EQIX

Drive to the outskirts of a major city — Ashburn, Virginia; Amsterdam; Singapore; Tokyo — and you will find clusters of large, nondescript buildings with massive electrical substations alongside them. These are Equinix data centers.

From the outside, they look industrial — concrete walls, few windows, heavy security perimeters. Inside is a different world. The ground floor is dominated by power infrastructure: utility feeds from the electrical grid, transformers stepping voltage down, UPS (uninterruptible power supply) rooms filled with battery banks, and rows of diesel generators for backup. Above that (or adjacent), the server halls — vast, climate-controlled rooms where rows of server racks stretch from one end of the building to the other, each rack humming with equipment.

Equinix does not own the servers inside the racks. Their customers — hyperscalers, cloud providers, enterprises, financial institutions — bring their own equipment (or have it installed by vendors like Supermicro). Equinix provides the space, the power, the cooling, the physical security, and critically, the network interconnections: the fibre optic cross-connects that link one customer's equipment to another's within the same building or across the Equinix campus.

The xScale program is Equinix's AI-era expansion. These are purpose-built facilities designed for hyperscaler deployments at 50-100+ megawatt scale per building — large enough to house tens of thousands of GPU servers. The xScale buildings have liquid cooling infrastructure designed in from day one, power densities that support 40-100+ kW per rack (compared to 5-15 kW in traditional facilities), and direct interconnection to Equinix's global fabric of 260+ data centers.

Three converging forces are transforming this business. Hyperscalers collectively guided over $300 billion in 2026 capex, and a significant share flows into leased colocation for interconnection-heavy workloads and geographies where hyperscalers lack their own real estate. Power scarcity has become the binding constraint — getting 100 megawatts permitted in Northern Virginia or Amsterdam takes 3-5 years, and Equinix's existing power contracts represent a time advantage no amount of capital can compress. And the emergence of enterprise AI inference is creating a second demand wave beyond hyperscale training: near-metro, low-latency workloads that need Equinix's interconnection density rather than a remote hyperscaler campus. Full year 2025 revenue was approximately $8.7 billion, with Q4 bookings up 42% YoY as AI-related demand accelerated into signed contracts. As a REIT, growth shows in FFO (Funds from Operations) rather than retained earnings.