRs (2025–2028). Quanta’s backlog increased from $19.3B (2021) to $44B (2025) as it expands transmission and data-center power work, with analysts projecting 17% revenue and 19% adjusted EBITDA CAGRs (2025–2028). Valuations are framed as not cheap but still reasonable (Vertiv ~34x, Quanta ~29x adjusted EBITDA) given strong growth tailwinds.

Analysis

This is a second-order capital-expenditure story, not a pure AI-earnings story. The near-term winners are the infrastructure bottleneck owners: power conditioning, cooling, transmission, and substation contractors should enjoy pricing power while hyperscalers absorb the budget shock. The losers are the cash-flow statements of the platform companies that have to fund this buildout; if AI monetization lags capex, multiple compression risk rises for the large cloud names before infrastructure demand even slows.

The market may be underestimating how sticky this demand is for PWR relative to VRT. VRT is more directly tied to server-room spend and therefore more exposed to competition, product substitution, and customer concentration if hyperscalers standardize designs; PWR has a broader utility/grid backlog that should convert over a longer cycle and is less vulnerable to a single OEM design win. That makes PWR the cleaner way to own the multi-year grid constraint theme, while VRT is the higher-beta expression with more valuation risk if order growth normalizes.

Contrarian view: the consensus is treating backlog as if it were earnings, but a lot can break between awards and margin realization—labor tightness, permitting delays, and utility rate cases can stretch conversion by quarters. The main falsifier is a slowdown in hyperscaler capex guidance or evidence that AI buildout shifts from brownfield retrofits to more efficient designs, which would reduce the need for incremental power/cooling intensity. Watch for any deceleration in backlog growth or margin compression in the next two earnings cycles; that would argue the trade is too crowded.

Key Decisions for Investors

• Long PWR vs short VRT on a 3-6 month horizon: prefer the more diversified grid/transmission beneficiary over the higher-multiple direct AI cooling name; target a 15-20% relative spread if AI capex remains strong.
• If buying VRT, do it on a 5-8% pullback and keep sizing modest: the stock is a good thematic expression but is vulnerable to multiple compression if backlog conversion slows or margins disappoint.
• Use XLU as a hedge against the ‘AI spend hurts everyone’ narrative only if hyperscaler capex headlines turn negative; otherwise the cleaner trade is against the platform names, not utilities.
• Watch NVDA/major cloud capex commentary over the next 1-2 earnings cycles; any guide-down in AI infrastructure spend is the clearest falsifier for both VRT and PWR.
• For risk-managed exposure, consider a small long PWR / short a hyperscaler basket (MSFT/AMZN/GOOGL) for 6-12 months to capture the margin transfer from software economics to physical infrastructure.