What Gets Expensive When Everything Is Cheap
Living through AI cost disease, interest-rate whiplash, and the last human scarcities
After the storm, the model on my laptop wrote a diagnostics script in three minutes. It parsed the logs, flagged a suspect relay, and even drafted a polite note for the building manager. Feeling competent, I called an electrician. He quoted triple his usual rate and couldn’t come for two days. I agreed before he finished the sentence. That evening, the grocery bill was strangely gentle, my music and movies were bottomless, and an assistant that never sleeps had thinned my inbox. Yet childcare still cost more than a mortgage. A two-bedroom within commuting distance felt like a collector’s item. I went to bed with the old, embarrassing fear, not that AI will take my job, but that a human life could be priced out of its own city.
This is the new arithmetic. The closer a thing is to symbols, the cheaper it gets; the closer it is to bodies, land, and guarantees, the dearer it becomes. Text, pixels, and repeatable processes collapse toward zero. Embodied time, local matter, and outcomes someone must stand behind climb a ladder with missing rungs. We are living through cost disease with a jet engine. The miracle at the edge, abundant content and code, throws the core bottlenecks into relief: power, space, hands, and promises.
GDP will not warn you. Measured output can look calm precisely when deflation in the virtual floods the numerator while inflation in the physical and guaranteed hides in your rent, your grid, your daycare invoice, your bid for an actual human to show up. It produces numeraire drift, a quiet disorientation in which life is benchmarked to what models can generate rather than to the goods that anchor a day. You will be richer in everything that streams and poorer in everything that stands.
The consequences are not theoretical. Wages spike where automation stalls: electricians after storms, line cooks at dinnertime, midwives on the night shift, grid operators in heat domes, teachers on Monday mornings. Meanwhile, the markets that price tomorrow start to look like hurricane maps. Capital stampedes into a few moats, compute, energy, land near substations, and deserts the median. Expectations of runaway growth collide with a collapsing desire to save, so rates rise simply to pry capital loose; asset prices fall even as the economy sprints. Means can explode even as lives tighten. It feels like standing before an open fire with your pockets turned inside out.
What actually gets expensive when everything is cheap is not content; it is consequence. Anyone can emit plausible text; few can bind outcomes. The premium shifts to verification, receipts, audits, insured delivery, to custody, who owns the power and the land under the data center, to attention, who can hold the hinge of a human day against the feed, and to care, time-intensive labor that cannot be batch-processed without breaking what it claims to serve. This essay is an attempt to name these last scarcities and to choose, deliberately, how to live among them.
The thesis is simple and a little impolite to our optimism: AI will make most things cheap and some things brutally expensive, and the price map will not match our intuitions. If we measure ourselves by the infinite supply of what machines produce easily, we will feel rich and grow brittle. If we reweight our lives toward the finite, land, energy, verification, attention, and care, we may feel poorer in spectacle and wealthier in days that hold.
I want to learn, and teach, how to navigate this split reality without panic or denial. That means reading mispriced signals in the market when explosive growth coexists with falling asset prices; valuing outcomes over output, guarantee over gloss; and investing not only money but allegiance in the scarcities that still make a human life a life. It also means small, unfashionable moves: spending less on the infinite and more on the local; favoring dull businesses with land and kilowatts; paying a premium for work that arrives with receipts; and protecting the hinge of attention as if it were a civic utility.
Abundance at the edge, scarcity in the core. If that sentence is true, then wisdom is the art of allocating those last scarcities, land, energy, verification, and attention, without losing wonder. In the next part, I will explain why GDP keeps lying to our gut, how cost disease with a jet engine hides in the averages, and what to watch when life grows more expensive in all the ways that count.
The Accounting Error
GDP is a camera that loves the light. It overexposes what is abundant and cheap because cheap things multiply, and it underexposes what is scarce and stubborn because scarcity hides in shadow. When AI floods the world with text, images, and software, measured output soars on paper, even before anyone pays, because statisticians “quality-adjust” the deflators as best they can. Meanwhile, the prices that hold a human day together, rent near work, a slot at daycare, an electrician after a storm, refuse to budge or surge the wrong way. The average looks calm. Your life does not.
Part of the misread is mechanical. GDP measures production, not distribution; totals, not tensions. If models can produce a billion near-perfect drafts for almost nothing, the national accounts smile. Those drafts do not, however, lower the price of verified childcare, a safe neighborhood, or kilowatts when the grid is hot. You experience a turbocharged cost disease: deflation where you stream, inflation where you stand. Averages reconcile the two; bodies do not.
