Surplus and the Shared Grid

Pooling is the oldest trick our species has. One fire warms a group. One kill feeds people who did not hunt it. One water wheel grinds for a valley. Each pooling raises the throughput available per person above what any person could produce alone, and each one buys the division of labour that makes cumulative culture possible.¹ Every technology in this book is downstream of somebody deciding to share a conversion rather than duplicate it.

And every pooling creates a valve. That is the part the celebratory account leaves out. A shared conversion has a physical location, a finite capacity, and a moment at which it is allocated. Someone performs the allocation. Chapter 3 established that dependency is the price of extension without standing. This chapter asks the structural version of the same question: at planetary scale, where exactly are the valves, and what would standing at each one look like?

The grid is a metabolism, and metabolisms are governed

An electrical grid is not a metaphor for a metabolism. It is one, in the strict sense the first chapter set out: a system that maintains an organised state by continuously converting a gradient and rejecting heat. It has an intake, a distribution network, storage that never quite suffices, and a bill that arrives whether or not anyone is watching. Like every metabolism, it can be starved, overloaded, or reallocated — and unlike a body, the reallocation is performed by identifiable people making decisions in rooms.

This is where Wittfogel's narrow claim earns its place in the argument rather than merely being asserted.² He was wrong that hydraulics determine political form; the archaeology and Ostrom's fieldwork settled that.³ He was right that when essential complexity passes through few enough physical points, the management of those points becomes the real constitution, regardless of what the written one says. The test of the claim is not whether it sounds ominous. The test is whether you can enumerate the points. So here they are.

Five chokepoints in the human–AI interface

Each layer below is a place where the shared metabolism narrows. For each I give the physical constraint, who currently stands at the valve, the specific way it fails the ledger of Chapter 3, and the lever that exists — not a wish, a lever that is presently available to engineers, regulators, or buyers.

The scale error that hides all of it

There is a single recurring mistake in public argument about this infrastructure, and it is worth naming precisely because it is made in good faith by careful people. The mistake is answering a basin-scale question with a national-scale number.

Data-centre water use is a small fraction of national freshwater withdrawal. That statement is true and it is not responsive. Agriculture dwarfs the total, which tells you nothing about whether a specific campus competes with a specific municipality during a specific drought. Worse, most of the water is not drawn on site at all: for most U.S. facilities the electricity-related draw exceeds the cooling draw, which means a site can report an excellent water-use effectiveness while its true footprint sits upstream in somebody else's basin.⁴

A first view from the valley floor

I was raised between San Francisco and a nearby suburb, and I did not know, the first time I entered Yosemite Valley, that I was walking into the middle of a water war. I knew only that the air felt different, that the walls of stone held the sky at a reachable distance, and that a stream ran through the meadow with a sound I had not heard in the city. My whole being felt contained, as if the valley were a room with no ceiling and no wish to leave it.

No one had told me that the water falling over those cliffs had already been claimed, re-claimed, diverted, and fought over for a century. No one had explained that the same gravity pulling the Merced River through its gorge had once pulled the Owens River south until a city hundreds of miles away dried the lake at its source. I did not need the explanation. The valley gave me something simpler and more urgent: the sense that water, stone, and light could be enough. The wars came later, in books and documents and the dry anger of people who had lost what I had only just seen.

That is the scale error from inside. To the child standing in the meadow, the water is a gift. To the ledger, it is an allocation. Both are true, and neither cancels the other. But the first feeling is the one the second feeling depends upon: if the valley had not contained me, I would not have cared who held the valve.

The work of this chapter is not to make anyone feel guilty for loving a valley. It is to keep the love and the ledger in the same frame, so that when we build the next shared metabolism — the next grid, the next cooling loop, the next training run — we do not build it in a way that requires a later generation to learn, too late, that the water they loved had already been seized.

