The AI energy lesson below is built for evidence-based classroom analysis, not for broad claims that artificial intelligence is automatically good or bad for the grid. Students compare two measurable pressures: electricity demand from data centers and efficiency gains inside AI hardware and software. That tension is useful for a critical infrastructure lesson because it shows why a single metric can mislead decision-makers.
Framing The AI Energy Lesson
AI Energy Lesson Objective
The main objective is for students to explain how AI can increase total electricity demand while also reducing the energy consumed per task. In 2025, worldwide data center electricity demand rose by 17%, while global electricity demand grew by about 3%, according to the IEA data center analysis. That comparison gives students a concrete starting point: sector-specific demand can grow much faster than the wider electricity system.
At the same time, the research packet states that power consumed per AI task has fallen by at least one order of magnitude annually in recent years because of software and hardware efficiency improvements. This does not cancel the demand problem by itself. It gives students a reason to ask a more precise question: are efficiency gains reducing total load, or are lower per-task costs enabling far more AI use?
Core Question For Students
The central question is: how should infrastructure planners judge the energy impact of AI when one dataset shows rapid data center growth and another shows better energy efficiency per computation? A strong student answer should avoid single-cause explanations. It should separate demand growth, equipment efficiency, grid capacity, location, and operating behavior.
This AI energy lesson works best when students treat the grid as a constrained engineering system. A data center does not only consume energy over a year. It also draws power at specific times and places. That distinction matters because a region can have enough annual energy in theory while still facing local transmission, distribution, cooling, or peak-load stress.
Data Center Demand As The Case Study
Reading The 2025 Electricity Signal
Students should first calculate the difference between 17% data center electricity growth and roughly 3% global electricity growth in 2025. The goal is not to memorize the numbers. The goal is to identify what kind of infrastructure question each number can answer. The global growth figure gives background conditions. The data center figure points to a concentrated load category that may require planning attention.
A useful classroom prompt is to ask students whether 17% growth is enough information to rank local grid risk. The answer should be no. Students still need location, generation mix, transmission availability, cooling technology, operating schedule, and demand flexibility. The research supports the demand-growth claim, but it does not give enough detail to judge every region or facility. That boundary should be stated clearly.
Separating Efficiency Per Task From Total Demand
The per-task efficiency claim helps students distinguish engineering efficiency from system-level consumption. If each AI task uses much less power, a model provider can run more tasks at the same energy cost, or it can reduce consumption if demand is stable. The research packet does not prove which outcome dominates in every setting. Students should be asked to diagram both possibilities.
For this AI energy lesson, a simple comparison table can keep the reasoning disciplined:
Students can then write one sentence for each row explaining why infrastructure decisions need more than a headline statistic. This turns the activity into a technical evidence exercise instead of a debate about AI in general.
Critical Infrastructure Analysis Tasks
Stakeholder Map
Students should map the parties affected by AI-related energy demand: grid operators, utility planners, data center owners, public agencies, local communities, and classroom-scale users of AI services. Each group sees a different part of the system. Grid operators focus on reliability and peak conditions. Data center operators focus on uptime, cooling, and power contracts. Public agencies must weigh economic activity, reliability, emissions, water use, and land use where those issues are part of the evidence packet.
A related classroom extension can compare this activity with an AI energy lesson for the U.S. grid, especially if students need more practice separating national electricity statistics from local infrastructure planning.
Evidence Checks
Students should be required to label every claim as observed, projected, or uncertain. The 2025 growth figure is an observed data point in the research packet. The U.S. 2028 estimate is a projection. The AP reported that U.S. data center power consumption was projected to reach up to 12% of national electricity usage by 2028, driven by AI-related expansion, in an AP report on AI data center energy planning. Students should treat that number as a planning scenario, not as a guaranteed outcome.
The class can then test how wording changes the strength of a claim. For example, students should avoid saying AI will cause a fixed amount of national electricity use. A better sentence is that one cited projection placed U.S. data center power consumption at up to 12% of national electricity usage by 2028. The difference is small in wording but large in technical accuracy.
Assessment And Classroom Controls
Student Deliverables
A strong assessment asks students to produce a short technical memo. The memo should define the problem, summarize the two main data points, explain the difference between per-task efficiency and total system demand, and identify what extra information a utility or public agency would need before approving infrastructure changes.
- Claim: State one evidence-backed claim about data center electricity demand.
- Counterpoint: State one evidence-backed claim about AI efficiency gains per task.
- Limit: Identify one fact that is missing from the research packet.
- Decision: Recommend what data should be collected before a grid planning decision is made.
This format rewards careful reasoning. It also prevents students from treating AI as a single device or data centers as identical facilities. A small regional facility, a large training cluster, and a flexible demand-response participant would not create the same engineering problem, even if all are described as AI infrastructure in general language.
What Claims Should Be Marked Uncertain
A cautious classroom standard is to mark claims uncertain when the research packet does not provide location, method, or operating conditions. Students should not infer local outage risk from global growth alone. They also should not assume that per-task efficiency gains automatically lower grid load. Both claims require additional evidence about deployment scale and demand behavior.
The teacher can also include a short note that related STEM resources are available through the NATEWIN network portal, but the graded analysis should depend on the assigned evidence packet and cited sources. That separation helps students see the difference between background reading and evidence used in a technical claim.
AI Impact On Energy Efficiency In Critical Infrastructure
What The Lesson Can Safely Conclude
The safest conclusion for this AI energy lesson is that AI creates a measurement problem for critical infrastructure planning. Efficiency per task can improve while total electricity demand rises. Data center electricity demand grew much faster than global electricity demand in 2025, and U.S. planning discussions included projections of data centers reaching up to 12% of national electricity usage by 2028. Those facts support serious analysis, not automatic alarm or automatic optimism.
The lesson should close by asking students to write one infrastructure question that cannot be answered from the available facts. Good examples include whether a specific utility territory has enough transmission capacity, whether a data center can shift demand during peak periods, and whether cooling requirements change the local resource picture. Those questions keep the analysis grounded in evidence and show students how technical decisions depend on scale, location, and system configuration.
Used this way, an AI energy lesson gives students a practical model for engineering judgment. They learn to compare metrics, challenge unsupported claims, and explain why efficiency improvements inside a computing system do not always translate into lower energy demand across the grid.
