Artificial intelligence is often discussed as a software or semiconductor investment theme. That framing misses a critical layer of the economic value chain: the physical infrastructure required to make AI usable at scale.

This is Post 3 of InSight Financial Planners’ eight-part AI investment series. The series begins with raw materials and power, moves through physical infrastructure and the semiconductor supply chain, and continues into compute and networking, intelligence, applications, and Physical AI. The anchor article, “AI Is Bigger Than AI Stocks: The Eight-Layer Value Chain Every Investor Should Understand,” explains why investors should analyze AI as an interconnected system rather than a narrow group of technology stocks.

This article focuses on the AI Physical Infrastructure layer: transformers, switchgear, cooling systems, construction, data centers, and fiber connectivity.

Why Physical Infrastructure Is Necessary

AI compute cannot operate in the abstract. It must be housed, powered, cooled, secured, and connected.

A large-scale AI facility requires far more than servers. It needs:

• Land with access to reliable electricity and telecommunications networks
• Transformers and switchgear to receive and distribute power
• Uninterruptible power supplies, backup generation, and battery systems
• Cooling equipment capable of managing high-density computing loads
• Buildings engineered for demanding electrical and mechanical requirements
• Fiber and optical connectivity to move data within and between facilities
• Skilled construction, engineering, maintenance, and operations personnel

In this sense, data centers are the factories of the AI economy. They convert electricity, computing equipment, networking capacity, and software into usable digital services. If the facility cannot obtain sufficient power or dissipate the heat generated by its servers, the most advanced chip in the world remains economically unproductive.

The physical infrastructure layer also helps explain why AI capacity cannot be added instantly. Semiconductor production, electrical equipment manufacturing, utility interconnection, permitting, and construction all operate on different timetables. Capital can be committed quickly, but physical capacity often takes years to plan and deliver.

That mismatch creates both opportunity and risk. Companies that supply constrained equipment may gain pricing power and strong order visibility. Developers may benefit from scarce powered land and contracted capacity. However, the same constraints can delay projects, increase costs, and reduce returns on invested capital.

The Main Demand Drivers

Hyperscaler Capital Expenditure

The largest cloud platforms are committing substantial capital to data centers, accelerated computing, networking, and power infrastructure. J.P. Morgan estimates that capital expenditures by the five largest U.S. hyperscalers could reach approximately $697 billion in 2026. The figure includes more than physical infrastructure, but it illustrates the scale of the investment cycle.

IDC reported that global AI infrastructure spending reached $318 billion in 2025 and projected spending of approximately $487 billion for 2026. These figures include servers and other infrastructure categories, meaning they should not be treated as direct revenue forecasts for electrical or construction companies. They do, however, demonstrate the increasing economic importance of the broader infrastructure ecosystem.

The relevant leading indicators for investors include:

• Hyperscaler capital expenditure guidance
• Data-center construction starts and powered land acquisitions
• Electrical equipment backlogs
• Utility interconnection approvals
• Leasing commitments and contracted capacity
• Cooling and mechanical-equipment orders

Data-Center Construction Pipelines

JLL projects that nearly 100 gigawatts of new data-center capacity could be added globally between 2026 and 2030, effectively doubling global capacity. The firm estimates that the real estate and tenant fit-out requirements could involve approximately $3 trillion of investment over that period.

Construction costs are rising as facilities become larger and more technically complex. JLL reports that average global data-center construction costs increased from approximately $7.7 million per megawatt in 2020 to $10.7 million per megawatt in 2025, with a further increase forecast for 2026. AI facilities can require significant additional spending for high-density power distribution, networking, and cooling.

The most valuable development attribute is increasingly speed to power rather than simply available acreage. A site without an executable utility connection may have limited practical value, regardless of its size.

Transformers and Switchgear

Transformers step voltage up or down so electricity can move efficiently through the grid and then be distributed safely inside a facility. Switchgear controls, protects, and isolates electrical circuits. Both are essential to data-center commissioning.

