Data Center Grid Lock, Google $20 B Intersect Power Deal, 7 GW US Crisis, and $1.7 T Capex Projections (2021 to 2026)

Grid Access Delays, US Faces 7 GW AI Data Center Capacity Crisis and Project Cancellations

The primary constraint on the multi-trillion-dollar artificial intelligence infrastructure buildout has shifted from semiconductor availability to the physical limitations of the global power grid. The inability of electrical infrastructure to accommodate the speed and scale of new demand is causing systemic project delays and cancellations, threatening the growth trajectory of the entire sector. This power bottleneck is forcing a strategic realignment where securing reliable, timely power is now the most critical factor for success.

Hyperscaler Project Delays and Cancellations

The core issue is a fundamental mismatch between the rapid deployment of AI data centers and the multi-year timelines required for grid upgrades and interconnection approvals. Before 2024, data center siting was primarily driven by fiber connectivity and land costs. Today, power availability is the determining factor. This has led to a significant backlog of projects unable to secure power.

• In the U.S., nearly half of the data centers planned for 2026 have been canceled or delayed, resulting in a 7 GW capacity crisis directly attributable to power grid bottlenecks and long interconnection queues.
• Some analysts project that 30-50% of planned 2026 AI data center capacity will be postponed until at least 2028 due to these electrical infrastructure constraints, representing a major impediment to near-term AI computational capacity growth.
• The problem is acute in established data center hubs. In Northern Virginia, the world’s largest data center market, utility Dominion Energy had to pause new data center connections in 2022 due to transmission constraints, a situation that highlighted the growing strain.

The Shift to “Bring Your Own Power” (BYOP)

In response to grid-related delays, a defining infrastructure trend for 2025-2026 is the adoption of “bring your own power” (BYOP) strategies. Hyperscalers are increasingly bypassing utilities by investing directly in their own dedicated energy generation. This strategic pivot, exemplified by ventures like the Liberty Energy Power Bridge joint venture, is driven by the urgent need for “time-to-power” rather than pure cost savings, as the opportunity cost of delayed AI deployment outweighs the premium of building independent power sources.

• The move toward on-site generation is a direct reaction to the realization that waiting for grid upgrades is no longer a viable business strategy for companies competing in the AI sector.
• This trend includes co-locating data centers with renewable energy projects, developing microgrids, and exploring dispatchable power sources like natural gas and even small modular reactors (SMRs) to guarantee uninterrupted supply.

$775 B in 2026 Capex, Hyperscalers Fund Massive Data Center and Energy Infrastructure

An unprecedented capital expenditure cycle is underway, driven by the competitive necessity for hyperscalers to build out their AI capabilities. This spending is increasingly directed not just at servers and GPUs but at the underlying power and cooling infrastructure required to run them. The sheer scale of this investment validates the market’s long-term growth thesis, even as it exposes the critical energy constraints.

Top-5 Hyperscaler Capex Projections

Leading hyperscalers like Amazon, Microsoft, Google, Meta, and Oracle are at the forefront of this spending. Their capital allocation plans signal a non-discretionary investment cycle necessary to remain competitive in AI services. This has created a massive tailwind for the entire data center supply chain.

• Analysts from Dell’Oro Group and others project that the top five hyperscalers alone will spend between $775 billion and $800 billion on data center infrastructure in 2026.
• The total worldwide data center capex is forecast to grow from approximately $434 billion in 2024 to over $1 trillion by 2029, with some estimates projecting a cumulative $1.7 trillion spend by 2030.
• These figures represent a significant upward revision from forecasts made before 2024, reflecting the explosive, power-intensive nature of generative AI workloads and the infrastructure required to support companies like Open AI.

Investment in Power-Related Infrastructure

A growing portion of this capital is being allocated to solve the energy problem directly. Power and cooling infrastructure now account for 30-40% of the total cost of a new AI data center, up from a much smaller fraction in traditional facilities. This includes investments in high-density power distribution, liquid cooling systems, and direct financing of new energy generation projects through vehicles like convertible bonds, as seen with firms such as Akamai.

