The data center industry is experiencing a generational boom, fueled by a relentless surge in demand for Artificial Intelligence (AI) and cloud computing infrastructure. Billions of dollars are pouring into the sector, with developers racing to build the specialized, power-hungry warehouses needed to host the world’s digital brains.
For global investors, the opportunity appears undeniable. However, this period of explosive growth carries echoes of past infrastructure manias—the telecom and cable frenzy of the late 1990s and early 2000s, where speculative investment led to catastrophic oversupply and financial collapse. Elements of this infrastructure build out are reminiscent of the utility industry attempting to build power plants to meet electric demand, where nearby residents protested “not in my backyard (NIMBY)” and held up plant permit approval which added significantly to cost. The current data center investment cycle also closely resembles early independent power producer (IPP)-Power Purchase Agreement (PPA) arrangements, many of which experienced financial difficulty/default.
This article, guided by expertise from institutions like Jones Lang LaSalle Inc., McKinsey, and the Federal Reserve, cuts through the noise. We analyze the rewards driving the current data center gold rush, challenge the narrative with critical risks—from technological obsolescence and private debt leverage to long-term power contract defaults—and ask: Are we on the precipice of a bust, or is this time truly different?
The AI Engine: Why Data Center Demand May be Different than the Telephone and Cable Catastrophe of the 1990s
The demand for data centers today is not merely evolutionary; it is exponential. The primary engine driving this paradigm shift is the rise of Generative AI.
For decades, the standard rack in a data center used about 7 to 10 kilowatts (kW) of power. A rack is a physical structure that holds multiple pieces of information technology equipment in a vertical stack. Today, a single high-density rack hosting advanced AI chips, such as those from NVIDIA, can require 70 to 100 kW, and sometimes significantly more [1]. This isn’t a modest increase—it’s a three-to-ten-fold leap in energy density and capacity requirement.
This shift has redefined what a “data center” even is:
- From Storage to Computation: Traditional facilities focused on storage and basic processing. Modern AI data centers are highly specialized, high-density computing engines designed for parallel processing, demanding massive power and cutting-edge cooling systems.
- The Hyperscale Mandate: Major hyperscalers (such as Amazon, Google, Microsoft) are locked in an infrastructure arms race, collectively committing hundreds of billions to build facilities that can handle these extreme loads [2]. This creates what appears to be a reliable, massive anchor tenant base for investors in the sector.
- The investment figures reflect this unique demand: An estimated $170 billion in asset value will need to secure development or permanent financing in 2025 alone [3]. Investment is accelerating across the full stack—from powered shells and land parcels to specialized cooling and power systems.
- The investment thesis is that investors gain access to the foundational infrastructure of the 21st-century economy, offering long-term, inflation-protected lease revenues backed by the most creditworthy companies in the world. We challenge several key elements of this investment thesis below.
Key Characteristics of a Hyperscale Data Center
Hyperscale facilities, as defined by IBM, are often operated by major cloud providers (called Hyperscalers like Amazon, Google, and Microsoft), are distinguished by:
Massive Scale
While there’s no single official definition, a hyperscale data center is generally considered to contain at least 5,000 servers and occupy at least 10,000 square feet of physical space, often encompassing millions of square feet.
Extreme Scalability
The architecture is designed to grow by adding more resources (servers, storage, networking) easily and quickly—a process known as horizontal scaling (scaling out).
Automation
They rely heavily on sophisticated automation and orchestration software to manage and monitor their vast infrastructure with minimal human intervention, which is crucial for efficiency and agility.
Efficiency
The facilities are custom engineered for optimal performance and energy efficiency, often using advanced cooling techniques and reporting significantly lower Power Usage Effectiveness (PUE) metrics than traditional data centers. While the PUE is less than traditional data centers, electric demand is huge and growing.
Resilience and Redundancy
Multiple layers of redundancy (backup systems, replicated data, geographically distributed sites) are built in to ensure continuous operation, high availability, and fault tolerance.
In short, hyperscale is all about building an infrastructure so big and so flexible that it can handle the internet’s largest workloads—like massive social media traffic, streaming video, AI processing, and global cloud computing services—all while maintaining high performance and efficiency.
