AI empowers the electricity market: Why electricity may be more important than GPUs

Electricity: A key bottleneck in AI infrastructure

A measure of the artificial intelligence (AI) boom is often the capital expenditures of NVIDIA GPUs and hyperscale cloud service providers. However, no amount of data centers will be useful if you can't get enough power to power these advanced chips. Because of this, electricity is becoming one of the most important constraints in AI infrastructure.

The International Energy Agency (IEA) predicts that electricity consumption in global data centers will grow from approximately 485 terawatt hours (TWh) in 2025 to approximately 950 TWh in 2030. Data centers focused on AI are expected to grow faster. The problem is not only how much power AI consumes, but also how quickly demand explodes. Data centers can be built in a few years, while transmission lines, substations and new power generation facilities often require longer construction cycles.

GPUs are only one layer of the AI architecture

The requirements of AI clusters go far beyond processors. Thousands of accelerators require networks, power conversions, cooling systems and reliable grid access. As server density increases, the requirements for supporting infrastructure increase.

This has spawned a broader AI supply chain:

GPU → network → power conversion → cooling → grid connectivity → power generation

Our AI infrastructure guide shows how investment themes have expanded from chips to power and thermal management. The key difference is that some bottlenecks cannot be solved simply by ordering more equipment. Hyperscale cloud service providers may be able to purchase more GPUs, but they cannot instantly establish new high-voltage connections.

Grid access is becoming increasingly scarce

Texas provides the clearest example yet. The proposed data center power demand has exceeded 700 gigawatts (GW), far exceeding the current total data center power consumption in the United States. Regulators have begun to tighten grid-connected rules because some of the demand may come from projects that will never be implemented. Reuters described the issue as a "ghost demand."

This creates a real financial dilemma: Utilities may invest in infrastructure for projects that have never occurred, while real AI parks face years-long power waiting periods.

Why it matters

• Grid connectivity: New projects may take years to wait
• Transformer: Long manufacturing cycle
• Cooling system: High-density GPUs generate more heat
• Power generation: AI requires reliable power supply around the clock
• Power transmission: Power production may be far away from demand centers

Power companies are becoming AI beneficiaries

This bottleneck is creating a new group of AI beneficiaries. For example, Vertiv provides power management and cooling systems for data centers and recently agreed to acquire microgrid expert Utility Innovation Group for up to $2.6 billion. The deal shows that on-site power generation and infrastructure independent of the grid are becoming extremely valuable.

Microgrids are important because they can reduce reliance on delayed utility connections by combining grid power with on-site generation and energy storage. Utility companies are also becoming part of the AI industry chain. Power providers such as NextEra Energy and Dominion Energy are increasingly benefiting from long-term demand created by hyperscale cloud service providers and data developers.

The IEA's energy outlook predicts that renewable energy will provide the majority of new data center demand, but natural gas, nuclear energy and energy storage will also remain important as AI workloads require reliable power supply around the clock.

This is also changing the economic model of Bitcoin mining. Miners have taken control of valuable grid connections in areas with cheap electricity. If AI companies are willing to pay higher prices for the same electricity, some operators may shift production capacity to high-performance computing. Coinpaper's mining analysis shows that electricity itself is becoming a scarce asset.

As a result, the AI investment story may no longer be limited to just "who makes the fastest chips" but may extend to competition across the energy ecosystem.