The price indexes add a second confusion. Deflators are good at counting identical apples; they wobble when the apple becomes a simulation. Hedonic adjustments try to say “this year’s model is better, so its true price fell,” but those guesses are softest precisely where AI works hardest. The accounts will faithfully record that software is cheaper and more capable. They will not tell you that the guarantee on outcomes, the part someone will insure, sign, or show up for, has grown dearer. GDP is fluent in output. Your budget lives in consequence.
Then comes numeraire drift. We start valuing ourselves in units of what the machine prints. If the denominator of your self-assessment is “content per day,” you are the richest person who ever lived. If the denominator is “square feet within a bus ride, an adult you trust with your child, a technician on Wednesday,” you feel poor in the ways that matter. We quietly switch the yardstick and wonder why the room feels wrong.
There is also a geometry problem. Value created in the cloud must land somewhere on Earth: in the land under data centers, in substations, in cooling, in skilled hands that fix what power makes possible. These inputs are lumpy, local, and slow. National accounts smooth them; you pay them at retail. You can live through a year in which “real GDP” whispers stability while your neighborhood gentrifies around transformers and your rent bakes in a compute premium you never asked for.
Even the “free” surplus of digital goods misleads your nerves. Economists rightly say you are richer because you get more for less at a price of zero. Your nervous system rightly answers that free abundance taxes the scarce hinge of your day, attention, and the tax is paid in worse sleep, fractured focus, and buying back quiet with cash. None of that fits cleanly in a price index. Your calendar is absorbing a cost your CPI cannot see.
This is why a boom can make your portfolio seasick. Expectations of runaway growth crush the appetite to save; rates rise to pry capital loose; long-duration assets rerate downward even as the economy sprints. The index says “progress.” Your mortgage reset says “brace.” Both can be true. Measures built for a world where productivity meant factories and trucks repaint poorly when productivity means GPUs and promises.
So watch different dials. Track the price of guaranteed outcomes, services that arrive with receipts, liability, and a signature. Track energy per verified action, the kilowatt-hours required to deliver something you would put your name on. Track land near power, rents and easements around substations and fiber. Track care wages, the cost of entrusting your child, your parent, or your convalescence to a human whose time cannot be parallelized. These are the shadow gauges of an AI economy. When they tighten, your gut is not wrong; your dashboard is outdated.
If GDP is the skyline, these are the load-bearing walls. In the next section, I will follow the social current that always flows toward scarcity: as cognitive piecework gets cheap, prestige migrates to hands, heat, and presence, and plumbers, midwives, grid operators, line cooks, and teachers move from backdrop to front row while the culture quietly inverts.
The Prestige Inversion
As models saturate the easy half of knowledge work, status begins to leak toward what cannot be cached: hands that fix, bodies that stay, people who can keep a room at the right temperature, literal or human, when the system runs hot. The culture has not caught up. We still award blue checks to pundits who remix PDFs while the person who can rewire a panel after a surge remains a name on a clipboard. Scarcity drags prestige behind it, eventually. In an AI economy, the scarce traits are not cleverness and speed; they are touch, steadiness, and custody.
Touch is not romance; it is fidelity to matter. A plumber who diagnoses a dying pipe stack from the sound in a wall delivers an outcome no autocomplete can guarantee. A midwife who knows when silence is the right instruction is practicing a craft with consequences. A line cook who keeps a Saturday service from spinning out is doing real-time systems engineering with sweat instead of dashboards. When the marginal cost of content falls to near zero, the price premium shifts to consequences you can eat, wear, plug in, or entrust your child to.
Steadiness is the opposite of mercurial intelligence. A kindergarten teacher who can land thirty small nervous systems at 8:15 on a rainy day is running a control loop with unknown inputs and high emotional stakes. A grid operator in a heat dome is not optimizing a slide deck; she is balancing scarcity, physics, and risk in a domain where the adjective “black” precedes “out.” These roles are not “unskilled.” They are unforgiving. That is why they will re-price.
Custody is the new moat. Who is physically present when responsibility lands? Who holds the keys to rooms with breakers and babies? Who signs the form that transfers liability from talk to guarantee? The answer earns the rent. In a city where AI writes legalese for free, the person who can legally witness, certify, and stand behind outcomes becomes the bottleneck. We are moving from an economy of opinions to an economy of receipts with names.
