Teaching Electricity and Emissions
One activity does most of the work here: two identical workloads on two different grids. Students who compute that difference themselves stop treating kilowatt-hours and carbon as the same quantity, permanently.
Why this is hard to teach.
Students arrive with energy and emissions fused into one idea, and everyday language encourages it — we say a device is “clean” or “dirty” as though the property lived in the device. It doesn't. It lives in the grid the device is plugged into, at the hour it is running.
The second difficulty is that this guide's headline number is a range, not a value. LBNL projects 6.7% to 12% of US electricity by 2028 — nearly a factor of two. Students trained to want the answer will ask which number is right, and the honest reply is that both are, under different assumptions about hardware efficiency and growth elsewhere in the economy. Sitting with that is the skill.
Third, this is the guide most likely to produce a student who wants to conclude something. The marginal-emissions question is open, and a class that ends on “so we don't know” feels unsatisfying. Name that feeling rather than resolving it — the discomfort of an honest open question is worth more than a tidy wrong one.
Target misconceptions.
“Kilowatt-hours are kilowatt-hours.”
A kilowatt-hour is a quantity of energy, not of carbon. Two identical facilities on two different grids have different carbon footprints and identical electricity bills. The emissions live in the generation mix, not in the load.
“If renewables are covering the growth, the problem is solved.”
Renewables meeting nearly half of demand growth is real, global, and about growth in aggregate. It says nothing about what plant ramps up at 7pm in July on a particular grid, or whether a coal retirement got postponed.
“A projection is a prediction.”
Projections are scenarios built on stated assumptions. The federal report here was published in December 2024, before the 2025–26 capital expenditure surge, and it could prove conservative on the upside or wrong about efficiency in either direction.
“The national percentage tells me about my area.”
It is a national aggregate, and the report's own limitations section says it says nothing about where the load lands. The unit that answers a local question is the balancing authority.
Two classroom-ready activities.
Same query, two grids
35 min · public dataGive the room two identical AI workloads — same model, same number of queries, same kilowatt-hours. Assign one to a coal-heavy balancing authority and one to a hydro- or nuclear-heavy one. Using published EIA fuel-mix data, groups compute the emissions for each.
The numbers come out strikingly different. Then ask the question that lands the lesson: which of the two facilities has the higher electricity bill? (They're comparable. The carbon is not.)
Debrief on where the difference actually came from. Nobody changed the AI, the hardware, or the number of queries. The only variable was geography — which is the guide's whole argument, arrived at by arithmetic rather than assertion.
Materials: EIA's public electricity data browser, one laptop or printout per group. No account required.
Write the headline
20 min · no techHand each group the same fact: US data centers are projected to use between 6.7% and 12% of national electricity by 2028. Each group writes a one-sentence news headline from it.
Post the headlines side by side. Some will have taken the top of the range, some the bottom, some the midpoint, and a few will have kept the range intact. Ask which headlines a reader could reconstruct the original fact from.
This is the drift mechanism from the flagship guide, reproduced live in twenty minutes by people who were not trying to mislead anyone. That is the point — the drift in this literature did not require bad faith, only ordinary compression.
Materials: index cards or sticky notes. Works well as a warm-up.
Discussion prompts.
Ordered from easy to charged.
- 1.Why does the same activity have a different carbon footprint in two places?
- 2.What would you need to know to answer “is this facility's electricity clean?”
- 3.The projection is a range spanning nearly a factor of two. Is that a weakness in the research, or information?
- 4.Renewables are meeting nearly half of demand growth. What does that settle, and what does it not?
- 5.Is it responsible to publish a number that could be off by half? What's the alternative?
- 6.CPAI says it doesn't know what clears at the margin. When is “we don't know” a good enough answer from someone teaching you?
Seeing whether it landed.
The two-grids explanation. Ask a student to explain to someone outside the class why two identical data centers can have different carbon footprints. If they reach for the grid rather than the building, it landed.
The range test. Give them a projection with a range and ask them to state it accurately in one sentence. Watch for whether the range survives.
The scale-tag test. Hand them the 4.4% figure and ask what question it answers and what question it does not. The second half is the harder and more useful one.
When someone asks “is this proven?”
In teacher voice
"The demand figures are the most solid thing in this whole debate — federal, transparent, and deliberately conservative. What they can't tell you is the emissions consequence, because that depends on which plants run to serve the load. The renewables finding is real and I'm not going to leave it out just because it complicates the story. And the marginal question — what actually burns when new load shows up — is genuinely unresolved. CPAI hasn't done that research and won't pretend otherwise. If somebody tells you confidently that AI is or isn't driving emissions at the margin, they're ahead of the evidence in whichever direction they're arguing."
Delivering a correction to a room that cares
Every guide in this series involves telling people that a claim they hold is imprecise. The flagship educator page carries the three-beat rule — the sequence that keeps a correction about scale from being heard as indifference. Read it before you teach any of these.
The three-beat rule →Where this leads
Teaching this is a different skill than knowing it.
Teaching AI Well is the facilitator library behind CPAI's Certified Applied AI Trainer Program — ten lessons on how to teach AI honestly, including the material on this page.