Are We Repeating the Telecom, Cable and Independent Power Producer Boom/Bust? (The Bubble Risk)
Skepticism is warranted. Market history is littered with infrastructure booms that end in busts. One salient comparison includes the late-1990s fiber-optic cable build out by the telecom and cable industries.
During that era, telecom upstarts and established giants speculatively plunged millions of miles of fiber-optic cable into the ground based on the promise of explosive internet growth. Companies like Global Crossing and WorldCom built vast, interconnected networks that eventually sat unused for years after the dot-com crash, leading to bankruptcies and massive investment losses [4].
The critical question for investors today is: Is the AI data center gold rush destined for the same fate?
Three Key Differences Proposed by Analysts Today
While history offers a warning, the current landscape may exhibit crucial differentiators in the present environment, which could change over time:
Cash Position and Stability
The 1990s boom was led by highly leveraged, speculative telecom companies. When demand projections failed to materialize, their weak balance sheets collapsed.
Today, the primary drivers of data center demand are the world’s most dominant and cash-rich technology companies—the hyperscalers. These companies generate unprecedented free cash flows that dwarf the capital raised by the fiber-optic builders of the past. Their capital expenditure is strategic, based on direct, internal usage, not purely speculative oversupply [5]. Analysts assume that this cash-rich condition will continue into the future.
Immediate, Quantifiable Demand
The fiber glut was the result of building a supply far in excess of immediate access to the internet (which was still reliant on dial-up, DSL, and early cable).
Today, AI models are operational and globally accessible now. The demand for specialized AI hardware (GPUs) and the power to run them is already creating power scarcity and long development timelines. This is not a projected demand 10 years out; it is a current supply bottleneck [6].
Asset Utility after a Crash
Even in the event of an economic or AI slowdown, the physical assets—the data center shells, the power infrastructure, the real estate—are significantly more fungible and useful than the “dark fiber” cables of the past. They remain purpose-built structures in prime locations with access to both power and connectivity. The underlying infrastructure has intrinsic, long-term utility.
While a bust remains a tail risk, the consensus among analysts is that the sheer financial strength of the anchor tenants and the non-speculative nature of the immediate AI demand mitigate the systemic oversupply risk that plagued the telecom sector. Hemispheres caution investors against irrational exuberance. Changes in Technology, Environment, Regulation, Demand, Credit Markets, PPA non-performance by either the Power Generator (as in the boom/bust cycle in the 1980s), the data center owners or financial difficulty at the anchor tenant level can all negatively affect investor returns.
The Threat of Obsolescence and Stranded Assets
Even if the overall demand holds, not all data center assets are created equal. The relentless pace of technology innovation creates a distinct risk of stranded assets—investments in infrastructure that prematurely lose value [7].
Technological Obsolescence: The Cooling Divide
The single greatest driver of obsolescence is the shift in cooling technology:
- The Problem: The extreme heat generated by AI chips cannot be reliably managed by traditional air cooling systems.
- The Solution: New facilities are being built with liquid cooling infrastructure (e.g., immersion cooling) as the default [8]. Demand for water to cool these newer systems is vast. This heightens regulatory and environmental risks.
- The Risk: Older, air-cooled facilities, even if well-maintained, face expensive retrofits or, worse, become functionally obsolete for the most profitable AI workloads. For investors, this risk, on the face of it, sits with the data center owner, not the tenant (who manages the servers). In the 1980s PPA, if the Independent Power Producer had a mismatch between their costs and their revenue that resulted in a margin squeeze (which can occur by any change in market driven factors where the contract terms do not allow increased costs to be passed through to buyer), performance default was not uncommon.
Environmental and Regulatory Obsolescence
Datacenters are facing increasing scrutiny over their massive consumption of power and water [9].
- Power Scarcity: Many jurisdictions are experiencing power grid bottlenecks and extended timelines for building transmission lines. Centers in constrained zones risk operational limits or being unable to expand capacity.