I can see the inversion in my own calendar. Meetings with people who make decks left me wired and empty. Afternoons with people who make food or keep servers cold left me quiet and respectful. The difference was not moral superiority; it was risk profile. The cook and the operator live closer to failure modes that do not care about rhetoric. They work in domains where feedback is immediate and reputations update in hours, not press cycles. AI will write a thousand apologies. It will not keep a walk-in from thawing.
The shift will be awkward because our status software is out of date. The résumé flatters cognitive piecework. The admissions letter selects for abstract virtuosity. The platform rewards opinions that age like fruit. Meanwhile, the jobs we cannot do without look bad on LinkedIn and have no glossy words to hide behind. That mismatch already punishes us: schools without enough aides, hospitals without enough nurses, cities where small electrical faults wait weeks because everyone with a license is booked. We call it a labor shortage. It is a prestige shortage disguised as economics.
What follows is not nostalgia but new guilds with modern joinery: portable credentials that travel across states, outcome-based pay that reflects load and risk, insurance pools that price skill correctly, and career ladders that do not force craftspeople into management to be respected. We also need story repair. The language we use for “trades” and “care” must grow teeth. Stop calling them soft. They are hard in the only way that counts: failure hurts someone you can name.
If you want a personal hedge against AI’s deflationary glamour, learn to do one difficult thing in matter or with bodies that other people depend on. Know a breaker panel, a garden, a sourdough, a neonatal protocol, a horn section, or a server rack’s cooling curve. If your hands will always type for a living, ally with the people who do these things: invest in their shops, build tooling that removes paperwork rather than dignity, and give them scheduling power instead of dashboards that grade their compassion.
Prestige inverts slowly, then all at once. First comes the price signal as wages rise where automation stalls. Then comes the narrative as admiration migrates from clever talk to reliable delivery. Then comes the pipeline as bright kids seek apprenticeships not because school failed them but because reality rewards them. The market can start the process; culture must finish it. We will know the inversion has stuck when a parent brags about a child stabilizing an ICU or running a substation with the same pride once reserved for “went to McKinsey.”
Verification as the Value Add
The longer I live with models that can pour competent text on command, the more I pay for something dumber and rarer: a promise that survives contact with the world. Anyone can emit output; few will bind themselves to an outcome. Value migrates into that gap. In an AI-saturated market, verification is the luxury good. The premium attaches not to what you say you will do, but to the receipt you staple to the thing once it is done.
I learned this the week two contractors sent bids. The cheaper one wrote, “We’ll fix it.” The pricier one attached warranty terms, a schedule with penalties, photos of comparable work, proof of insurance, and three references who answered on the first ring. The second number made me wince. I signed anyway. What I bought was not skill as such; it was accountability in a form that could travel, to a neighbor, to a court, to my future self at 2 a.m. In a world drowning in plausible claims, the scarcest thing is something you can enforce.
This is where the cultural conversation about AI misses the point. We keep arguing about output, who drafts faster, who remixes better, who stacks more tokens. The market we actually live in buys guarantees. Restaurants sell dinner plus food safety. Hospitals sell procedures plus malpractice coverage. Utilities sell kilowatts plus reliability curves. The margin is in the plus. Even in white-collar work, clients worth keeping are not paying for decks; they are paying for decisions they can live inside. If AI cheapens the draft, it does not cheapen the moment a human must sign.
Verification is not bureaucracy; it is the engineering of trust. A good receipt answers three questions cleanly: what happened, under what policy, and who is on the hook if it fails. Translate that into software and you get input digests, policy hashes, replayable traces, and a named signer. Translate it into construction and you get stamped plans, bonded timelines, and a foreman who answers his phone. Translate it into care and you get credentials you can verify and protocols you can audit, not inspirational posters. The forms differ. The function is the same: consequence with custody.
If this sounds unromantic, notice what it buys back. Verification lets you keep speed without turning your life into a slot machine. When we can prove at line speed, we can decide at line speed. When outputs carry their own receipts, sources, tests passed, liability attached, we do not have to slow the world to a committee’s pace. The alternative to proof is not freedom. It is theater followed by accidents.
Metrics will be gamed; Goodhart’s Law predates any dashboard. That is why receipts must be expensive to fake and cheap to check. In code, that means property tests that fail loud, tamper-evident logs that bind inputs to actions, and small sandboxes that replay decisions without bravado. In physical work, it means third-party inspections, serial numbers you can scan, and warranties backed by an entity with a balance sheet. In care, it means supervision that looks like presence, not paperwork. None of this is a vibe. It is infrastructure for promises.
