Data centre infrastructure is the fastest-growing segment of the global technology supply chain. As artificial intelligence, cloud computing, and digital services demand more computational capacity, the physical facilities that house, power, and cool that compute are in unprecedented demand.
The global data centre infrastructure market exceeded $250 billion in annual capital expenditure in 2024, with hyperscalers (AWS, Microsoft, Google, Meta) alone committing over $200 billion in annual capex. The sector encompasses three main categories: hyperscale data centre operators, colocation providers (Equinix, Digital Realty), and infrastructure equipment vendors (Vertiv, Super Micro Computer).
Data Centre Infrastructure Business Models
Hyperscale Data Centre Operations
Amazon Web Services, Microsoft Azure, Google Cloud, and Meta build and operate their own data centres globally. These facilities are classified as: Availability Zones (redundant clusters of data centres), regions (groups of zones), and edge points-of-presence. Capital expenditure ranges from $1B–$5B+ per campus; useful life is 15–20 years.
Revenue from the hyperscale operator perspective is captured through cloud services (IaaS, PaaS, SaaS) priced per compute-hour or subscription — not by selling infrastructure capacity as a standalone business.
Colocation (Colo) Services
Equinix and Digital Realty lease space, power, and cooling within their data centres to enterprises and cloud providers who bring their own servers. Revenue = square footage × monthly recurring rent (typically $200–$500/kW/month for premium interconnection). Colocation provides the physical layer without operating the compute.
Equinix’s interconnection business — the ability for companies to connect directly to dozens of network providers and cloud on-ramps within the same building — is a particularly sticky and high-margin offering.
Data Centre Equipment (Power and Cooling)
Vertiv designs and manufactures the critical infrastructure inside data centres: uninterruptible power supplies (UPS), thermal management (computer room air conditioners, cooling distribution), and power distribution units (PDUs). Revenue is driven by data centre build-out volume; the shift to higher-density AI compute (NVIDIA GPUs draw 10–30× more power per rack than CPU servers) is creating demand for entirely new power and liquid cooling infrastructure.
AI Server Manufacturing
Super Micro Computer (Supermicro) manufactures high-density server platforms optimised for NVIDIA GPU deployment. Supermicro’s rack-scale solutions — complete GPU-optimised server racks with integrated cooling — benefit from being early partners with NVIDIA in delivering the complete AI compute stack. Revenue grew dramatically in FY2024 driven by AI infrastructure demand.
Revenue Models Compared
| Model | Revenue Basis | Gross Margin |
|---|---|---|
| Colocation (Equinix) | kW capacity × monthly rate | 45–55% |
| Critical power/cooling (Vertiv) | Equipment units × price | 25–35% |
| AI server assembly (Supermicro) | Server racks × ASP | 10–15% |
| PDU and UPS products | Capital equipment sales | 30–40% |
| Hyperscale data centre (as % of cloud revenue) | Cloud service revenue allocation | N/A (internal cost centre) |
Key Companies in Data Centre Infrastructure
- Vertiv — critical data centre power and cooling infrastructure; UPS systems, thermal management, power distribution; major beneficiary of AI infrastructure build-out; 2021 IPO via SPAC, now significantly outperforming
- Super Micro Computer — high-density server manufacturing; NVIDIA GPU platforms; rack-scale AI solutions; rapid revenue growth with some accounting controversy
Key Metrics for Data Centre Infrastructure Companies
Revenue Growth and Order Backlog
Data centre infrastructure companies are experiencing demand surges driven by AI capex. Order backlog provides forward revenue visibility; Vertiv’s backlog grew from $2B to $7B+ between 2022 and 2024 as hyperscalers and AI companies accelerated build-out.
Gross Margin and Margin Expansion
Vertiv’s gross margin improvement (from 22% to 35%+) is the single most important fundamental story in the sector — demonstrating that elevated demand allows price increases that flow directly to margin. Supermicro’s gross margins are structurally lower (hardware assembly vs design) but benefit from volume scale.
Power Density per Rack
The shift from 8–15 kW per rack (traditional CPU servers) to 40–100+ kW per rack (AI GPU clusters) is the key technical driver of equipment replacement demand. Higher power density requires liquid cooling (direct liquid cooling, immersion cooling) instead of air cooling — creating a new product category where traditional data centre cooling vendors must adapt or lose market share.
Free Cash Flow Conversion
As demand surges, data centre equipment companies are investing in working capital and supply chain to meet orders. FCF conversion may temporarily lag earnings; but as the backlog normalises and inventory is worked through, FCF should accelerate.
The AI Infrastructure Super-Cycle
AI data centres differ fundamentally from traditional cloud data centres in their infrastructure requirements:
- Power: A 100-MW AI campus can require as much electricity as a small city. Power procurement — through long-term PPAs, on-site generation, or nuclear agreements — is the new constraint on hyperscaler build plans.
- Cooling: NVIDIA H100 GPUs draw 700W each; a rack of 8 GPUs draws 5.6kW. A 1,000-rack AI cluster draws 5.6 MW — far beyond what air cooling can manage. Liquid cooling (cold plates, rear-door heat exchangers, direct liquid cooling) is rapidly becoming standard.
- Physical space: Permitting, constructing, and energising a large data centre takes 3–5 years. The physical build pipeline is now the primary bottleneck on AI compute supply, not chip manufacturing.
Vertiv, Eaton, and Schneider Electric are the primary beneficiaries of this infrastructure upgrade cycle.
Key Comparisons
Related Glossary Terms
- Gross Margin — Vertiv’s margin expansion is the key investment thesis
- Free Cash Flow — working capital-adjusted FCF growth as demand normalises
- Capital Expenditure — hyperscaler capex drives equipment demand
- Return on Invested Capital — colocation colo economics depend on ROIC exceeding cost of capital