Analysis of the Impact of Power Infrastructure Bottlenecks on Data Centers and AI Investment Returns

Based on the latest data and research, I will systematically analyze the profound impact of power infrastructure bottlenecks on the data center strategies of tech giants and AI investment returns for you.

U.S. power infrastructure is facing unprecedented bottleneck pressure. According to JLL’s 2026 Global Data Center Outlook Report [1],

, with insufficient power supply being the direct cause of most of these delays.

Region	Average Wait Time
Major U.S. Data Center Markets	At least 4 years
Northern Virginia (World’s Largest Data Center Market)	Approximately 7 years
Atlanta	Approximately 5 years
Chicago	Approximately 5 years

The core driver of this phenomenon is the rapid development of artificial intelligence. AI data centers require far more power than traditional computing applications, forcing grid upgrades and power generation facility construction to proceed in tandem, but the actual progress is far behind the pace of demand growth.

The U.S. federal government has elevated the data center power issue to a national strategic level. On October 23, 2025, the U.S. Department of Energy sent a “Large Load” letter No. 403 to the Federal Energy Regulatory Commission (FERC), requesting the initiation of a rulemaking process to accelerate the grid connection process for large loads (including AI data centers) [2]. This marks that delays are no longer regarded as a purely technical issue, but have risen to a

.

---

Facing power bottlenecks, tech companies are adopting diversified strategies to ensure the stability and controllability of energy supply.

Company	Investment/Collaboration Direction	Scale
Alphabet/Google	Acquire Intersect Power	$4.75 billion [3]
Meta	Nuclear Energy Agreements (with Kairos, NextEra, etc.)	6.6 GW [4]
Microsoft	Restart Nuclear Power Plant with Constellation Energy	835 MW [5]
OpenAI/Oracle/SoftBank	Stargate Project	$500 billion [6]

Alphabet’s acquisition of Intersect Power is particularly noteworthy. The deal includes not only a $4.75 billion acquisition price but also the assumption of corresponding debts. Intersect has $15 billion worth of assets under construction or in operation, and is expected to bring online approximately

of power generation capacity by 2028 — a scale that exceeds [7].

Nuclear energy is becoming a “hot commodity” pursued by tech companies:

• Meta
  : has signed three new nuclear energy agreements, and together with its previous layout, will obtain a total of
  6.6 GW
  of nuclear power supply, with the goal of achieving this by 2035 [8]
• Google
  : has signed a nuclear energy agreement of approximately
  1.1 GW
  with Kairos Power and NextEra Energy
• Microsoft
  : has reached an agreement with Constellation Energy to restart the Three Mile Island Nuclear Power Plant, which can provide
  835 MW
  of power
• Oracle
  : is exploring small modular reactor (SMR) technology

However, nuclear energy projects typically require

from planning to commissioning. During this period, tech companies are filling the energy gap through the following methods:

• Large-scale procurement of
  liquefied natural gas (LNG)
  for on-site power generation
• Deployment of
  battery energy storage systems
• Construction of
  on-site microgrids

OpenAI and SoftBank recently announced a

investment in SB Energy to support the construction of data centers and energy infrastructure for the Stargate project [9]. SB Energy is developing a data center park in Milam County, Texas, with supporting solar energy and energy storage facilities.

The core logic of this model is:

. As Wood Mackenzie pointed out, hyperscale data center developers are through a combination of “power generation + energy storage” [10].

---

The capital expenditures of tech companies are rising at an unprecedented rate:

Year	Capital Expenditures of Six Major U.S. Hyperscale Enterprises	Remarks
2025	Nearly $400 billion	Microsoft, Amazon, Alphabet, Oracle, Meta, CoreWeave
2026	Projected $500 billion	Continued acceleration
2027	Projected $600 billion	Moody’s Prediction [11]

Moody’s analysis points out that

is required to meet the AI data center expansion demand by 2030, covering construction, IT infrastructure, and power costs [12].

However, the realization of investment returns faces severe challenges:

• MIT research shows that 95% of enterprise organizations have not yet received any returns from AI investments
  [13]
• OpenAI recorded a net loss of at least $11.5 billion in Q3 2025 (as of September 30)
  [14]
• McKinsey warned in May 2025 that demand forecasts for AI investments are mostly “experience-based guesses” with great uncertainty

Power infrastructure bottlenecks are having multiple financial impacts:

1. Increased construction costs
   : Grid upgrades and on-site power generation systems can increase the cost of large-scale data center projects by
   tens of millions of dollars
   and extend the construction period by more than one year [15]
2. Electricity price fluctuation risk
   : Electricity prices have risen in some markets due to the surge in power consumption by data centers, triggering public concerns and regulatory pressure
3. Declining capital efficiency
   : A large amount of capital is locked in energy infrastructure instead of being directly used for AI capability building

The impact of power bottlenecks on AI investment return cycles can be summarized in the following transmission chain:

Power Bottlenecks → Grid Connection Delays (4-7 years) → Data Center Commissioning Postponement
     ↓
Early Lock-in of Capital Expenditures → Extended Cash Flow Recovery Cycle
     ↓
High Fixed Costs (Energy, Facilities) → Increased Operating Leverage
     ↓
Revenue Growth Below Expectations → ROI Under Pressure

Taking OpenAI as an example, it has signed infrastructure agreements exceeding

in 2025 [16], but its commercialization process has just begun. This “invest first, earn later” model poses enormous pressure on participants with weaker financial strength.