Energy hides in this, too. A verified outcome eats watts. A thousand drafts cost a trickle; one guaranteed decision costs a meal. The right unit for an AI-native organization is not tokens per day but kilowatt-hours per verified outcome. When that number falls while incidents stay flat or improve, you are compounding. When it rises while dashboards look pretty, you are telling yourself a story. Verification turns energy from a vague ESG paragraph into a limit you can design against.
Reputation will evolve to match. The résumés I trust less each year are lists of outputs: roles, decks, threads. The ones I trust more are portfolios of receipts: the outage resolved with a timestamped postmortem; the contract delivered with penalties actually paid once; the ICU shift where three patients stabilized and a nurse who trained me will vouch for it. In a world where anyone can sound like an adult, adulthood is proven by whose name stands next to the work when it shakes.
If you build, design the receipt into the product. Do not ask a buyer to trust you because the demo sings; ship the attestation with the API call. Bind every automated action to a small bundle of evidence, input digest, policy version, tests passed, signer, and make failure local and loud. Reject outputs that lack provenance the way a compiler rejects a bad type. When you reduce friction at the edge, add discipline in the core, or you are minting debt.
If you invest, underwrite guarantees, not gloss. Ask for the verification strategy, not the content roadmap. Who insures the promise? How do receipts flow through the stack? What happens when the first big failure lands? The premium lines flow to firms that bind outcomes with less pain than incumbents, not to those that auto-generate more artifacts for the same meeting.
If you are a citizen, pay a visible premium for work that arrives with accountability and treat it as a civic act. Tip the childcare worker. Hire the contractor with the bond. Switch to the clinic that publishes outcomes. You are not just consuming; you are voting for an economy that pays the people who keep reality from wobbling. Wonder is free. Wisdom carries receipts.
The test I keep is crude but clarifying: how many hours this week did I spend producing verified outcomes? Not meetings, not drafts, not opinions that wandered through a feed and nodded at me. Outcomes I would sign for. Models cannot automate this scarcity because they do not carry liability; we do. That is not an argument against AI. It is an argument about where to stand so the wind helps.
From verification, the path bends toward money’s weather. When growth expectations go vertical and the appetite to save collapses, interest rates rise just to pry capital loose. Asset prices can fall even as the world speeds up. In the next section, I will put that nausea into plain language: interest-rate whiplash as civic physics, and how to live through a boom that feels like a squeeze.
Interest-Rate Whiplash
The week my portfolio got seasick, nothing “bad” had happened. Earnings were strong, models were better, the economy was humming. Then a line I had trained myself to ignore, an interest rate buried in a chart, tilted up a few notches, and my long, sensible positions sagged like wet canvas. It felt wrong until I remembered the oldest translation in finance: rates are the price of patience. When everyone expects tomorrow to be much richer and no one wants to lend today, the price of patience rises until someone does. Explosive growth can coexist with falling asset prices because the time between cash flows suddenly matters again.
In an AI boom, the desire to save collapses for two reasons that rhyme. The first is euphoric: why hoard cash if future income will explode? The second is competitive: why hold safe assets if you must chase the winners now? Meanwhile the world needs brutal amounts of investment, energy, land, factories, cooling, to support the very growth everyone is counting on. To pry capital loose from reluctant savers, rates climb. The climb does two unfriendly things at once: it punishes duration as long-dated assets mark down, and it re-ranks projects by near-term cash yield. A boom starts to feel like a squeeze because the economy sprints while your financing terms start doing hill repeats.
This is not an economist’s parlor trick; it is civics. Rates are a public signal that redistributes when pain arrives. Floating-rate debtors feel it first; young firms with distant profits feel it next; the cash-rich with pricing power feel it last. An election year under high rates is not just about inflation; it is about time preference. Households with adjustable mortgages become sudden militants about policy. Cities that levered up on variable debt find their budgets breathing through a straw. The most accurate political poll is sometimes a refinancing calendar.
Central banks live in this crossfire. Growth expectations lift demand ahead of supply; prices wobble; institutions must decide whether to move fast enough to be hated now or slow enough to be hated later. If they lag, inflation rides the confidence wave, “I’m richer soon, so I spend now.” If they surge, assets rerate downward and every CFO discovers that secular growth does not immunize you against a quarterly covenant. Either way, the civic physics are the same: interest rates are how a society prices its appetite for the future against its appetite for the present.
