AI - The AI Supply Chain - Part 2

Data Center Power Overview

Two of the most critical components in data center operations are:

Power Supply
Cooling Systems

Power is the backbone of every data center, especially as AI workloads push energy demands to new extremes. Most data centers rely on local utility companies for their electricity needs, and they consume vast amounts of it. As data center capacity grows, so too does the power required to operate them.

Currently, data centers are estimated to consume 3–4% of the total U.S. electricity supply. That share is expected to rise sharply, potentially reaching 8–12% by 2030, and as high as 16% by 2039. To put this in perspective, powering data centers at scale will require the energy equivalent of multiple large metropolitan cities, each consuming 10 gigawatts (GW) or more.

https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/investing-in-the-rising-data-center-economy

Managing power is more than just paying an electricity bill—it involves navigating an entire energy ecosystem consisting of:

Power generation (e.g., natural gas, nuclear, solar, wind)
Power transmission and distribution
Power storage and redundancy systems

In high-demand regions, utility providers are now quoting multi-year lead times for delivering new power to data center developments. These delays often exceed construction timelines and have become a critical bottleneck in scaling data center capacity. As a result, companies are re-evaluating:

Where they build data centers
How they secure and supplement power (e.g., private energy grids, renewables, microgrids)

The chart below (adapted from Visual Capitalist: Ranked – Top 50 Data Center Markets by Power Consumption) illustrates how major data center markets compare in terms of power draw. Many of the most saturated regions, including Northern Virginia, Silicon Valley, and parts of Texas, are now facing long wait times for new electrical capacity.

Time to Power - As AI infrastructure expands, the ability to secure reliable, large-scale power will become a defining factor in global data center strategy and competitiveness.

https://www.visualcapitalist.com/cp/top-data-center-markets/

The Elephant in the Room: Renewable Energy

Let’s address the elephant in the room when it comes to powering the data centers of the future, renewable energy. As demand for electricity grows exponentially due to AI and digital infrastructure, the role of solar, wind, and hydroelectric power is often viewed as the clean and sustainable solution. And to be clear: investment in renewable energy is essential. But it’s also critical to understand its limitations, technical, environmental, and logistical, especially when scaled to meet the demands of modern data centers.

1. Renewables Can’t Scale Fast Enough

While renewables are expanding rapidly, they’re not growing quickly enough to meet projected AI-driven energy demand. Data centers already consume 3–4% of U.S. electricity and may require up to 16% by 2039. That kind of exponential demand growth can’t be matched by current renewable energy deployment timelines, especially when considering the infrastructure needed to distribute and store that power reliably.

2. The Wind Energy Debate

Wind energy, in particular, faces significant resistance:

Aesthetic objections - Offshore and rural wind farms often spark local opposition over disrupted views and landscape impact.
Health concerns - Residents living near wind turbines report side effects such as sleep disruption, nausea, vertigo, and headaches, commonly referred to as “wind turbine syndrome.”
Environmental impact - Bird populations and other local wildlife can be affected by turbine placement and operation.

Despite improvements in turbine design and placement, public sentiment and regulatory hurdles continue to slow development.

3. The Intermittency Challenge

The biggest hurdle for renewables is intermittency:

Solar doesn’t generate at night or on cloudy days.
Wind turbines don’t work when the air is still, or when wind conditions are too extreme.
In colder climates, ice buildup on blades (as seen in the Pacific Northwest) can shut turbines down entirely.

This variability creates reliability issues for data centers that require constant, uninterrupted power. Without robust energy storage systems, renewable sources can’t provide the on-demand baseline power these facilities need.

4. Overgeneration and Transmission Limits

Even when renewables do produce power, like during strong wind conditions, there’s often nowhere for the excess energy to go:

The electrical grid lacks sufficient long-term storage to bank surplus energy.
Transmission lines often can’t handle sudden power surges.
In some regions, turbines are shut down to avoid overloading the grid.

This leads to wasted generation capacity and reinforces the need for better energy storage and grid modernization.

5. Location, Location, Location

Even if we overcome generation and storage issues, there's still a geographic mismatch between where renewable energy is produced and where data centers are built. Remote wind farms and solar fields often exist far from urban data hubs, which means transmission capacity becomes the next bottleneck. High-voltage infrastructure upgrades are costly and face their own permitting challenges.