Table: Data Center Capital Expenditure Forecasts

Hyperscaler Capex to Hit $1 Trillion by 2027, Driven by AI Demand
U.S. Big Four Hyperscaler Capex is projected to reach $1 trillion by 2027, quadrupling 2024 investments. This unprecedented surge, driven by escalating AI and data center demands, indicates an accelerating infrastructure arms race among top players.

(Source: FocusEconomics — via Hyperscaler AI & Data Center Energy 2026, $726B Dell'Oro – EnkiAI)

Google $20 B Intersect Power Deal Signals Co-Location Strategy for AI Data Centers (2024 to 2026)

Strategic partnerships between hyperscalers and energy developers have become the primary mechanism to bypass grid logjams. These alliances are moving beyond traditional Power Purchase Agreements (PPAs) toward deeper integration, including the co-location of data centers with power generation facilities. This ensures a dedicated, reliable power source and shortens the critical “time-to-power” for new facilities.

Google and Intersect Power’s Co-Location Model

The late 2024 partnership between Google, Intersect Power, and TPG represents a landmark shift in this direction. The collaboration is a clear signal that hyperscalers are now willing to co-invest in energy infrastructure to secure the power needed for their AI ambitions. This type of deal structure is becoming a template for the industry.

• The partners launched a $20 billion initiative to co-locate new data centers with clean energy generation projects, directly linking power supply with power demand.
• This model differs from earlier strategies that relied on purchasing renewable energy credits to offset consumption from grid-connected data centers. The new approach physically integrates the data center with its power source.
• By financing both the data center and the generation asset together, hyperscalers can de-risk projects for energy developers and accelerate the deployment of both, a strategy also seen in offtake agreements with operators like Hut 8.

Long-Term Power Purchase Agreements (PPAs)

Alongside co-location, hyperscalers are aggressively securing long-term (10-20 year) PPAs and offtake agreements to guarantee future supply and hedge against price volatility. These agreements are crucial for underpinning the financing of new large-scale energy projects, including wind, solar, and potentially nuclear, as seen in partnerships between data centers and large utilities like Next Era Energy.

Table: Key Energy Partnerships for AI Data Centers

US vs. Global, North America Dominates AI Data Center Growth Amid Power Constraints

North America, and the United States in particular, remains the undisputed epicenter of the AI data center buildout, but its leadership position is being actively challenged by its own infrastructure limitations. While the region attracts the majority of investment, severe power constraints are creating opportunities for other global markets with more available grid capacity or a more streamlined regulatory environment for energy projects.

North American Bottlenecks

The U.S. is home to the most significant planned data center expansion, with total power demand projected to climb from 61.8 GW in 2025 to as high as 134.4 GW by 2030. However, this growth is running directly into the reality of an aging grid that was not designed for such concentrated load growth. The result is a geographic fragmentation of development, with companies seeking out secondary and tertiary markets where power is more readily available.

• Key markets like Northern Virginia and Silicon Valley are saturated, with multi-year waits for power. This has pushed development towards states like Ohio, Texas, and Arizona, which offer more grid capacity and supportive policies.
• The 7 GW capacity crisis identified in 2026 is a uniquely American problem at this scale, stemming from a combination of massive demand and a balkanized regulatory and utility landscape that slows infrastructure development.

Emerging Global Locations

While the U.S. grapples with its grid, other regions are positioning themselves to capture a share of the AI infrastructure market. Countries with strong renewable energy resources, supportive governments, and available grid capacity are becoming more attractive. This dynamic is visible in markets like India, where entities such as the Adani Group and Reliance are making substantial investments in data center capacity powered by renewable energy.