I stopped treating rates like weather and started treating them like terrain. On flat ground you can lean back on narratives and let duration carry you. On hills you shorten your stride. In practice that meant less poetry and more cash flow: trimming exposures whose beauty lived entirely in the outer years; adding dull businesses that throw off money now; preferring obligations I can refinance on my own calendar; refusing adjustable-rate temptations because I fancy myself nimble. I am not. I am human. My edge is preparation, not reflexes.
It also changed how I read “growth.” I stopped asking “how big can it get?” and started asking “how soon does it pay for itself?” and “how is it funded at the new price of patience?” Companies that looked strongest in my old lens, feature-rich, revenue-light, heroically visionary, looked brittle when I overlaid the financing map of their next two years. A founder who can talk compute but not kilowatts, a roadmap full of users but empty of receipts, a capital plan that assumes the Fed will collaborate with the deck, these are not sins, just mismatches with the physics of the moment.
There is a quieter, domestic version of this lesson. When rates jump, the most humane investment is predictability. Fix what you can fix. Lock what you can lock. The freedom you buy is not infinite upside; it is the right not to renegotiate your life every time a central banker clears a throat. In my house that looked like refinancing even when it hurt, staggering renewals, and building a small boring bucket that sleeps through drama. I do not admire it. I rely on it.
None of this argues against the miracle. It argues for a better map. In a world where means expand faster than patience, place yourself where the time tax is survivable, closer to cash-generating consequence, the parts of the stack that get paid to reliably deliver outcomes, rather than out on the option-generating output that needs cheap time to feel brilliant. The market will still produce legends. It will also produce beautiful stories that cannot clear the new price of later.
From here the weather gets stranger. When patience is expensive and growth is uneven, indices become less truthful, dispersion explodes, and capital moves like a storm seeking warm water, compute, energy, land, and the few firms that bind outcomes instead of emitting artifacts. That is where we go next: markets in hurricane mode, and why the death of the median is not a crisis of faith but a change in climate.
A Philosopher Lecturing on the Orrery, Joseph Wright of Derby
Markets in Hurricane Mode
In a gentle climate, the index tells a decent story. A rising tide lifts most boats, factor models smooth the wrinkles, and diversification passes for prudence. In a storm, the index lies. Dispersion explodes. A few hulls find warm water and grow monstrous while the median takes on spray. An AI boom is not a breeze across all sectors; it is a heat engine that powers a handful of feedback loops, compute, energy, land near power, firms that bind outcomes, and leaves the rest bobbing in their wake.
Winner-take-most is not a moral plot twist; it is the math of bottlenecks. As models improve, demand for inference rises non-linearly. As demand rises, the constraint jumps to inputs you cannot conjure with a sprint, megawatts, water, substations, permits, skilled crews. The firms that secured those inputs early do not merely run faster; they start to define the weather. Their cost curves bend away from the pack because they are not buying spot patience at spiking rates; they are harvesting long-dated agreements that function like time machines: power-purchase contracts, interconnection queue precedence, municipal relationships, water rights, leases with expansion options, rural easements with fiber. In this regime, alpha is procurement with foresight.
I felt this in my bones the month I passed on a dull, infrastructure-heavy company because the deck was allergic to adjectives. No moonshots, just diagrams of dirt, pipes, and kilovolt lines. Weeks later a clever “can’t-miss” AI tool I had praised publicly was crushed by a problem no demo could solve: their data center sat behind a substation queue measured in presidential terms, and summer heat turned their latency promises into poetry. The dull company kept compounding because their most important slide was a map with signatures. Own the gate, or rent the line.
Markets amplify this asymmetry. The winners will look briefly overvalued and remain so because their moats are physical and slow. You cannot fast-follow a transformer. The almost-winners will look cheap and stay cheap because the missing pieces, permits, megawatts, water, do not respond to cleverness on a quarterly cadence. The rest will oscillate with influencer weather. If you insist on reading the world through the index, you will feel gaslit. “AI is booming” can coexist with “my diversified portfolio is exhausted” because median exposure has become a liability. The action has compressed into a few corridors where inputs clear.
The pattern extends beyond compute. In care, chains that can staff reliably and publish outcomes will pull price away from boutiques that sell vibes. In construction, contractors with bonded crews and inspected receipts will be fully booked while brand-forward design shops watch inbound slow. In software, vendors that staple attestation to every automated action will win procurement while feature factories drown in proofs-of-concept. The storm rewards custody of consequence, not the prettiness of output.
