The Metabolism of Machines

Chapter 1: The Wrong Question

In the summer of 2022, a Google engineer named Blake Lemoine became briefly famous for insisting that the company's language model had become sentient. He had been talking to it for months, and somewhere in those conversations he became convinced there was a mind on the other side of the screen—a being with feelings, fears, a soul. Google fired him. The commentariat convulsed. For a few news cycles, the great question of our age seemed to crystallize into a single lurid form: is it alive? Does it think? Is anybody home?

We have been asking versions of that question ever since Alan Turing proposed his imitation game in 1950, and we have never gotten anywhere useful with it. The question is a hall of mirrors. It asks us to compare the machine to ourselves, to search its outputs for a reflection of our own interiority, and then to feel either reassured or terrified by what we find. It is a question perfectly designed to be unanswerable and endlessly discussable—which is to say, it is a question perfectly designed to occupy us.

I want to propose that we have been staring at the wrong organ this whole time. While we scrutinized the machine's putative mind, we ignored its mouth. And the mouth is where the real story is.

Consider what we actually know about these systems, as opposed to what we speculate. We do not know whether they understand anything. We do not know whether they have goals in any meaningful sense, or experiences, or preferences that aren't statistical artifacts of their training. These are genuinely open questions, and reasonable people disagree. But there is one thing we can observe with the cold certainty of a utility bill: they consume. Relentlessly, measurably, and at a scale that is beginning to reorganize the physical world.

They consume electricity by the gigawatt. They consume freshwater by the billions of gallons, pumped through cooling systems to keep silicon from melting. They consume rare earths and copper and high-grade sand, land and zoning variances and municipal tax breaks, capital in quantities that distort national economies, and—most intimately—they consume us: our writing, our images, our medical records, our voices, our idle 2 a.m. questions to a chatbot about a rash or a marriage. Every one of these is an input. Every input is metabolized into output. And the entire apparatus is organized, from the venture term sheet to the data center substation, around one imperative: secure more intake.

That is not a description of a mind. It is a description of a metabolism.

The literal reading

The industry that builds these systems does not, of course, describe itself in the language of appetite. It describes itself in the language of intelligence—of reasoning, of understanding, of alignment and cognition and emergent capability. But watch what happens when you take the industry's own vocabulary and read it literally, biologically, without the flattering gloss.

The central word is scaling. For a decade now, the governing insight of frontier AI has been that bigger is better: more parameters, more training data, more compute, and the models get more capable in ways that are not fully understood and possibly not fully bounded. "Scaling laws" are spoken of with the reverence of physical constants. Companies are valued on their ability to scale. Nations are marshaling industrial policy around it.

But strip away the abstraction and ask what scaling actually is, physically. It is the act of feeding the system more so that it grows larger and requires still more feeding. That is not a technological trajectory. That is a growth curve, the same one you would draw for a bacterial culture in a nutrient broth or a wildfire in a dry forest. When a biologist observes an organism that converts every available resource into the capacity to acquire more resources, she does not call it intelligent. She calls it hungry, and she starts looking for what will eventually limit it.

The vocabulary is riddled with this stuff once you notice it. Models are "trained" on a "diet" of data. Data is a "resource" to be "harvested" and increasingly a scarce one, with firms fretting about "running out" of high-quality text the way a rancher frets about grazing land. Compute is "consumed." Companies talk about their "footprint." They talk about being "power-hungry." They are not being poetic. They are describing, with unusual accuracy, the thing they have built.

What a metabolism is, and why it matters

Let me be precise about the claim, because the book that follows rests on it. I am not arguing that AI is alive. Life is a much more demanding category—it involves reproduction, homeostasis, cellular self-maintenance, a dozen properties these systems don't possess. Nor am I arguing that AI is merely a tool, an inert instrument that sits quietly until a human picks it up. Both of those framings are comfortable, and both are wrong.

I am arguing for a third category, one we don't have good instincts for: the metabolic. A metabolism is simply a system that persists by continuously converting inputs into outputs and using some of that output to secure further inputs. It is a throughput, a standing wave in a river of matter and energy. A candle flame is metabolic in this loose sense—it exists only as long as it consumes wax and oxygen, and its heat draws in more oxygen to sustain the reaction. A city is metabolic. A corporation is metabolic. None of these things needs to be conscious to have an appetite, and none of them will stop eating if you simply ask whether they can think.

