Data Centers Don't Just Need Power -- They Need Cooling. Trane and Carrier Sell It.

Trane and Carrier are seeing rising cooling demand from AI hyperscalers, supporting multi-year growth expectations. Trane’s backlog surged 70% year over year to a record $12.1B (2Q26) and analysts expect 2025–2028 revenue/EPS CAGRs of 10%/16%, helped by data-center liquid-cooling and modular plant solutions; Carrier’s data-center revenue is expected to rise ~50% in 2026 and backlog grew 40% YoY to $8B (2Q26). However, both trade at ~30x this year’s earnings (historically high multiples), so the article is constructive on Trane vs. Carrier but warns against chasing valuation.

Analysis

This is less an “AI beneficiary” story than a thermal bottleneck trade. The market is paying for the option value that hyperscaler capex stays elevated and that cooling remains a constrained input, but at ~30x earnings the bar is now execution, not narrative. The near-term upside is mostly sentiment-driven over the next 1-3 months if bookings, backlog conversion, and margin mix continue to beat; the risk is that project timing slips quickly if power interconnects, permitting, or data-center capex pauses slow deployment.

TT looks higher quality because its exposure is more directly tied to custom applied systems and next-generation liquid cooling, so it should capture more of the value chain than a broader HVAC mix. CARR’s larger residential and light-commercial exposure makes it more sensitive to rate-driven housing weakness, so if the AI cycle cools it has a second headwind from the cyclical side of the book. In other words, the market may be overpaying for “AI adjacency” while underestimating how much of these earnings can still be re-rated by housing, pricing, and service mix.

The contrarian miss is that hyperscalers will try to dual-source and standardize thermal systems once designs mature, which can compress margins and shift economics toward EPCs and integrators rather than OEMs. Over 6-18 months, the key falsifier is any slowdown in data-center backlog growth or a mix shift toward lower-margin equipment/service work. If that happens, both names can de-rate even if unit shipments stay healthy, because the multiple has already priced in a durable scarcity premium.

Key Decisions for Investors

• Prefer TT over CARR on a relative-value basis: long TT / short CARR for 6-12 months, targeting a modest multiple gap widening if data-center mix keeps outperforming and residential remains soft. Falsifier: CARR’s data-center growth outpaces TT or housing re-accelerates meaningfully.
• Do not chase outright longs at current multiples; wait for a 10-15% pullback or a quarter of slower backlog conversion before adding exposure. Risk/reward is better on weakness than on headline momentum.
• For investors seeking pure AI infrastructure beta, favor NVDA over thermal-adjacent industrials; the cooling thesis is a second-order beneficiary and more vulnerable to commoditization. Use TT/CARR only as complementary exposure, not core AI positions.
• Set a watch item on upcoming bookings/backlog and gross margin commentary: any sequential slowdown in data-center order growth is a sell signal, especially if management leans on backlog rather than new demand.
• If TT and CARR remain bid into earnings, consider a covered-call overlay rather than initiating fresh directional longs; implied premium is likely cheaper than paying full multiple for the next leg of upside.