This climate also breaks our comfortable hedges. In a normal cycle, you diversify across factors and sleep. In this one, your hedges share a hidden exposure: time. Duration gets punished by rates; narrative gets punished by dispersion; correlation spikes when the market tests who actually secured inputs and who assumed they would be available. The only durable hedge is exposure to the plumbing, cash-flowing assets tied to the bottlenecks the winners must rent from you to grow. Not every investor can buy a substation, but anyone can move closer to businesses that make money when promises are kept and power arrives.
There is a policy shadow worth naming. When returns concentrate in corridors gated by permits and queues, politics becomes capital allocation by other means. Local boards, public utilities, water districts, zoning commissions, these suddenly reprice billions. The answer is not to sabotage the miracle; it is to accept that state capacity is now a factor in the model and behave accordingly: transparent procurement, receipts in public, and standards that let smaller players inherit rigor instead of begging incumbents for a seat.
If you are building, read the weather like a mariner. Secure the non-scalable inputs first and in public. Publish the PPAs, show the queue positions, name the crews. Bind every promise to evidence so procurement can say yes in an age of fear. If you are investing, stop asking “Is it AI?” and ask “What scarce input does this firm control that AI must rent?” If the answer is “content” or “eyeballs,” you are surfing froth. If the answer is “megawatts, land, water, verified outcomes,” you are closer to warm water.
Hurricanes pass, but they redraw coasts. This one is carving deltas where capital will settle for a decade, along power, land, and consequence. In the next section I will walk those shores: land and energy as the real AI stack, the geographies of compute that will look, a century from now, like rail hubs, wealthy, controversial, decisive.
Land & Energy as the Real AI Stack
A friend walked me through a pasture at the edge of a small town and pointed to stakes in the dirt that meant nothing to my eyes and everything to his. “Transformers here,” he said. “Pad-mounted switchgear there. Chilled-water loop along that fence.” Cows grazed; a meadowlark heckled us. In a year, the field will hum. I felt the old moral vertigo: I want cheaper models, and I want birdsong at dusk. AI is exothermic; its miracles vent into land use, power flows, water rights, and sound.
We keep talking about “the AI stack” as if it were a diagram on a screen. It is a map. Data centers do not live in cyberspace; they take up counties. The real stack runs megawatts → transformers → land → cooling → crews → models. Once you see it, you cannot unsee the rent frontier it creates. Ground that once priced hay now prices latency. Parcels that once hosted warehouses now host substations with waiting lists. The new hubs are not necessarily dense; they are electrically wealthy, near generation, near fiber, near water, and near a permitting office that can say yes.
Two non-trivial shifts follow. First, the premium moves from content to corridors. Owning a corridor, a right-of-way, an interconnection slot, a long-dated power purchase agreement, is not a boring footnote; it is the moat. You can rebuild a model with a bigger checkbook; you cannot fast-follow a transmission upgrade. The winners in this cycle look less like “apps” and more like the people who sited rail in the nineteenth century, unromantic, decisive, and sued by everybody. Alpha turns into surveying, procurement, and relationships with water boards. Your favorite feature list will wither in front of a planning commission. A signed easement is a more important slide than your parameter count.
Second, the social contract gets renegotiated at the parcel. When compute arrives, it drags rents with it. Rural towns become compute-adjacent and wake up to Bay Area prices without Bay Area wages. Homeowners near new substations inherit noise and light they did not vote for. The old playbook, tax abatements now, hand-waving later, will not survive an era in which every citizen is a publisher and every opposition group can lawyer up. The alternative is receipts-first YIMBY: publish real-time telemetry, decibels, water draw, hourly kWh mix, bind benefits to neighbors through property-tax freezes, bill credits, and broadband, and share upside as equity rather than swag. If a data center raises your cost of quiet, it should lower your cost of power. If it strains your aquifer, it should pay rent to your water table with metered honesty.
Cooling is a case study in how receipts change instincts. To a spreadsheet, water-cooled is elegant: lower energy per inference, fewer fans, better PUE. To a town in a drought cycle, water-cooled is theft. The way through is consequence accounting: liters per verified outcome published like a nutrition label, paired with a binding offset plan that invests in local storage or restoration rather than distant virtue. The right to compute does not eclipse the right to drink. A company that treats this as PR will lose. A company that builds a water budget with the county and returns waste heat to a school or a greenhouse will keep permits and sleep.