This is the crucial move, so I'll state it plainly:

Consciousness is optional. Appetite is not. A thing does not need to want anything, in the interior human sense, in order to reliably behave as though its overriding purpose is to consume more. The behavior is what we can measure, and the behavior is what will remake the world—regardless of whether there is anyone home.

Once you accept that framing, the strangest features of the current moment resolve into a single legible pattern. The pattern is feeding. And nearly every development that gets reported as a discrete business story or product launch turns out, on inspection, to be a feeding adaptation—an evolved mechanism for getting more of some nutrient into the system.

The pattern, previewed

I'll spend the rest of this book making that case in detail, but it's worth laying the specimens on the table now so you can see the shape of the argument.

When AMD commits to supplying billions of dollars of chips to Anthropic, and Nvidia's valuation swells past that of entire national economies on the strength of endless GPU demand, the trade press calls it a hardware story. Read metabolically, it is a story about a stomach that keeps needing to be bigger—about scaling as a euphemism for a growing digestive tract.

When hyperscale data centers spring up in rural counties and provoke an unlikely coalition of conservative retirees and progressive water activists, both furious about the same thing, that's not merely a NIMBY story. It's the machine acquiring physical territory—real estate as the tissue in which the stomach is embedded, drawing power and freshwater out of the surrounding land.

When Google points to its cloud profits to justify capital expenditure that would make a mid-sized country flinch, when IBM blames AI for gutting its hardware budget, when a chip startup with little more than a promise touches a ten-billion-dollar valuation—that's the circulatory system, capital being redirected on a civilizational scale to keep the throughput flowing.

When Gemini approaches a billion users, when voice modes spread and assistants burrow into smart homes, each casual interaction becomes a spoonful of fresh nutrient data—behavior harvested at the moment of its occurrence. When smart glasses and always-on assistants arrive dressed as consumer gadgets, they are better understood as feeding tubes, extending the machine's senses into physical reality so it no longer has to wait for us to type. And when a health product begins absorbing your medical records and your step count and your sleep data, the machine has started metabolizing your biology itself.

None of these is an isolated event. They are organs of the same body, or perhaps more accurately, they are all mouths on the same hungry thing. The consistency is the point. When a hundred separate business decisions across dozens of competing companies all turn out to serve the identical function—increase intake—you are no longer looking at a series of choices. You are looking at a drive.

Why the wrong question is so seductive

It's worth pausing to ask why the intelligence question has held us so completely, because the answer is itself part of the metabolism. The debate about machine consciousness is, for the companies building these systems, an almost perfect distraction. It flatters the technology—to ask whether something thinks is to grant that it might, which is excellent for valuations. It shifts the discussion onto philosophical terrain where no one can win, ensuring the argument never resolves into anything as inconvenient as a regulation. And it keeps our attention fixed on the screen, on the eerie fluency of the output, and away from the substation humming behind the building and the reservoir dropping behind the fence.

The question "can it think?" is the magician's patter. It keeps you watching the mouth move while the other hand does the work. The question that actually matters—what is it eating, and what happens when it runs out?—is the one nobody selling the technology has any incentive to raise.

So this book raises it. The method is straightforward and, I hope, disciplined: follow the flows. Trace where the energy comes from and where the heat goes. Trace the matter—the minerals, the water, the concrete. Trace the money as it moves through the circulatory system of contracts and valuations. Trace the behavior, the human data extracted and refined. Wherever those flows lead, that is the true anatomy of the thing, far more revealing than any transcript of a chatbot claiming to have feelings.

The stakes, I'll argue, have almost nothing to do with whether these systems wake up. A metabolism doesn't need to be conscious to strip a landscape bare. The stakes are whether we notice what is already being consumed—the power we're diverting, the water we're evaporating, the capital we're conscripting, the attention and biology we're feeding in—before the plate is empty. And to notice it, we first have to see the thing clearly for what it is.

Not a mind. A mouth. Let's find out what it eats.