AI Data Center Bottlenecks Shift to Power and Cooling; Scaled Delivery Capability Becomes the Key Differentiator for Suppliers

Overview

Bernstein invited Raj Parihar, who previously worked on data center infrastructure procurement at Meta and Microsoft and has more than 15 years of industry experience, to participate in an expert webinar. The discussion covered procurement sequencing, project timelines, supplier selection, delivery bottlenecks, and architectural evolution for AI data center power and cooling equipment. The core conclusion is that current deployment speed depends more on whether power and cooling infrastructure can be secured in advance and deployed on schedule than purely on compute hardware supply.

Core views

Power planning typically precedes finalization of chip and rack designs, while cooling decisions are highly linked to chip type, rack density, and air- or liquid-cooling approaches. Insufficient grid capacity and transmission and distribution capabilities are prompting hyperscalers to increase behind-the-meter and backup solutions such as gas turbines, gas reciprocating engines, and batteries. At the same time, as campuses expand from hundreds of MW to multiple GW, modular designs that can be replicated across phases, regions, and GPU generations are more attractive than highly customized solutions. Procurement is forming a model that combines dual suppliers, architectural redundancy, and long-term capacity reservations, with suppliers that have global capacity, stable delivery, and complex project management capabilities receiving higher priority.

Analysis framework

The report centers on expert interviews and breaks down the procurement cycle in the sequence of project approval, campus planning, power procurement, rack finalization, cooling procurement, pilot commissioning, and traffic go-live. It also compares grid, behind-the-meter generation, and backup power solutions; evaluates technology trends such as prefabrication, modularization, 800VDC, and liquid cooling; and forms stock-level investment views by combining supplier delivery capabilities with valuation methodologies.

Methodology notes

• Qualitative researchExpert interview

  Using an industry expert with procurement experience at Meta and Microsoft to validate market operating mechanisms

  The interview focused on actual procurement sequencing, equipment lead times, project delays, supplier assessments, and architectural choices. The conclusions reflect expert experience and Bernstein's analytical judgment, rather than a large-sample statistical survey.

• Industry chain analysisProcurement timeline breakdown

  Analyzing project progress based on the dependencies among power, racks, primary cooling, and facility-side cooling

  Power equipment is usually procured relatively early after project initiation, rack architecture is refined around the third quarter, CDUs and cold plates are then finalized, and facility-side cooling solutions are typically confirmed in the fourth quarter.

• Competitive analysisSupplier execution capability comparison

  Evaluating suppliers based on delivery speed, capacity fulfillment, global coverage, project management, and service capabilities

  In an environment of long equipment lead times and multi-GW campus expansion, historical delivery records and the ability to expand capacity across regions are more important than individual product specifications.

• Valuation methodsEV/EBITDA, EV/EBIT, P/E ratio, and sum-of-the-parts valuation

  Selecting relative valuation or sum-of-the-parts methods based on each company's business structure

  VRT, TT, and JCI are valued using EV/EBITDA; CARR is valued using EV/EBIT; ETN is valued using a discounted 2030 P/E multiple; and NVT is valued using a sum-of-the-parts approach.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• VRT

  A core beneficiary of AI data center power and cooling infrastructure, rated Outperform with a target price of $368.

  Strengths

  Rated strongly in delivery, capacity fulfillment, and project management for hyperscale customers, with supplier preference ranked second only to Schneider.

  Weaknesses

  Faces supply chain and staffing challenges and is relatively sensitive to the NVIDIA-related ecosystem and the evolution of liquid cooling and 800VDC technologies.

  Comparison

  The expert believes its overall execution capability is stronger than Eaton and Delta, but its overall assessment is below Schneider.

  Risks

  Improved cooling efficiency reducing demand, slower data center expansion, customers accelerating their shift to custom chips, and failure to adapt in time to DTC or 800VDC technology changes.

• NVT

  A beneficiary related to liquid cooling and electrical connections, rated Outperform with a target price of $220.

  Strengths

  Participates in data center construction through its electrical and fastening business portfolio, with sum-of-the-parts valuation reflecting differences in growth and valuation across businesses.

  Weaknesses

  The CDU production line is still in the expansion stage, and hiring cycles for key roles may constrain execution.

  Comparison

  The report uses sum-of-the-parts valuation rather than a single group valuation multiple.

  Risks

  Commoditization of CDU or OCP solutions faster than expected, CDU production ramp-up below expectations, and delays in hiring for key supply chain positions.

• TT

  A stock related to data center chiller and liquid cooling demand, rated Outperform with a target price of $555.

  Strengths

  Has an established HVAC product base and has the opportunity to benefit from data center facility-side cooling and broader liquid cooling demand.

  Weaknesses

  Volatility in traditional residential, transport, and chiller businesses may offset incremental liquid cooling growth.

  Comparison

  Valuation uses 22x one-year-forward EV/EBITDA after the next twelve months, above levels common for traditional industrial companies, reflecting growth expectations.

  Risks

  Escalation of price-fixing litigation, competitors catching up in liquid cooling innovation, deterioration in residential or transport businesses, and chiller headwinds exceeding liquid cooling benefits.

• JCI

  A stock related to facility-side cooling and building systems, rated Outperform with a target price of $173.

  Strengths

  Has chiller and building infrastructure capabilities and can participate in the facility-side cooling segment of data centers.