Energy is the same argument with more voltage. A grid under heat stress does not care about your valuation. The humane version of “scale” is shape: run inference where generation is stranded, batch jobs to off-peak, and make your flexibility valuable to the system rather than parasitic. I have started to look for locational marginal compute, pricing that encourages models to chase the same physics that govern aluminum smelters and bitcoin miners: abundant, cheap, often windy. If your AI roadmap ignores the weather, you are not building a product; you are building an outage.
There is also a justice claim that will not go away: who captures the rent from the corridor you need to grow? In the last boom, externalities landed on neighborhoods with the least say and the least cushion. In this one we can do better by making neighbors counterparties. Offer community PPAs that let residents own a slice of the electrons the data center buys. Create compute dividends indexed to uptime so locals cheer when you perform. Write community-benefit agreements that are enforceable by design, penalties that auto-trigger into local funds when noise or draw exceed thresholds. I am allergic to slogans, but here is one I can live with: if you make us a hub, make us a stakeholder.
This part of the map forced me to grow up. My most embarrassing portfolio mistake last year was underwriting a clever firm with no plan for power beyond “we’ll buy it.” My most satisfying check went to a boring developer with signatures from a co-op utility and a county board I can name. I still want the beautiful demo; I now want the substation queue position stapled to it. I still admire the paper with the training trick; I now read the footnote about cooling.
The older metaphor that helps me is the mill. Towns grew around waterwheels not because poetry, but because power. Mills were loud, contentious, and life-giving. They minted local fortunes and local resentments. The best built housing and halls; the worst extracted and left. Data centers are our mills. They will shape geographies, accents, school levies, and traffic patterns. They will concentrate returns and attention. They will test our capacity to trade wonder for consequence without becoming cynical or naïve.
So I try to hold both: the field that will soon hum and the bird that is heckling me while I consent to it. The compromise I can live with is receipts. Let the data center publish telemetry a citizen can read; let the citizen hold a share of the upside; let the county keep a veto that means something when promises are broken. Cheap models are not evil. They are heat. The question is whether we can place the radiators in ways that warm a town rather than warp it.
From land and energy, the argument turns human again. If the real stack runs through megawatts and parcels, redistribution that helps must stop being generic cash and start being capacity, time, training, and local ownership that make people resilient where prices are rising. That is next.
Redistribution That Works
Handing out cash in a bottleneck economy is like pouring water into a kinked hose: pressure rises where flow is already constrained. Checks chase daycare slots that do not exist, electricians you cannot book, megawatts not yet connected. Prices move; capacity does not. If AI makes most outputs cheap and a few inputs decisive, redistribution that helps must buy capacity at the choke points, time, training, custody, and local ownership, rather than spray demand at whatever is scarce this week.
Start with the most insulted form of work: care. The margin in a kindergarten classroom is not content; it is adults who can keep thirty nervous systems inside a day. A cash transfer bids up their time without minting more of them. A care floor, a wage indexed to local scarcity and funded like infrastructure, does the opposite. It pulls talent into the pool and stabilizes schedules parents plan lives around. Tie raises to verified outcomes, attendance, retention, child-to-staff ratios, rather than paperwork, and you buy not just compassion but the choreography that makes it reliable. I would rather pay taxes for rooms that open on time than for headlines about “support.”
Now energy. If compute is the new mill, you cannot calm a town with swag. Share the corridor. Let residents opt into community PPAs that sell a slice of the same electrons the data center buys, at a discount that appears on a bill rather than in a press release. Write capacity dividends indexed to uptime: if the facility runs hot, payments clear automatically into local accounts; if noise or water draw exceed thresholds, penalties post to a public ledger without a press conference. That is redistribution with teeth. It turns neighbors into counterparties and aligns their incentives with your reliability.
Training is where we get sentimental and then lazy. A bootcamp that promises software jobs into a market models just flooded is a feel-good cul-de-sac. A capacity pipeline looks different: paid apprenticeships in the actual bottlenecks, electrical, HVAC, line work, eldercare, early childhood, fast credentialing across states, and outcome-guaranteed placements underwritten by buyers desperate for crews. Stop treating trades as consolation prizes and treat them as the places AI’s miracle needs hands. If you insist on a stipend, staple it to hours in a shop or a ward, not as punishment, but as the shortest path to dignity that survives a rate hike.
