Are Data Centers the Backbone of AI's Future?

The servers never sleep. Somewhere right now, a GPU cluster is running inference on a large language model, a training job is consuming more electricity than a small town, and a hyperscaler is quietly signing another land lease for a campus that doesn't exist yet. The AI boom didn't just create demand for data centers — it fundamentally redefined what a data center needs to be.

For infrastructure investors and developers, that distinction matters enormously.

What Data Centers Actually Are (and Why the Definition Is Shifting)

Strip away the marketing language, and a data center is straightforward: a facility housing computing hardware, power infrastructure, and cooling systems that process, store, and transmit data. What's less straightforward is that "data center" now describes facilities with almost nothing in common except the name.

A colocation facility serving regional enterprises operates on entirely different economics than a hyperscale campus built for a single cloud tenant. An edge data center the size of a shipping container solving latency problems for autonomous vehicles has almost nothing structurally in common with a 500MW AI training campus. The category has exploded in both scale and specificity.

The facilities being built today for AI workloads would have been considered power plants a decade ago — the compute just happens to be inside.

That matters for anyone underwriting land, securing permits, or structuring acquisition deals. The specs that made a data center viable in 2015 — 5-10 MW of critical IT load, traditional air cooling, standard utility interconnect — are increasingly irrelevant to the buyers and tenants driving today's market.

The AI-Data Center Connection Is Not Metaphorical

When people say AI runs on data centers, they're describing a hard physical dependency. Training a frontier AI model requires coordinated computation across thousands of accelerators running continuously for weeks or months. That computation generates heat. That heat requires cooling. The cooling requires water and power. The power requires grid capacity, substations, and increasingly, on-site generation.

A single Nvidia H100 GPU draws roughly 700 watts. A cluster of 10,000 of them — not unusual for serious AI training infrastructure — pulls 7 megawatts from the facility. Add cooling overhead, and you're looking at a power usage effectiveness (PUE) load that pushes total facility draw significantly higher. For context, the average American home uses about 1.2 kilowatts. That 10,000-GPU cluster consumes as much power as roughly 5,800 homes, continuously.

AI inference — running a trained model to generate outputs — is actually where the sustained data center demand lives, because it never stops.

Training happens in bursts. Inference runs 24/7 every time someone uses ChatGPT, generates an image, gets a recommendation, or triggers an automated decision in an enterprise system. The inference market is enormous and growing faster than most forecasts anticipated, and it's exactly the workload that data center operators are now designing their facilities around.

Where Capital Is Flowing

Investment in data center infrastructure has moved from significant to staggering. Major hyperscalers — Microsoft, Google, Amazon, Meta — have collectively announced hundreds of billions in capital expenditure commitments oriented around AI infrastructure. These aren't aspirational budget line items. They're land acquisitions, utility agreements, and construction contracts.

The upstream effect on real estate and infrastructure is substantial. Sites with existing power access, fiber connectivity, and favorable permitting environments have become scarce. Developers who locked in large land positions near transmission infrastructure three or four years ago are now sitting on assets with dramatically different valuations.

What's driving urgency on the investment side isn't just AI adoption — it's the recognition that lead times are brutal. Utility interconnection queues in major markets run 3-5 years. Permitting and construction for a large campus takes 18-36 months minimum. The data center capacity that needs to exist in 2028 has to be in the ground now.

Secondary and tertiary markets are absorbing demand that primary markets can no longer accommodate. Places like the Carolinas, the Mountain West, and parts of the Midwest are seeing serious developer attention precisely because they offer available land, utility capacity, and lower regulatory friction — even if they lack the legacy network density of Northern Virginia or Silicon Valley.

Mergers, Acquisitions, and the Consolidation Playing Out in Real Time

The data center sector is consolidating, and the deals reflect a strategic logic that goes beyond simple scale. When large operators acquire smaller competitors or regional platforms, they're typically buying something specific: permitted capacity, utility relationships, experienced operating teams, or geographic positions that would take years to replicate organically.

Recent M&A activity in the sector has involved not just traditional data center operators but infrastructure funds, sovereign wealth capital, and REITs that have restructured their portfolios to weight digital infrastructure more heavily. The buyer universe has expanded significantly, which has compressed cap rates and pushed valuations to levels that would have seemed implausible five years ago.

What's interesting about technology mergers in this space is that the strategic value often has little to do with existing cash flows — it's about optionality and queue position.

A facility with 50MW of approved capacity and a signed utility agreement is worth far more than its current revenue suggests because that approved capacity represents years of development work that a buyer cannot shortcut. Acquirers are effectively paying for time.

On the disposition side, the story is equally telling. Enterprises that built private data centers in the 2000s and early 2010s are selling or subleasing those assets at an accelerating rate — not because the facilities lack value, but because maintaining them pulls capital and operational attention away from core business. The beneficiaries are operators who can absorb those assets, upgrade them, and deploy them for AI-adjacent workloads.

What Comes Next

Liquid cooling is the near-term inflection point most operators are grappling with. Traditional air cooling hits physical limits at the power densities AI hardware demands. Direct liquid cooling — running coolant directly to the chip — is no longer a niche engineering curiosity; it's a requirement for high-density AI deployments, and it's forcing facility redesigns at scale.

The data centers being designed today are increasingly purpose-built for specific computational tasks rather than general-purpose hosting. That specialization changes the risk profile for investors: higher barriers to entry, longer useful asset lives, but also less flexibility if tenant requirements shift.

The AI-data center relationship is self-reinforcing in a way that most infrastructure cycles aren't — AI tools are now being used to optimize data center operations, cooling efficiency, and power management, which in turn enables more capable AI deployments.

For developers and investors watching this market, the non-obvious insight is this: proximity to power is now more valuable than proximity to population. The traditional calculus of siting data centers near end users is being disrupted by latency-tolerant AI training workloads that simply need electrons and fiber — wherever those can be sourced cheaply and reliably. That reordering of site selection criteria is quietly reshaping which land assets are worth holding and which infrastructure bets pay off.

The facilities that process tomorrow's AI breakthroughs are being sited, permitted, and financed right now. The developers and investors who understand the technical requirements — not just the financial structures — are the ones positioned to capture that value.

Explore our marketplace for investment opportunities in data centers today!

[INTERNAL LINK: data center investment]

[INTERNAL LINK: AI infrastructure trends]

[INTERNAL LINK: liquid cooling technology]