  Weaknesses

  The investment thesis depends on organizational lean transformation, data center business growth, and improved operating leverage.

  Comparison

  The report uses 20x one-year-forward EV/EBITDA after the next twelve months for valuation.

  Risks

  Lean transformation failing to take hold effectively across the organization, pressure on the chiller business, and operating leverage below management guidance.

• CARR

  A stock related to HVAC and data center cooling, rated Market-Perform with a target price of $78.

  Strengths

  In addition to data center opportunities, it may also benefit from a U.S. residential and light commercial cycle recovery and European heat pump policies.

  Weaknesses

  Rated lower than other covered stocks, and the contribution from data center liquid cooling still requires further validation.

  Comparison

  The report uses 18x one-year-forward EV/EBIT after the next twelve months for valuation.

  Risks

  Escalation of price-fixing litigation, slowdown in hyperscaler capital spending, and R-410A supply tightening falling short of expectations.

• ETN

  A beneficiary of data center electrification, power distribution, and utility capital spending, rated Outperform with a target price of $534.

  Strengths

  The long-term earnings thesis is supported by structural factors such as data center load growth, electrification, utility investment, and manufacturing reshoring.

  Weaknesses

  Valuation requires strong realization of long-term earnings growth and structural demand.

  Comparison

  The target price is based on a 28x P/E ratio on 2030 EPS and discounted back, corresponding to about 33x 2027 P/E.

  Risks

  Load growth below expectations, disruptions in global data center construction, slowdown in utility capital spending, deceleration in manufacturing reshoring, and delays or cancellations of large projects.

Key data

• Overall deployment cycle24 to 30 monthsFrom project concept approval to actual traffic go-live.
• Pilot, commissioning, and testing cycle12 to 18 monthsOccurs after major equipment procurement and solution finalization.
• Grid interconnection cycle24 to 36 monthsApplies to projects above 50 to 100MW, with constraints more pronounced above 200MW.
• Period of tight utility capacityThrough 2028 to 2029The expert said capacity commitments at many utilities are close to fully allocated.
• Common project delay rateAbout 20%May result in additional capital expenditures and commissioning overruns.
• Potential delay for GW-scale projects6 to 9 monthsDoes not yet include additional delays from regulatory or compliance issues.
• Gas turbine lead time18 to 24 monthsHyperscalers may reserve capacity well in advance.
• Gas turbine costApproximately 2 times grid powerDespite the higher cost, it remains a relatively efficient behind-the-meter generation option.
• Potential 800VDC adoption rateAbout 30% to 40% by 2030Actual progress depends on component supply, standards maturity, and migration from traditional facility-side architectures.
• VRT rating and target priceOutperform, $368Valuation uses 27x one-year-forward EV/EBITDA after the next twelve months.
• NVT rating and target priceOutperform, $220Uses sum-of-the-parts valuation.
• TT, JCI, CARR, and ETN target prices$555, $173, $78, $534TT, JCI, and ETN are Outperform, while CARR is Market-Perform.

Impact & implications

The beneficiaries of AI infrastructure investment will expand beyond GPUs to power distribution, transformers, behind-the-meter generation, backup batteries, CDUs, cold plates, chillers, and modular facilities. Equipment suppliers' order durability may benefit from long construction cycles, dual-supplier configurations, redundant designs, and campus expansions. Infrastructure equipment has a long service life, so even if both suppliers complete deliveries, operators may expand originally planned capacity rather than cancel orders. However, slower capital spending, improvements in technology efficiency, product commoditization, and project delays could still weaken order and earnings realization.

Risks

• Hyperscalers slow AI data center capital spending or cancel large projects.
• Delays in grid interconnection, transmission and distribution construction, regulatory approvals, or equipment delivery postpone project go-live and increase costs.
• Improvements in chip and system energy efficiency reduce the demand for power or cooling equipment per unit of compute.
• Accelerated commoditization of CDUs, cold plates, and open-standard solutions weakens supplier differentiation and pricing power.
• Technology roadmaps shift from current DTC or 800VDC solutions to other architectures, causing mismatches in supplier R&D and capacity.
• Highly customized solutions cause supplier lock-in, premiums, and reduced flexibility for upgrades across GPU generations.
• Long lead times for key equipment such as gas turbines and solid-state transformers, as well as shortages of specialized talent, constrain order conversion.
• Dual-supplier strategies and campus expansions fall short of expectations, leading to cancellations or deferrals of reserved equipment.

What to watch

• Available capacity, interconnection schedules, and transmission and distribution expansion progress at U.S. utilities through 2028 to 2029.
• The proportion of projects above 200MW adopting gas turbines, gas reciprocating engines, and other behind-the-meter generation solutions.
• The pace at which lithium-ion batteries and second-life EV batteries replace diesel backup generators.
• Standardization and supply improvements for 800VDC, Meta/OCP side-mounted architectures, and solid-state transformers.
• Orders, lead times, and capacity expansion for CDUs, cold plates, chillers, and facility-side cooling equipment.
• Delivery performance and supplier share changes for Schneider, Vertiv, Eaton, Delta, ACT, CoolIT, and Boyd.
• Actual adoption of modular and dual-supplier designs as 500MW campuses expand to 1GW and above 2GW.
• Whether GW-scale project delays, capital expenditure overruns, and commissioning progress deteriorate.