How Technology Risk Is Reshaping Global Data Center Investing

Obsolescence can be a silent killer of data center returns. This note breaks down why certain facilities fall behind and what smart investors and operators are doing to stay ahead.

Jun 18, 2025

Welcome to Global Data Center Hub. Join 1300+ investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.

Executive Summary

Technology risk and obsolescence are now defining threats to data center performance and investor returns.
AI workloads are pushing rack densities from 8–15kW to 50–100kW+, exposing critical limitations in legacy infrastructure.
Cooling, power, and physical architecture built even five years ago are no longer fit for high-density AI clusters.
Operators who fail to pre-design for liquid cooling, modular upgrades, and flexible density are seeing declining tenant retention and IRR compression.
Emerging markets are building smarter, often delivering AI-ready infrastructure with modularity and advanced cooling from the start, but energy access, permitting delays, and supply chain gaps still pose localized risks.
Obsolescence is now a capital planning issue. It’s no longer about how long a facility lasts, it’s about how long it stays competitive.

The Infrastructure That Quietly Fails

What’s Happening

The AI era has changed the rules of data center design and it’s exposing billions in sunk infrastructure that can’t evolve fast enough.

Today’s hyperscale AI clusters require 50–100kW per rack, liquid cooling, and distributed power systems that can scale in real time. In contrast, most legacy facilities were built for densities under 15kW and rely on air cooling and rigid electrical topologies.

Operators face a hard truth: even data centers built in the last decade are at risk of being obsolete before lease-up ends.

Cooling systems that were efficient in 2018 now fail under next-gen GPU loads. Floorplates can’t support heavy racks. Traditional containment layouts limit density segmentation. Older UPS systems are mismatched to the demands of modern high-power draw workloads.

The most alarming part? These assets often still show full uptime and tenant occupancy, right until renewal fails or a retrofit proves economically unviable.

1. Obsolescence Is Already in the IRR

Facilities that can’t support AI workloads are already missing renewals or accepting rent discounts, often quietly, without showing up in headline metrics.

This directly impacts NOI, compresses exit multiples, and creates residual value risk at disposition.

Investors are seeing IRRs fall below threshold not because demand is soft, but because infrastructure can’t deliver what tenants now expect.

And with tenant needs evolving faster than underwriting models, the obsolescence drag is compounding across portfolios.

2. Liquid Cooling Isn’t Optional Anymore

AI workloads are generating thermal loads that traditional air cooling can’t handle.

The latest GPU racks exceed 700W per chip, pushing total rack loads to 50–100kW or more. In that range, standard CRAC units and raised-floor airflow designs aren’t just inefficient, they’re unworkable.

Direct-to-chip cooling, rear-door heat exchangers, and even immersion systems are no longer “advanced options.” They’re baseline requirements for serving next-gen AI clients.

And yet, most facilities built even five years ago lack the physical provisions, chilled water loops, slab strength, piping access, to retrofit without disruptive and costly overhauls.

3. Design Timelines Are Misaligned

Most data center projects still take 24–36 months to move from land acquisition to go-live.

But chip cycles and AI system designs are evolving every 12–18 months. This means many facilities are functionally outdated by the time they open.

Designs optimized for 20kW racks in 2023 may be incompatible with the 80kW workloads demanded by 2026.

Without flexible density zoning, spare capacity for mechanical systems, and modular upgrade paths, new builds risk locking in obsolescence on day one.

4. Hidden CapEx Debt Is Growing

Many portfolios carry an invisible liability: the future capital expenditures required to bring legacy infrastructure up to AI-ready standards.

This “obsolescence debt” builds quietly as operators defer cooling and power upgrades to preserve short-term margins or avoid tenant disruption.

But delaying lifecycle CapEx doesn’t eliminate the need, it compounds the cost.

When the retrofit finally becomes urgent (due to tenant turnover, compliance changes, or performance failures) operators are forced into reactive spending under time pressure. This drives lower ROI, greater disruption risk, and impaired asset value at exit.