The supply of large transformers and medium-voltage switchgear is constrained by manufacturing capacity, specialized materials, engineering requirements, and long production cycles. J.P. Morgan identifies power availability, supply-chain constraints, and permitting as factors that can materially extend project schedules. JLL also reports that equipment lead times remain significantly longer than pre-2020 levels.

These constraints make electrical equipment a potential bottleneck in the AI buildout. A data-center shell can be complete, and servers can be available, yet the facility may remain idle until the required power equipment is delivered, installed, and commissioned.

Liquid Cooling and Rising Rack Densities

As computing density increases, traditional air cooling becomes less efficient for some AI workloads. Direct-to-chip liquid cooling, coolant distribution units, rear-door heat exchangers, and immersion systems are becoming more relevant as rack power requirements rise.

Liquid cooling does not eliminate the need for conventional heating, ventilation, and air-conditioning systems. Instead, it adds complexity to the mechanical design and operating model. Facilities must manage coolant quality, redundancy, maintenance, leak detection, and heat rejection.

This creates demand for specialized thermal-management providers as well as mechanical and electrical contractors capable of integrating these systems into large facilities.

Fiber and Connectivity

AI training and inference require high-bandwidth, low-latency connections. Data must move between servers within a facility, between buildings on a campus, and across regional networks.

Fiber providers, optical-component manufacturers, network-equipment suppliers, and specialized contractors therefore form another important part of the physical infrastructure layer. Connectivity is not an optional enhancement. Without sufficient network capacity, expensive computing resources cannot operate efficiently as a coordinated cluster.

Major Investment Opportunities

Investors can access this layer through several types of companies. The following examples are representative, not recommendations.

Electrical Equipment Manufacturers

Companies such as Eaton, GE Vernova, Hubbell, Powell Industries, and nVent Electric participate in markets involving power distribution, switchgear, electrical components, enclosures, and grid infrastructure.

These businesses may benefit from strong order activity and pricing, but investors should evaluate backlog quality, margins, manufacturing capacity, geographic exposure, and the extent to which data-center demand contributes to total revenue.

Cooling and Thermal Management

Vertiv, Modine Manufacturing, Johnson Controls, Carrier Global, Trane Technologies, and other industrial companies have exposure to data-center cooling, thermal management, HVAC, and related systems.

Important due-diligence questions include whether a company provides differentiated equipment or competes primarily on installation volume, how much revenue is tied to AI-specific demand, and whether rapid growth is creating execution or working-capital pressure.

Data-Center REITs and Operators

Equinix, Digital Realty, and Iron Mountain provide publicly traded exposure to data-center real estate, colocation, and interconnection services.

The economic characteristics differ from those of equipment manufacturers. Investors must consider occupancy, contracted power, rental escalators, debt levels, development pipelines, customer concentration, and the cost of financing new capacity. A data-center operator with strong demand but excessive leverage may not provide the same risk profile as an equipment manufacturer with a net-cash balance sheet.

Construction and Engineering Contractors

Companies such as Quanta Services, EMCOR Group, and Comfort Systems USA participate in electrical, mechanical, utility, and engineering work connected to data centers and other critical infrastructure.

Contractor analysis requires attention to backlog conversion, labor availability, contract structure, project execution, and exposure to fixed-price contracts. Revenue growth can appear attractive while project-level margins deteriorate if costs rise faster than expected.

Fiber and Optical Connectivity

Corning, Ciena, Coherent, Lumentum, Arista Networks, and other communications companies provide exposure to fiber, optical components, switching, and high-speed connectivity.

The opportunity is tied to sustained data movement and network upgrades, not exclusively to AI. That distinction matters because it can provide broader diversification, while also making the investment case less directly connected to AI spending.

Principal Risks

The physical infrastructure theme has substantial secular support, but it is not risk-free.

Construction Cyclicality

Data-center development can be delayed or canceled when demand forecasts change, financing becomes less available, or local opposition increases. Contractors and suppliers may face sharp changes in order activity after a period of rapid expansion.