Bottom line: Renewable energy is part of the future, but it won’t be enough on its own to support the data center boom, especially in the near term. Until power storage and transmission technologies catch up, AI infrastructure will continue to rely heavily on traditional energy sources. This makes energy strategy, not just energy type, a key factor in global digital competitiveness.

Three Macro Approaches to the Data Center Power Challenge

As demand for compute continues to rise, especially from AI workloads, data center operators are exploring new power strategies. Three prominent approaches have emerged:

1. Nuclear Power: The Hyperscaler Bet

Hyperscalers are beginning to invest directly in nuclear energy. In 2024, Amazon signed a “behind-the-meter” Power Purchase Agreement (PPA) with Talen Energy, securing data center energy capacity directly from a dedicated nuclear reactor in Pennsylvania. This means the power is sourced directly from the generation site, bypassing the traditional utility grid. The deal was temporarily halted by the Federal Energy Regulatory Commission (FERC), which argued that such agreements might set a precedent for large power consumers to benefit from grid infrastructure without contributing to its maintenance, raising cost risks for other utility customers. The matter remains under appeal.

Also in late 2024, Microsoft entered a deal with Constellation Energy to help reactivate Three Mile Island Reactor Unit 1, specifically to power its growing AI and data infrastructure.

In parallel, companies are heavily investing in Small Modular Reactors (SMRs). These next-gen nuclear solutions are:

Compact and scalable
Prefabricated for faster deployment
Well-matched to the power requirements of individual data centers

U.S.-based companies like NuScale and TerraPower (co-founded by Bill Gates) are leading the charge, viewing SMRs as ideal for collocating with large compute campuses. Some evidence that the SMR market may be ready for some increased traction is the increased news around enriched Uranium startup General Matter.

2. Natural Gas and Microgrids

Many data centers are tapping directly into natural gas pipelines or deploying fuel cell systems to become self-sufficient microgrids. In this model, the data center generates its own electricity on-site, independent of the main utility grid.

Natural gas is favored for:

High energy efficiency
Lower carbon emissions compared to coal or diesel
Stable and abundant domestic supply

Operators often contract with third-party Power-as-a-Service (PaaS) providers who install, own, and manage these gas-powered turbines, reducing the capital expenditure and maintenance burden on the data center itself.

3. Long-Term Energy Storage

Renewable energy integration hinges on solving the energy storage problem. Unlike fossil fuels or nuclear, renewables are intermittent, solar depends on sunlight, wind on air movement. Data centers require 24/7 reliability.

A few major storage strategies are gaining attention:

Pumped Storage Hydropower (PSH)

PSH stores energy by using excess electricity to pump water uphill into a reservoir. When demand spikes, the water is released downhill through turbines to generate electricity. While efficient, PSH requires specific geographic conditions and vast amounts of water, limiting its scalability.

Pumped Storage Hydropower | Department of Energy

Lithium-Based Battery Storage

Lithium-ion batteries are common in states like California, where surplus solar energy is stored during the day and discharged during evening demand peaks. However, their limited duration (typically a few hours) doesn’t match the continuous power requirements of data centers.

Thermal Energy Storage: A Promising Frontier

Thermal energy storage systems are gaining momentum as a longer-duration, lower-cost alternative to batteries. Companies like Exowatt (backed by OpenAI’s Sam Altman) and Fourth Power (supported by Bill Gates) are pioneering this space.

Here’s how these systems work:

Renewable electricity is used to heat liquid tin via a closed-loop heat transfer system.
The tin transfers heat to carbon blocks (graphite), which store the energy at temperatures exceeding 2,400°C (4,350°F).
The carbon blocks emit intense light, which is captured by thermophotovoltaic (TPV) cells.
TPV cells convert this light into electricity, which can be used immediately or stored in traditional batteries.

Thermal Energy Grid Storage (TEGS) Concept - MIT ASE

This approach offers:

Longer storage duration
Reduced reliance on chemical batteries
Efficient conversion of excess renewable power into dispatchable electricity

These three approaches, nuclear, natural gas microgrids, and advanced energy storage, represent the evolving playbook for solving the power demands of tomorrow’s data infrastructure. Operators are increasingly becoming energy strategists, designing custom solutions tailored to their scale, location, and sustainability goals.