The pattern is not theoretical. In California, the same structural move has been made twice at two different scales. In the early twentieth century, Los Angeles acquired Owens Valley water rights through methods one historian described as "chicanery, subterfuge, and a strategy of lies," diverting the river so completely that Owens Lake dried by 1926.⁸ A century later, documented reporting describes how a single billionaire couple acquired control over a large share of the state's water system through private meetings, converting publicly financed infrastructure — including the Kern Water Bank, built with hundreds of millions of taxpayer dollars — into a private holding while urban water systems received a fraction of what they requested during drought.⁷⁹ The point is not the individuals; it is the structural form. When a shared metabolism narrows to a few private valves, the governance question becomes who owns the valve, not who uses the water.

If there is a grim joke in the history, it is that every generation builds the valve and then professes surprise that someone stands at it. The surprise is itself part of the scale error: look at the national average and the valve disappears; look at the basin and it is the only thing that matters.

Electricity shows the same pattern in reverse. Globally, data centres remain a low single-digit share of demand. Regionally, they have reached shares of national and corridor demand that no other single industrial category holds.⁵ Ireland and the Northern Virginia corridor are the standing examples. The figures move every year and I will not anchor an argument to one of them. The concentration is the durable fact.

A quantity averaged across a nation cannot detect a chokepoint. That is not a limitation of the data. It is what averaging is for.

Once you see the scale error you see why it persists. Every party with an interest in the chokepoint has an interest in the national number. It is not a conspiracy; it is a gradient, and arguments roll downhill the same way water does.

What Ostrom would ask of a data centre

Ostrom's design principles were derived from irrigation systems that survived for centuries without a central authority, and they map onto this infrastructure with very little strain.³ Stated as questions a regulator, a buyer, or an engineer can actually ask:

• Clearly defined boundaries. Does the facility's draw — electricity and both water pathways — have a published figure at the basin and feeder level, not the national one?
• Congruence with local conditions. Is the cooling design matched to the water status of the basin it sits in, or imported unchanged from a wetter site?
• Collective-choice arrangements. Do the people whose water and power it shares have standing in the permit, or only a comment period?
• Monitoring. Is consumption metered by someone other than the operator, and is the count visible?
• Graduated sanctions. Is there any consequence short of shutdown for exceeding an allocation during a drought or a peak?
• Conflict resolution. Is there a forum where a municipality and an operator meet as parties rather than as applicant and objector?
• Nested enterprise. Does regional capacity exist, or does every failure travel continentally?

None of these is exotic. Every one is a thing some jurisdiction already does for water, and none is currently standard for compute. The gap between those two sentences is the governance deficit this chapter exists to name.

Efficiency as the practical form of self-government

Chapter 2 argued that brute force is the first item off the ledger because it is the most tractable to deliberate choice. The chokepoint analysis gives that argument its second and stronger leg.

Every watt and every litre a system does not require is a lever nobody gets to hold over it. Architectures that unify memory and compute, sparse and event-driven methods, higher operating temperatures, closed-loop cooling, regional capacity in place of a few continental campuses — these lower the bill, and in lowering it they widen the set of hands that can pay it. Substitutability is manufactured by efficiency. There is no other way to manufacture it.

This is also why the physics matters and cannot be left to a later volume. Low-energy resonance approaches — commanding a system by finding its invariant rather than by overwhelming it with throughput — are not gentler versions of brute force.⁶ They are the only class of method that produces capability without producing a valve. A civilisation that learns to do more with less has not merely economised. It has removed the physical basis of the thing Wittfogel was afraid of.

The prediction

One claim, stated so it can be checked. A chokepoint is a low-energy position from which to command high-energy throughput. Under the accounting used throughout this book, such positions are the cheapest available source of leverage in any complex system — and cheap leverage attracts whoever is least constrained by the costs borne downstream. This is not an allegation about anyone. It is a statement about a niche: build a valve and the valve will be filled, and it will not reliably be filled by the person who cares most about the people below it.

The remedy that follows is structural rather than moral. Do not attempt to select better valve-holders. Build fewer valves. Where valves are unavoidable, meter them publicly, nest them regionally, and give the downstream parties standing before the concrete is poured. That is the ledger which sustains the warning, and it is the only part of this argument that is actionable this decade.

Chapter 5 turns to tempo — why genetic change cannot arrive in time to solve any of this, and why that fact makes the architectural question the only one left open.