Interest Rates and Financing Costs

Data centers require significant upfront capital. Higher interest rates increase the cost of development, refinancing, and corporate debt. They also reduce the value of long-duration infrastructure cash flows and can make marginal projects uneconomic.

Power Availability

Power is often the binding constraint. Grid interconnection delays, transmission limitations, utility capacity, environmental requirements, and local opposition can postpone projects well beyond initial schedules.

Supply Chain and Labor Constraints

Transformers, switchgear, cooling equipment, construction materials, and specialized labor may remain constrained. Shortages can increase costs and create commissioning delays.

Customer Concentration

Many equipment suppliers, contractors, and operators depend on a small number of hyperscale customers. This concentration can produce strong visibility during expansion but significant downside if a major customer reduces capital expenditures or renegotiates commitments.

Potential Overbuild

AI demand is expanding, but the timing and profitability of that demand remain uncertain. If developers build capacity ahead of monetizable workloads, occupancy, pricing, and asset values could weaken. J.P. Morgan notes that long construction timelines can collide with rapidly changing technology cycles and uncertain end-user demand.

Near-Term Buildout, Long-Term Duration

The AI Physical Infrastructure layer is the most visible near-term buildout in the series. Orders for electrical equipment, construction services, cooling systems, data-center capacity, and fiber are already translating into measurable revenue and backlog.

That does not mean the opportunity is limited to the next few quarters. JLL’s projections point to a multi-year expansion in global capacity, while the transition from AI training to inference could create demand for more geographically distributed facilities over time.

The investment distinction is important:

• The near-term case is supported by visible capital expenditure, construction pipelines, equipment backlogs, and contracted data-center demand.
• The long-term case depends on sustained AI adoption, continued cloud growth, inference workloads, and the economic returns generated by those investments.
• The principal uncertainty is whether infrastructure capacity will be built in the right locations, at the right cost, and at a pace aligned with actual demand.

Portfolio Relevance: Avoiding Double-Counted AI Exposure

Physical infrastructure can provide a different form of AI exposure than owning application or semiconductor companies. However, it does not automatically diversify an AI allocation.

A portfolio may appear to hold several independent positions while remaining exposed to the same underlying driver. For example, a data-center REIT, an electrical-equipment manufacturer, a cooling supplier, a contractor, a chip company, and a cloud platform may all depend on continued hyperscaler capital expenditure. If that spending slows, multiple holdings could decline together.

Investors should map exposure across the entire value chain and distinguish between:

• Direct revenue exposure to AI infrastructure
• Broader exposure to industrial, utility, real estate, or communications markets
• Customer concentration and shared capital-spending dependencies
• Balance-sheet sensitivity to interest rates
• Valuation assumptions that already reflect strong future growth

This is the central portfolio lesson of the series: AI is a massive economic value chain, and every layer depends on the layers underneath it. Owning multiple layers can improve participation in the theme, but it can also create accidental concentration and double-counting.

For households with substantial assets, AI exposure should be evaluated alongside liquidity needs, tax considerations, retirement funding, estate objectives, and overall risk capacity. A position that appears modest in isolation may represent a meaningful concentration when viewed across related funds, employer equity, private investments, and individual securities.

The next article will examine the AI Semiconductor Supply Chain, including the materials, equipment, packaging, memory, and manufacturing capabilities required to produce advanced processors. New installments will be available through the InSight Financial Planners Articles and News archive.

Disclosure

This article is provided for informational and educational purposes only and should not be interpreted as investment, tax, or legal advice. References to companies, securities, sectors, or investment themes are illustrative and do not constitute a recommendation, solicitation, or offer to buy or sell any security. Past performance does not guarantee future results. Investing involves risk, including the possible loss of principal.

InSight Financial Planners is a Registered Investment Advisor. Investment decisions should be evaluated in the context of a client’s complete financial circumstances, including objectives, time horizon, liquidity requirements, tax situation, risk tolerance, and existing portfolio exposures. Consult with qualified financial, tax, and legal professionals before implementing any investment strategy. Category: Articles and News.