After the Build: The Long Game of AI Infrastructure Operations

By Matt Keefe, Managing Director 

During Goldman Sachs’s recent earnings call, ⁠CEO David Solomon said the US “is in the middle of an AI capex super ​cycle” due to AI infrastructure buildouts.  There are other insiders that suggest we are not in the middle, but rather the early innings of a long capex cycle tied to AI.  Regardless of your view, it’s hard to ignore the tremendous capital being deployed to build the AI infrastructure.  But what happens when the buildout slows down (will it ever be done?)  Smart investors are beginning to ask that question in pursuit of long-term value plays.

The headlines have fixated on the staggering capital expenditure surge fueling the AI boom — $300 billion annually into data centers, chips, and power infrastructure. Tech giants and cloud providers are locked in an arms race to erect fortress-like facilities packed with GPUs and sprawling cooling systems. But there's a critical inflection point that tends to get overlooked: once all this hardware is in the ground and energized, the real economic story shifts from one-time construction spending to perpetual operational economics. The AI CapEx boom is an exciting phase. The decades-long operational phase is where the actual business of keeping artificial intelligence running unfolds.

The scale of this operational burden is staggering. A 100-megawatt data center doesn't simply run itself after the grand opening. It requires around-the-clock technician staffing, preventive maintenance schedules, and constant monitoring of thermal management systems that are already operating at the edge of physical cooling limits. Cooling infrastructure alone — liquid cooling loops, high-density power distribution, thermal sensors — degrades, clogs, and fails with regularity. HVAC units, redundant power supplies, networking hardware, and control systems all have finite lifespans measured in years, not decades. The shift from a $100 million construction project to a $50 million-plus annual operations budget happens quietly but inevitably.

This operational phase opens a massive aftermarket for services and parts. Specialized maintenance contractors will command premium rates for expertise in handling cutting-edge thermal management and power delivery systems that few technicians truly understand. GPU memory controllers fail, and replacement modules don't grow on trees. Networking cards, power supplies, storage controllers, and cooling components will require continuous refurbishment and replacement. A single large facility can consume hundreds of thousands of dollars monthly in replacement parts alone. Specialized data center service providers are already positioning themselves to capture this enormous annual aftermarket opportunity. Every GPU installed today is a future service call, upgrade, or replacement revenue stream for somebody.

Perhaps most consequential is electricity, which transforms from a construction-phase cost into the defining operational expense.  The on-going operations of a fully utilized data center creates an insatiable demand for stable, cheap, and abundant power — sparking a quiet infrastructure revolution in grid upgrades, renewable energy buildout, and nuclear capacity retrofits. It also locks in the geographic advantages of regions with low-cost power and available transmission capacity.

The ripple effects extend into labor markets and supply chains that have been largely invisible in the CapEx boom narrative. Building and staffing thousands of data centers require legions of electricians, HVAC technicians, controls engineers, and operations managers. These aren't one-time construction hires; they're permanent organizational costs. Supply chains for specialty components, maintenance contracts, and spare parts inventory must shift from episodic to continuous. Power providers, already strained by demand growth, face reliability pressures they've never encountered. The maintenance economy that emerges from this infrastructure creates fewer headline-grabbing opportunities than the CapEx boom, but substantially more enduring financial value for operators and service providers.

The AI infrastructure boom will eventually transition from a capital-intensive buildout phase into a permanent operations phase dominated by services, parts replacement, and electricity demand. Annual operational expenses — including maintenance, technical staffing, component replacement, and power — will likely rival or exceed initial construction costs indefinitely. The unglamorous work of keeping massive AI facilities running 24/7 for decades will ultimately generate more durable economic value than the current CapEx race that captures all the attention.  Smart investors have already begun searching for companies that will benefit from the long term operational phase.

© Copyright by Matt Keefe, Managing Director, EdgePoint Capital, merger & acquisition advisors. All rights reserved. Matt can be reached at 216-342-5863