No Disassemble 4: The Recalibration

Or: How I Learned to Stop Comparing Almonds to AI and Start Asking Who Decides

One Month Ago:

I was arguing about whether AI images were "real art." That felt urgent. It felt like the center of the conversation.

Today:

I am writing about aquifers. I am writing about zoning variances. I am writing about data center cooling loops and the fact that a single large data center can evaporate more water in a day than a family of four uses in a decade. I am writing about the Lenape, who managed the Delaware Valley for ten thousand years without draining it, and about Pennsylvania, which does not officially recognize any Native group within its borders, and which is currently rubber-stamping the next data center campus that will draw down the same watershed the Lenape kept alive for a hundred centuries.

I am writing about the question nobody in the internet argument is asking:

What are we actually optimizing for?

---

The Problem with Internet Arguments

This all started with a very simple question:

"How much water does AI use?"

That is a reasonable question. I asked it myself. I got numbers, and I put them in an infographic. The infographic looked something like this:

---

HOUSEHOLD

80–100 gallons per person per day

300–400 gallons per family of four per day

DIGITAL

20 AI prompts ≈ 0.13 gallons

1 hour streaming ≈ 0.13–3.17 gallons

FOOD

1 almond ≈ 3.2 gallons

1 lb chicken ≈ 500 gallons

1 lb beef ≈ 1,800 gallons

INFRASTRUCTURE

Typical data center ≈ 300,000 gallons/day

Large data center ≈ 5,000,000 gallons/day

---

Those numbers are real. The sources are solid: EPA WaterSense, Mekonnen and Hoekstra's foundational work on water footprints, Li and Ren's 2025 paper on AI water consumption. The infographic is honest. It recalibrates. It forces both sides to stop and look at what the numbers actually are.

But here is what I learned:

The infographic is not the argument. The infographic is the ruler.

Most internet arguments happen when people compare different scales and call it the same thing. Someone says "AI uses too much water!" Someone else says "almonds use way more!" Both are true. Both are irrelevant. They are comparing a per-query sip to an agricultural firehose.

One almond contains roughly the same water footprint as 25 days of AI use at 20 prompts per day. One pound of beef contains roughly the same water footprint as 14,000 days of AI use. But what does that tell you? It tells you that food and computation are different systems. It does not tell you which one is "bad." It tells you they are not the same thing.

---

The Real Question Emerges

If you keep pulling on the thread, you eventually get to something that is not about water at all.

The water question was the doorway. The governance question is the room.

Water alone does not tell you what matters:

System Produces

Households People

Food Nutrition

Three Sisters Resilience

Hopi Blue Corn Cultural continuity

Libraries Knowledge access

Universities Research

Streaming Information & Entertainment

AI Computation

Data Centers Digital infrastructure

Those are not the same output. They are not optimizing for the same thing. Comparing them only makes sense if you first answer a harder question: what are we trying to achieve, and who decides?

Water footprints tell us what we use.

Productivity tells us what we get.

Optimization tells us what the system values.

Governance tells us who decides.

---

The Governance Question Is Not New

The Lenape inhabited the Delaware Valley for roughly 10,000 years. Their existence demonstrates that different optimization targets are possible.

They practiced sophisticated fire ecology. They cultivated the Three Sisters: corn, beans, and squash—each plant supporting the others, none extracting without returning. They operated on the seventh generation principle: decisions involving the land must account for their effects seven generations into the future.

That is not poetry. That is a longer discount rate than any data center lease agreement or municipal zoning variance in the history of the Commonwealth of Pennsylvania has ever used.

Pennsylvania is one of a handful of states that neither contains a reservation nor officially recognizes any Native group within its borders. It is the only state without a university-level Native American studies program or cultural center.

The state that sits on top of 10,000 years of sophisticated ecological management cannot be bothered to acknowledge it happened, let alone learn from it, while it rubber-stamps the next data center campus that will draw down the same watershed the Lenape kept alive for a hundred centuries.

This is not a coincidence. This is the operating system.

The Lenape example is not a historical sidebar. It is a case study in governance. It shows that the question "who decides what the water is for" is not new. It has been asked, and answered differently, for thousands of years. The answer we are giving now is not the only possible answer. It is simply the one we are choosing.

---

The Chain That AI Did Not Invent

One of the things that became clear during this investigation is that digital infrastructure did not begin with generative AI. The chain is not:

Food → Streaming → Search → AI → Data Centers

It is more like a layered system:

Households

↓

Food Systems

↓

Digital Services

↓

Streaming / Search / AI

↓

Data Centers

↓

Water + Energy Infrastructure

Streaming was using data center water long before ChatGPT existed. Your Netflix habit and your Facebook scrolling have been drawing on aquifer water for years. AI just increased demand on the same infrastructure that was already running. AI did not invent the water question. It amplified an existing one. It made visible something that was already happening.