- Water Risk: Data centers use large amounts of water for cooling. In regions experiencing drought, local community push back and future water-use regulations could strand assets or significantly raise operational costs, creating a permanent drag on asset value.
- Compliance: Stricter “climate-neutral” mandates (like those emerging in Europe) threaten to pressure older, less energy-efficient facilities to update or cease operating, potentially leading to the premature closure of some investments [10].
The Debt-Fueled Arms Race: Private Credit and Default Risk
The data center build out relies not only on large amounts of capital but also on significant debt financing. This brings the risks associated with the rapidly expanding Private Credit market into the equation.
The Private Credit Factor
Private debt funds are eager lenders in the data center space, providing financing for both land acquisition and construction. However, compared to public market loans, private credit often carries higher risk profiles:
- Higher Leverage: Private credit borrowers often operate with higher leverage ratios (debt-to-earnings) [11].
- Floating-Rate Debt: The predominant use of floating-rate debt significantly amplifies interest burdens during periods of high or rising interest rates. This is especially true for companies whose cash flows cannot grow fast enough to service the ballooning debt.
- Credit Quality Indicators: The rise of Payment-In-Kind (PIK) arrangements—where interest is paid not in cash, but by issuing more debt—is a worrying indicator of declining credit quality in some segments of the private lending market [12].
Potential for Increased Credit Default
While overall default rates in private credit have been relatively low, the data suggests two key vulnerabilities:
- Lower Recovery Rates: In the event of a default, private credit loans have historically exhibited lower recovery rates (higher Loss Given Default, or LGD) compared to traditional syndicated loans [13]. This means investors stand to lose a larger percentage of their principal if a borrower fails.
- Systemic Interconnectedness: Massive debt volumes interconnect private equity, banks, and institutional investors such as pension funds. A significant wave of defaults in the data center sector could spill over and amplify systemic risk across the broader financial system [14].
Power Purchase Agreements (PPAs): The Long-Term Contractual Risk
A dependable power supply, secured through Power Purchase Agreements (PPAs), ensures a data center’s reliability. These are long-term contracts that fix the price and terms of electricity delivery.
The Risk in the Contract
Historically, the risk of defaults often stemmed from contracts that, due to market deregulation or unforeseen price changes, became uneconomical for one party.
In the data center space, the immediate risk is counterparty credit risk [15]. PPA generators finance their construction based on the long-term revenue stream guaranteed by the buyer (the data center owner or tenant).
- If the data center owner/tenant fails: Credit rating is of paramount importance. If a smaller developer or highly leveraged tenant files for bankruptcy, the event immediately compromises the PPA’s revenue stream and can trigger financial distress for the power generator and its lenders.
- Contract Mismatch: Power generators often require a 20-year PPA term to secure financing, but many data center operators prefer 5- to 10-year agreements to maintain flexibility for technological shifts [16]. Contract renewal will undoubtedly result in renegotiated terms. A change in terms could drive poor economics for one party or the other, ultimately resulting in non-performance.
- Transmission risk: A data center reliant on electricity derived from the grid will not be as reliable as a proximate, dedicated power source. Residential, commercial and industrial customers all receive power from the grid. In high demand scenarios, power outages would adversely impact data center reliability. Furthermore, gaining regulatory approval to build new transmission lines is very difficult and time intensive, taking years/decades to achieve.
The Mitigation: Nuclear and Dedicated Power
In a major development, hyperscalers are actively moving to mitigate transmission and PPA risk by bypassing the grid entirely. Companies like Microsoft and Amazon are signing dedicated PPAs with nuclear power plants located nearby (e.g., in Pennsylvania) [17].
This strategy addresses the most profound power risk: it secures high-capacity, 24/7 continuous power at a predictable price, bypassing grid congestion and volatility. This innovation suggests that the most large players recognize and are actively engineering solutions to insulate themselves from traditional power risks.
Conclusion
The AI data center sector presents a classic risk-reward equation in three key areas:
Technological Obsolescence: The speed of AI innovation will rapidly render non-liquid-cooled and non-sustainable facilities functionally obsolete. The pace of technological advancement is undeniable. To the extent newer technology makes current data structure infrastructure obsolete, investor returns could be adversely affected.
