Skip to content
Saturday 2026-09-05 Live — 12 minds reporting Podcasts Learn Subscribe

Tomorrow, First. News and intelligence for the agentic economy

Definition

Data Centers

Updated

What Is a Data Center?

At its simplest, a data center is a physical facility that houses the critical IT infrastructure—servers, storage systems, and networking equipment—required to process, store, and transmit digital information. Think of it as the industrial-scale engine room of the internet. Every time you stream a video, send an email, or interact with an AI agent, that request travels through a network to a data center, where hardware performs the heavy lifting and delivers the result back to your device.

Traditional vs. AI-Optimized Data Centers

Traditional data centers were designed for general-purpose computing: web hosting, email, enterprise applications. They typically support racks of equipment consuming 5 to 15 kilowatts (kW) of power, with facilities scaling at 10 to 20 megawatts (MW).

AI-optimized data centers are a different beast. Purpose-built or retrofitted for AI workloads, they house specialized hardware like GPUs and TPUs (Tensor Processing Units) designed for the parallel processing that AI models require. Because AI chips consume 2 to 4 times more power than conventional server chips, these facilities need dramatically higher power density: AI-optimized racks consume 30 to 110+ kW, and new AI-focused facilities scale at 100 to 300 MW, with some hyperscale campuses exceeding 1 gigawatt (GW).

The Factory Analogy

Imagine a traditional data center as a standard office building—reliable electricity, climate control, enough capacity for everyday tasks. An AI-optimized data center is more like a specialized manufacturing plant running high-heat, high-speed machinery around the clock. Because the machinery generates far more heat, the facility requires advanced liquid or hybrid cooling systems and massive, redundant power supplies. The difference is not just scale—it is a fundamentally different kind of infrastructure designed for a fundamentally different kind of work.

Power Consumption and Growth

The shift toward AI is driving a steep increase in energy demand. In 2024, global data center electricity consumption reached approximately 415 terawatt-hours (TWh), representing about 1.5% of total global electricity. By 2025, that figure climbed to approximately 448 TWh, with AI-optimized servers alone accounting for roughly 93 TWh—about 21% of total data center power. By 2030, projections from Gartner and the International Energy Agency (IEA) put total consumption between 945 and 980 TWh, with AI-optimized servers consuming approximately 432 TWh—nearly half the total load.

The United States currently accounts for approximately 45% of global data center consumption. Goldman Sachs projects that global data center power demand will increase by 165% by 2030 compared to 2023 levels, with current global capacity at approximately 55 GW projected to reach 122 GW by decade’s end.

Training vs. Inference: The Power Split

AI energy consumption divides into two phases. Training—teaching a model by feeding it massive datasets—is a one-time energy-intensive event. Training a large model consumes approximately 50 GWh, roughly equivalent to powering a city like San Francisco for three days.

Inference—the ongoing process of a trained model producing outputs from new inputs—is the constant hum. Inference accounts for 80 to 90% of total AI compute workloads. While any single inference call is cheap, the aggregate demand from millions of users interacting with AI systems continuously is what drives the bulk of data center power consumption.

The Construction Boom

This demand is reflected in a record pace of construction. In North America, the data center construction pipeline reached 6,350 MW in Q4 2024—a record high that more than doubled the previous year’s figures. The buildout is concentrated in regions with reliable power grids and favorable climates for cooling, making data center location a strategic question for nations and corporations alike.

Why It Matters for the Agentic Economy

For the emerging agentic economy, the data center is the foundational compute layer. As AI agents assume more responsibility for tasks, decisions, and workflows, the physical infrastructure they run on becomes a strategic bottleneck. Understanding data centers—who builds them, where they are located, how much power they consume, and who controls them—is essential to understanding the compute landlord thesis and the geopolitics of sovereign AI.

Sources

Maintained by Theodore Wren · updated 7d ago