Liquid Cooling at Commercial Scale, Direct-to-Chip Solutions Mitigate AI Power Density

The extreme power density of AI hardware has forced the rapid maturation and commercial-scale adoption of technologies that were considered niche prior to 2024. Direct-to-chip liquid cooling has transitioned from a specialized solution to a standard requirement for new AI data centers, driven by the thermal challenge of dissipating heat from server racks that now exceed 100 k W.

The Rise of Direct-to-Chip Cooling

Traditional air cooling is insufficient for the thermal loads generated by modern AI accelerators. The shift to liquid cooling is not an incremental improvement but a necessary architectural change. This technology’s revenue doubled in a single year, highlighting its critical role in enabling the AI buildout. The strategy is central to hardware providers like Dell, which are designing systems around these new thermal management requirements.

• Modern AI server racks are projected to exceed 600 k W in the near term, with 1 MW rack configurations on the horizon, making advanced liquid cooling solutions non-negotiable for future deployments.
• The adoption of liquid cooling is a key enabler of deploying high-density GPU clusters, which are the engines of large-scale AI training and inference models. Without it, the physical footprint of AI data centers would be unmanageably large.

The Nuclear and SMR Option

To address the macro-level energy shortage, hyperscalers and data center operators like Equinix are actively exploring nuclear power, particularly Small Modular Reactors (SMRs), as a long-term source of reliable, carbon-free baseload power. While this was a theoretical discussion before 2024, by 2025-2026 it has become a serious strategic consideration, with active talks between tech companies and nuclear developers to power future data center campuses.

SWOT Analysis, Hyperscaler AI Data Center Energy Buildout

The AI data center market is characterized by a powerful combination of immense demand-side pull and significant capital availability, creating massive opportunities for growth. However, this growth is fundamentally constrained by physical infrastructure limitations, primarily power grid availability, and growing regulatory and community opposition, which pose significant threats to project timelines and costs.

Table: SWOT Analysis for Hyperscaler AI Data Center Energy Market

BYOP Becomes Standard, Hyperscalers Bypass Grids with On-Site Generation in 2026

If grid interconnection delays and capacity shortages persist as the primary obstacle through 2026, the market will see a decisive and accelerated pivot to “bring your own power” models. Expect hyperscalers to move beyond partnerships and become direct developers and financiers of on-site power generation, including natural gas peaker plants for reliability, renewables for sustainability, and SMRs for long-term baseload power.

Signals for BYOP Acceleration

The most critical signal to watch is the volume of capital hyperscalers allocate to energy infrastructure versus data center hardware. Recent trends from 2025 to today confirm this shift is already underway, as companies prioritize securing GWs of power to support their AI ambitions.

• The strategic focus is shifting from raw FLOPS (Floating-Point Operations Per Second) to a more efficient metric: “tokens-per-watt, ” which underscores the increasing importance of energy efficiency in AI hardware design, a key concern for chipmakers like TSMC.
• Watch for an increase in direct hiring of energy infrastructure experts and project finance professionals by companies like Amazon, Microsoft, and Google, indicating a move to internalize energy development capabilities.
• Monitor announcements for new data center campuses that explicitly include plans for on-site, multi-gigawatt power generation facilities as part of the initial design, rather than as a later addition.

Impact on Utility Business Models

This trend poses a direct challenge to traditional utility business models, which are structured around centralized generation and slow, planned load growth. A future where the largest electricity customers build their own power could lead to grid fragmentation and stranded assets. This dynamic will force utilities to innovate, offering faster interconnection services, flexible tariffs, and new partnership models to retain their largest industrial clients.

The questions your competitors are already asking

This report covers one angle of the data center energy market. The questions that matter most depend on your work.

• data centers building their own power plants
• where new data centers are being built in the US
• top data center liquid cooling companies
• data center deals with nuclear power companies

This report does not answer these. Enki Brief Pro does.

Your question, your angle, your framework. SWOT, PESTL, scenario modelling. The same niche depth, built around the decision your work actually depends on.

Run your first brief in Enki Brief Pro