If you are angry about AI water use, compare it to streaming. Compare it to food. Compare it to household use. Compare it to data centers. If you are dismissing AI water use, compare it to local municipal supplies. Compare it to data center concentrations. Compare it to electricity generation water.

Per-user AI water use is tiny. AI infrastructure water use is not tiny. Food production uses vastly more water than individual AI use. Data centers can create real local water stress. Streaming was using water long before generative AI existed. Most comparisons depend on what scale you are measuring.

All of those statements are simultaneously true.

---

The Lead-Time Gap

Part of what emerged from this investigation was the realization that infrastructure systems often know what is coming long before the public does.

Major projects leave footprints. They require property transfers, LLC formations, utility studies, environmental reviews, noise variance requests, transmission planning. These events often occur months or years before widespread public awareness.

I started calling this interval the Lead-Time Gap.

The Falls Township data center project, for example, left detectable signals before most residents had any idea what was happening. The same pattern applies to data center development anywhere. If citizens wish to participate meaningfully in decisions that affect their communities, they must learn how to recognize those signals.

Here is a simple checklist for anyone trying to see the future:

· Property transfer records

· New LLC registrations

· Planning commission agendas

· Zoning hearing board notices

· Conditional-use applications

· Noise variance applications

· DEP environmental permits

· DRBC water withdrawal permits

· Utility infrastructure filings

· PECO substation projects

· PJM transmission planning documents

The earliest signal of a future project may not mention a data center at all. It may appear as a utility study, a zoning amendment, a variance request, or a property transaction.

---

What John Henry Knew

I wrote earlier about John Henry racing the steam drill and dying. I said the story is not "machines bad." It is about the fear that the people controlling the machines have decided you are a cost to be minimized.

That fear is back. It is at infrastructure scale. The old questions about labor and displacement are real, and they intersect with new questions about ontology and personhood in ways we haven't fully mapped.

The argument about whether AI is "art" was really about dignity. The argument about data centers is about survival. One is about whether the work matters. The other is about whether there will be water in August.

---

The Recalibration

What I am offering is not an answer.

It is a recalibration.

It is not a manifesto. It is a ruler. It is a thing you can hold up to the numbers and say: "Oh. That is where that sits." It does not tell you which system is "best." It tells you that they are not optimizing for the same thing.

If you are optimizing for... Look at...

Protein efficiency Chicken, not beef

Cultural resilience Hopi Blue Corn

Knowledge access Libraries

Computation AI

Digital infrastructure Data Centers

---

What to Read Next

This is the fourth post in a series. If you are coming in late, here is the order:

Part 1: John Henry, Photoshop, and the AI Question

A measured essay about AI anxiety, dignity, and the pattern of moral panic around new tools.

Part 2: No Disassemble 2: Data Center Boogaloo

The moment I realized the conversation was no longer about art and had become about municipal zoning, aquifers, and industrial infrastructure.

Part 3: No Disassemble 3: The Lead-Time Gap

An investigation into how major infrastructure projects become visible—and how to see them coming.

Part 4: No Disassemble 4: The Recalibration (This post)

Where we finally ask: what are we optimizing for, and who decides?

---

The Point

The numbers are real. The scale matters. The comparison matters. Most internet arguments fail because people compare different scales and call it the same thing.

Water footprints tell us what we use.

Productivity tells us what we get.

Optimization tells us what the system values.

Governance tells us who decides.

The future probably is not "humans OR AI." It is much more likely humans working with increasingly powerful cognitive tools while society struggles to adapt ethically, economically, and culturally.

I am enthusiastically rooting for humans and AI to educate each other and work together to fix as many problems as we can. That adaptation process is going to define a lot of the rest of this century.

But first, we have to ask the right question. Not "how much water?" That is Layer 1. The internet is stuck at Layer 1.

The real question is Layer 4: "Who decides what the future is built for?"

---

Because every gallon, every acre, every kilowatt, and every dollar is ultimately a vote for the future we are choosing to build.

That is what this piece is. Not an article about AI. Not an article about water. An article about how decisions become reality.

Which is why, after all this research, I ended up somewhere completely different from where I started.

I started by asking:

"How much water does AI use?"

I ended by asking:

"Who gets to decide what the water is for?"

That is the actual story.

---

*This post synthesizes research from EPA WaterSense, USGS, Mekonnen & Hoekstra (2011, 2012), Fulton et al. (2018), Li/Ren et al. (2025), Obringer et al. (2021), LBNL (2024), Brookings (2025), and the Delaware River Basin Commission. The infographic it refers to is available as a standalone resource. All numbers are approximate. Methods vary by source. Direct water use unless noted.

**Averaged numbers do not show dry wells. They do not show who gets water and who does not. They do not show the difference between a wet year and a drought year. They are a ruler, not a complete picture. That is why the post ends with governance. The numbers tell us what we use. The governance question tells us who decides, and who bears the cost.