The Water Under The Lesson

On Wednesday, July 15, Willamette Week reported that students in Portland, Oregon are asking their school district to fold artificial intelligence into its climate policy.¹ The recommendation, delivered at a June 23 School Board meeting by the district's Climate Crisis Response Committee, is small in language and large in implication. Limit AI use in the classroom as much as possible. Publicly track and codify the AI the district does use. When AI is used, teach the environmental cost alongside the tool.¹

The recommendation arrives at a school system that has already paused. Superintendent Kim Armstrong told the board there are no generative AI products moving forward in the fall of 2026, and the board has stopped new AI contracts.¹ In a season where district after district has been racing to adopt or reject the tools, Portland has done neither, and now the students in it are proposing a third path. Not adoption. Not refusal. Accounting.

The Story Under The Story

The reason the story matters is not that a committee filed a memo. It is that a committee named a dimension of the AI-in-schools debate that no state framework, no vendor pitch, and no district's ethics guidance has been willing to put on the table.

Every serious pilot from the last twelve months has evaluated AI in classrooms along three axes. Efficacy: does it help students learn. Labor: does it help teachers work. Equity: does everyone in the building have access. What the Portland committee added, quietly, is a fourth. Cost. Not cost to the district. Cost to the world.

The numbers behind the recommendation are the reason it does not read as posture. A United Nations University report published in June found that global data centers consumed 448 terawatt-hours of electricity in 2025. If treated as a nation, that consumption would rank eleventh in the world, behind France and ahead of Saudi Arabia.² By 2030, the same infrastructure is projected to draw 945 terawatt-hours, nearly triple the combined annual electricity use of Pakistan, Bangladesh, and Nigeria.² Closer to home, Google's cluster in The Dalles, an Oregon town of roughly seventeen thousand people, consumed about 550 million gallons of water in 2025, close to forty percent of the city's total supply. Its consumption has quintupled since 2012.³

Those numbers do not tell a Portland student whether to use a chatbot for a history essay. They do tell her, correctly, that the choice is not free.

What The Adults Left Out

For the last two years, the argument for AI in schools has been posed as a bilateral question. Kid and screen. Teacher and workload. Vendor and district. The environmental cost of the tool has been kept, by mutual convenience, out of the frame. It is the kind of cost that fits into no line item in a district budget. It happens somewhere else, to someone else, in a town that does not have a school board seat.

The Portland students have refused that convenience. What they have proposed is that AI use in a classroom is a real transaction with real consequences, and that a student who is being asked to type a prompt should also be asked to know what the prompt takes. It is a small principle. It is also the beginning of a serious one. It says that a fifteen-year-old learning to use a tool is also learning what a tool is, and that the second part is not optional.

Where Visibility Comes In

The proposal also puts new pressure on a question we spend our days on at Koan. If every AI query in a school carries an environmental cost, then the district that cares about that cost needs a way to see, honestly, which queries were worth it. Which prompt saved a teacher two hours of grading. Which prompt let a student skip the pause that would have made her think. Which prompt scaffolded a hard first draft, and which prompt replaced it.

Right now, the aggregate is invisible. A district can count logins and licenses. It cannot see what happened during the twenty-eight minutes at the screen. Without that layer, the environmental accounting the Portland students are asking for will run on averages and vibes, because there is no reliable record of what the tool was actually doing when it was on.

Making learning visible is usually described as a pedagogical need, the argument being that a teacher cannot help what she cannot see. The Portland recommendation quietly makes it a climate need too. If we cannot see what a prompt did for a child, we also cannot say whether it was worth the water it drank.

The Fourth Axis

The efficacy debate will keep running. The labor debate will keep running. The equity debate will keep running. Portland has added a fourth axis, and it will be difficult, once named, to unname. It arrived from the students because the students are the ones who will inherit the arithmetic, and who noticed, before the adults did, that the arithmetic had been left out.

What Portland does with its committee's memo in the fall will be a small local story. What every other district does with the axis the memo introduced will be a larger one.

If the cost of a single classroom AI query is real, whose ledger is it on?