Insight and analysis on the data center space from industry thought leaders.
As AI infrastructure grows more complex, companies are rethinking how they acquire, own, and finance assets that operate on dramatically different economic timelines.
The scale of AI infrastructure buildout is often measured in dollars. Gartner forecasts global spending on data centers will reach $653 billion in 2026, a 31.7% increase over last year. At the same time, rising memory costs and supply constraints have pushed server prices up 15% to 20% in recent months. The industry remains focused on securing capacity, acquiring GPUs, and meeting surging demand.
What receives less attention is how AI is changing the economics of infrastructure ownership itself.
A data center building may have a useful life measured in decades. Power and cooling infrastructure may remain productive for 10 to 15 years. GPUs, servers, and other compute assets can follow entirely different refresh cycles. Increasingly, organizations are discovering that treating all those assets the same way no longer reflects operational reality.
This shift in focus – from simply acquiring infrastructure to rethinking how it is owned and financed – reflects a deeper transformation in the data center industry. As AI infrastructure grows more complex, organizations are being forced to align their investment strategies with the unique economic timelines of each asset.
One Infrastructure Investment, Multiple Economic Clocks #
Five years ago, infrastructure ownership was often the default assumption. Companies traditionally purchased hardware with the expectation of long-term ownership. Today, many organizations are becoming more selective about what they own, what they finance differently, and where flexibility creates strategic value.
A useful analogy is homeownership. Most homeowners would not finance a kitchen renovation on the same timeline as their mortgage because the assets have different useful lives. Yet many organizations continue to approach data center investments as though buildings, power systems, cooling equipment, and GPUs all depreciate at the same rate. While this approach simplifies procurement and accounting, it often creates inefficiencies when assets follow different lifecycles.
This shift also reveals a growing divide in how infrastructure buyers approach the market.
Hyperscalers increasingly place equipment orders 10 to 12 months in advance, leveraging their scale and planning horizons to secure pricing and supply. By contrast, newer AI cloud providers often prioritize speed over optimization, acquiring capacity as quickly as possible to meet immediate demand.
The divide between these groups is becoming increasingly pronounced. Some organizations are scaling in a measured way, while others operate in a “hair-on-fire” environment, where delays carry significant business consequences.
Infrastructure Constraints Are Driving New Forms of Innovation #
Power presents another example of how quickly conventional assumptions are changing.
Just a few years ago, discussions about data center expansion often centered on utility constraints and grid availability. Those challenges remain real in many markets, but the industry’s response has been remarkably innovative.
Operators are increasingly adopting independent generation strategies, geothermal technologies, advanced turbine systems, and carbon capture initiatives to reduce reliance on traditional utility timelines.
While these developments do not eliminate power challenges, they suggest the industry’s long-term trajectory may be more adaptable than many assume.
Old Assumptions About Asset Life Are Being Challenged #
The same applies to depreciation and asset life assumptions.
Conventional wisdom once held that GPUs would quickly become obsolete as new generations entered the market. In practice, many organizations are finding that these assets remain productive far longer than expected. Some companies have extended GPU depreciation schedules to seven or even eight years, reflecting both the durability of the hardware and the continued value of prior-generation systems.
That reality highlights an important distinction between the investment thesis and the lending thesis surrounding AI infrastructure. Market commentators often focus on potential bubbles, shifts in future demand, or technological disruption.
Lenders focus on actual performance data, asset utilization, and repayment behavior. From that perspective, many of the assumptions surrounding rapid GPU obsolescence simply have not materialized.
Today’s data center is no longer a single asset operating on one economic timeline. Buildings, power systems, cooling infrastructure, and compute assets each have their own useful lives, depreciation profiles, and ownership considerations. As those differences become more pronounced, the financing structures supporting them are evolving as well.
Organizations that embrace this shift will be better positioned to maintain flexibility, allocate capital efficiently, and adapt to the next phase of AI infrastructure development.