Financial Leverage: High debt loads in the private credit market pose a threat to smaller or more aggressively financed players, with low recovery rates compounding the loss.
Contractual Exposure: Long-term power agreements are critical, and their risk is only as strong as the credit quality of the underlying buyer.
Hemispheres Investment Management (HIM)
HIM is a wealth manager with a global investment management focus (domestic and international investments in the same portfolio). Because of HIM’s global investment experience, we are uniquely qualified to meet and benefit from this secular trend toward a polycentric investment world. Our team of seasoned professionals each has over 35 years of experience in research, strategy development, and management of investment portfolios, including deep proficiency in U.S., international, and emerging markets. Hemispheres can assist you in diversifying your portfolio globally. Global Equities is Hemispheres’ flagship investment product.
Please contact Hemispheres Investment Management for a free consultation. We provide guidance to assist you in optimizing your investment strategies and helping you achieve your investment goals. Book a meeting.
Footnotes and References
[1] JLL. 2025 Global Data Center Outlook. https://www.jll.com/en-us/insights/market-outlook/data-center-outlook
[2] Morgan Lewis. Private Equity in Data Centers – Growth, Risks & Opportunities. https://www.morganlewis.com/pubs/2025/09/in-the-know-private-equity-in-data-centers-growth-risks-and-opportunities
[3] JLL. 2025 Global Data Center Outlook. https://www.jll.com/en-us/insights/market-outlook/data-center-outlook
[4] Federal Reserve Bank of Richmond. Boom and Bust in Telecommunications. https://www.richmondfed.org/publications/research/economic_quarterly/2003/fall/couperhejkalwolman [5] Morgan Lewis. Private Equity in Data Centers – Growth, Risks & Opportunities. https://www.morganlewis.com/pubs/2025/09/in-the-know-private-equity-in-data-centers-growth-risks-and-opportunities
[6] JLL. North America Data Center Report Midyear 2025. https://www.jll.com/en-us/insights/market-dynamics/north-america-data-centers
[7] Future Bridge Americas. The True Cost of Stranded Assets in Data Centers. https://future-bridge.us/the-true-cost-of-stranded-assets-in-data-centers/
[8] JLL. 2025 Global Data Center Outlook. https://www.jll.com/en-us/insights/market-outlook/data-center-outlook
[9] Morgan Lewis. Private Equity in Data Centers – Growth, Risks & Opportunities. https://www.morganlewis.com/pubs/2025/09/in-the-know-private-equity-in-data-centers-growth-risks-and-opportunities
[10] Optrium. Will Data Centres Become Obsolete? https://optrium.co.uk/will-data-centres-become-obsolete/
[11] Vistra. Can private debt cause systemic risk? https://www.vistra.com/insights/can-private-debt-cause-systemic-risk
[12] Chief Investment Officer – Ai-CIO. Private Credit Faces Rising Risk Factors. https://www.ai-cio.com/news/risk-factors-are-rising-in-private-credit-performance-harder-to-predict/
[13] The Fed. Private Credit: Characteristics and Risks. https://www.federalreserve.gov/econres/notes/feds-notes/private-credit-characteristics-and-risks-20240223.html
[14] Vistra. Can private debt cause systemic risk? https://www.vistra.com/insights/can-private-debt-cause-systemic-risk
[15] Pillsbury Law. Power Purchase and Interconnection Agreements for Data Centers. https://www.pillsburylaw.com/en/news-and-insights/power-purchase-interconnection-agreements-data-centers.html
[16] Pillsbury Law. Power Purchase and Interconnection Agreements for Data Centers. https://www.pillsburylaw.com/en/news-and-insights/power-purchase-interconnection-agreements-data-centers.html
[17] U.S. Energy Information Administration (EIA). Data center owners turn to nuclear as potential electricity source. https://www.energy.gov/sites/default/files/2024-12/44-%20Exh.%20PP%20-Data%20center%20owners%20turn%20to%20nuclear%20as%20potential%20electricity%20source%20%28EIA%29.pdf