Proactive CapEx planning is no longer optional, it’s strategic risk management.

5. Emerging Markets Are Smarter But Still Vulnerable

In many high-growth markets, developers are leapfrogging legacy designs and building AI-ready infrastructure from the ground up.

In places like Navi Mumbai, Lagos, Jakarta, and São Paulo, new builds increasingly include high rack weight tolerance, chilled water distribution, and modular layouts with flexible density zones.

But smart design alone isn’t enough. These markets still face macro risk: permitting delays, energy availability, and supply chain bottlenecks that can delay upgrades or restrict tenant onboarding.

And in some cases, value-engineered “AI-ready” facilities still cut corners, leaving investors exposed to future CapEx surprises.

Billions are being lost to obsolete infrastructure, often in silence.
Here’s what the smartest investors, operators, and governments are doing about it, and how you can apply it.

You’ve reached the premium section of this article.

The full breakdown below includes tactical takeaways for investors, operators, and policymakers.

Upgrade now to unlock the full breakdown

For Investors

The AI infrastructure wave is exposing a fundamental flaw in conventional underwriting: most models assume a 20–25-year asset lifespan, yet today’s compute and thermal dynamics suggest many facilities may be functionally obsolete in 7–10 years or sooner.

This shift demands a rethinking of core investment assumptions:

Discounted Cash Flow (DCF) models should incorporate accelerated CapEx for power and cooling refresh cycles.
Terminal value assumptions must be sensitivity-tested for obsolescence risk, especially for facilities with rack densities <20kW.
Stranded CapEx should be modeled like stranded power, pricing in scenarios where a retrofit fails to attract AI tenants and reduces resale liquidity.
Asset scoring frameworks should now include “AI readiness” metrics: rack weight tolerance, chilled water loop integration, and power distribution flexibility.
Preferred sponsor diligence should prioritize firms with proven track records in modular retrofits, hybrid cooling deployment, and lifecycle CapEx management.

In the near term, expect bifurcation: AI-optimized facilities might trade at higher EBITDA multiples (25x+). Everything else might see compressed valuations, or be excluded or divested from strategic portfolios altogether.

For Operators

Operators who treat obsolescence like an afterthought are already behind.

This is no longer a discussion about Tier ratings or N+1 power. It’s about infrastructure agility and density adaptation.

To stay competitive in the AI era, operators must:

Redesign for 50–100kW rack densities with segmented thermal zones and flexible containment strategies.
Pre-install or provision for liquid cooling, including chilled water distribution, CDUs, and space for immersion tanks or cold plates.
Upgrade electrical topologies to support 415V distribution and redundant feeds that can flex with dynamic load curves.
Model refresh cycles as a standard part of budgeting, not a CapEx exception.
Use digital twins for airflow, load balancing, and real-time optimization of underutilized space, especially when preparing for tenant expansions.

Hyperscalers are already asking for this in their RFPs. If your site can’t deliver, you won’t even get a meeting.

For Policymakers

Technology obsolescence is no longer just a private sector concern. It’s increasingly a national competitiveness issue.

The economic value of AI infrastructure (jobs, innovation, and compute power) is flowing toward regions that can host AI-ready facilities, not just offer incentives.

To remain investable, governments should:

Update permitting frameworks to accommodate liquid cooling, modular blocks, and high-rack weight tolerances.
Streamline transmission approvals and grid upgrades to meet hyperscaler load requirements.
Fast-track permitting for AI retrofits tied to density, ESG, or energy performance upgrades.
Coordinate AI infrastructure strategy with utilities, operators, and global cloud firms.
Create and/or support blended financing tools (green bonds, ESG tax credits, and digital infrastructure loans) for future-proof builds.

Regions that treat obsolescence mitigation as public infrastructure policy will win private capital and tenant demand.

The Bottom Line

Data centers are no longer just steel and concrete, they are dynamic platforms for AI deployment.

In this market, returns will accrue not to the biggest footprint, but to the most adaptive infrastructure.

If your facility isn’t AI-ready, it’s already obsolete.

Global Data Center Hub