---

Traditionally, data center site selection mainly considered factors such as land costs, network connectivity, and talent reserves. However,

, with location and cost factors taking a back seat [17].

• Ireland
  : Implemented a
  moratorium
  on data centers in the Dublin area in 2024, prohibiting new approvals until 2028, due to data centers already consuming approximately
  21% of the region’s electricity
  [18]
• UK
  : Launched “AI Growth Zones” in 2025 to attract data center investment by streamlining approval processes
• West Virginia
  : Legislated to create “Certified Microgrid Zones” in 2025, allowing data centers to be paired with dedicated power generation facilities while isolating cost burdens from other utility users [19]

This regional differentiation means that tech companies are facing the awkward situation of “having capital, technology, and demand, but nowhere to build”.

---

• Grid bottlenecks will continue to restrict the expansion speed of data centers
• Tech companies will accelerate the deployment of on-site power generation, energy storage systems, and microgrids
• Natural gas power generation will become an important transitional energy option
• Some projects will adopt the “operate while expanding” model

• The first batch of newly built nuclear power plants is expected to come online (although the number is limited)
• Small Modular Reactors (SMRs) may achieve commercial deployment
• The problem of grid interconnection queues is expected to be alleviated through policy intervention
• The surge in AI inference demand will drive the construction of distributed data center networks

• Superconducting power transmission technology may enter commercial applications, reducing transmission losses
• AI-native grid dispatching systems will improve overall energy efficiency
• Nuclear energy may become a standard configuration for AI data centers
• The interaction between data centers and the power grid will shift from one-way power consumption to two-way balance

---

Power infrastructure bottlenecks are fundamentally reshaping the data center expansion strategies of tech giants and AI investment return expectations.

mean that tech companies must prioritize and vertically integrate their energy strategies, shifting from simply “renting electricity” to “controlling energy assets”.

From the perspective of capital efficiency, this transformation is both a passive choice to deal with bottlenecks and an active layout to establish long-term competitive advantages. However, the data showing that

reminds us that there is still high uncertainty in the realization of huge investments.

For investors, the focus should shift from “who is building more data centers” to “who can obtain stable power supply at a lower cost and faster speed”. In this dimension, tech companies with vertically integrated energy capabilities will gain structural advantages, while competitors relying on the traditional grid-dependent model will face increasingly severe constraints.

---

[1] JLL’s 2026 Global Data Center Outlook Report (data on data center project delays and grid connection wait times)
[2] TechnoStatecraft - “Behind the Federal Power Grab to Fast-Track AI” (https://www.technostatecraft.com/p/behind-the-federal-power-grab-to) (content related to FERC Letter No. 403)
[3] Reuters - “Alphabet to buy clean energy developer Intersect in $4.75 billion deal” (https://www.reuters.com/technology/alphabet-buy-data-center-infrastructure-firm-intersect-475-billion-deal-2025-12-22/) (details of Alphabet’s acquisition of Intersect Power)
[4] Fierce Network - “Meta goes all-in on nuclear power with 6.6 GW plans” (https://www.fierce-network.com/cloud/meta-goes-all-nuclear-power-66-gw-plans) (scale of Meta’s nuclear energy agreements)
[5] Various sources - Microsoft Constellation Energy nuclear plant restart agreement, 835 MW capacity
[6] Reuters - “OpenAI, SoftBank invest $1 billion in SB Energy” (https://www.reuters.com/business/energy/openai-softbank-invest-1-billion-sb-energy-2026-01-09/) (commitment of $500 billion investment in the Stargate Project)
[7] Alphabet Investor Relations - Intersect acquisition details, 10.8 GW power capacity by 2028
[8] Los Angeles Times - “Meta signs multi-gigawatt nuclear deals to power AI data centers” (https://www.latimes.com/business/story/2026-01-09/meta-signs-multi-gigawatt-nuclear-deals-to-power-ai-data-centers) (details of Meta’s nuclear energy strategy)
[9] CNBC - “OpenAI and SoftBank announce $1 billion investment in SB Energy” (https://www.cnbc.com/2026/01/09/openai-and-softbank-group-announce-1-billion-investment-in-sb-energy-.html) (OpenAI and SoftBank’s investment in SB Energy)
[10] PV Magazine USA - “SB Energy secures $1 billion from OpenAI and SoftBank” (https://pv-magazine-usa.com/2026/01/12/sb-energy-secures-1-billion-from-openai-and-softbank-for-stargate-datacenter-expansion/) (analysis of grid bottlenecks and bypass strategies)
[11] The Register - “$3T AI infrastructure boom rolls on amid profit doubts” (https://www.theregister.com/2026/01/14/ai_investment/) (Moody’s capital expenditure forecast data)
[12] Moody’s 2026 Global Data Center Market Outlook - Global investment requirement of $3 trillion
[13] MIT Research - Enterprise AI ROI statistics, 95% no return
[14] The Register - OpenAI Q3 2025 financial results, net loss of at least $11.5 billion
[15] Engineering News-Record - “Grid Access, Not Land, Emerges as Bottleneck for Data Center Construction” (https://www.enr.com/articles/62227-grid-access-not-land-emerges-as-bottleneck-for-data-center-construction) (analysis of data center construction costs)
[16] CNBC - OpenAI infrastructure deals exceeding $1.4 trillion
[17] JLL 2026 Global Data Center Outlook - Power availability as primary site selection criterion
[18] AP News / International Energy Agency - Ireland data center moratorium, 21% electricity consumption
[19] Engineering News-Record - West Virginia certified microgrid districts legislation