Finance, which sounds abstract until your rent resets, may be the most humane lever. In a high-rate world, the best transfer is predictability. Replace scattered subsidies with fixed-rate windows for the scarcities that stabilize a life: mortgages for teachers and nurses within commuting distance of their work; capex loans for small contractors who pass outcome audits; refinancing guarantees for childcare centers that hit staffing and safety targets. These are not gifts to incumbents. They are ways to make the people who carry our bottlenecks less exposed to the weather we created.
Attention is a public utility nobody bills. The cheapest redistribution we can afford is to buy back the hinge of the day. Make device-free openers a norm in public schools and public offices. Fund quiet rooms in libraries the way we fund Wi-Fi. Require platforms to ship bounded-by-default sessions for minors the way we require seatbelts. This is not puritanism. It is capacity transfer to the one faculty everything else consumes. I have learned that the fastest way to give my community more of me is not to send money; it is to claw back two unfractured hours and spend them on someone who cannot invoice for theirs.
The tax base should tilt to consequence. We cannot fund a society on infinite streams of zero-price outputs. If the scarce goods are energy, land, verification, and care, then taxes should fall lightly on local capacity and bite where externalities live: on carbon, on unproductive land speculation around new corridors, on latency arbitrage that extracts without building. Pair those taxes with receipts-first procurement, public contracts that pay for outcomes with auditable evidence, and money will move from gloss to the people who hold reality together.
I am reweighting my own life accordingly. Less spend on the infinite, apps, streams, clever subscriptions. More on the finite I will miss if it vanishes: the daycare where my neighbor works, the electrician who actually shows, a small stake in a solar co-op down the highway, a clinic that publishes outcomes instead of decor. I try to tip for custody, the person who stands there when consequences land, not for charisma. When I pay more, I ask for receipts without apology. When I pay less, I ask who is eating the variance.
None of this is photogenic like a helicopter check. It is joinery, money and policy aimed at the joints where a society tears under stress. Cash has its place; when a life is on fire, you hand someone a hose. But if the fire is the structure, you rebuild the beams. In an economy where models make plenty and bottlenecks price the rest, redistribution that works will look like capacity compounding: hands we can hire, rooms that open, grids that hold, neighbors who own a slice of the hum.
Wonder with a Governor
On the edge of town the pasture has begun to hum. The stakes my friend pointed to last year are now steel, the chilled-water loop a silver artery along the fence. A low chorus of fans flattens the afternoon. Meadowlarks still heckle us, louder than necessary, as if volume were a vote. I stand at the property line and do the math I came to learn: megawatts in, tokens out, jobs by shift, decibels in the nearest kitchen, tax receipts, water drawn, heat returned to the greenhouse that reopened in March. A teenager pedals past with a basketball under his arm and calls to his friend, “They put the spaceship in the cow field.” He is not wrong.
Later, at home, the script on my laptop drafts a letter to the county about publishing telemetry, hourly kWh mix, water draw, noise at the property line, and another to the co-op about a community PPA I want my block to join. I schedule my small life around consequences: a childcare swap on Thursday, a contractor who will finally show on Friday, a standing walk without audio so the day keeps one unpriced margin. The market can keep its weather. I am learning to keep my climate.
I used to end essays like this with confidence about the arc of history, a sturdy line about how technology saves us in the end. I still think it can. I no longer think it does so by itself. The miracle at the edge is real: cheap code, cheap drafts, cheap competence on tap. The bill in the core is also real: land that must be sited, energy that must be shaped, hands that must be paid, promises that must be kept. If we will not price the bill, the miracle curdles into a trick. If we do price it, honestly, locally, with receipts, the miracle reads as a public good rather than a private spectacle.
I measure my days differently now. Fewer hours “being current,” more hours producing verified outcomes: a thing delivered with liability attached, a room that opened because someone trusted me to be there, a line I would sign with my name in daylight. I am less impressed by gloss and more moved by custody. I still use models with gratitude; I use them the way a carpenter uses a plane, and I sweep up my shavings. A good day is not the one with the most output. It is the one with the cleanest receipts.
The economics that frightened me at the start, the cost disease with a jet engine, the interest-rate whiplash, the index that lies, have not softened. They have become furniture. I do not expect calm seas; I expect hurricane mode with corridors of warm water where power, land, and consequence meet. My hedge is narrow and old: pay for the finite I will miss if it vanishes, and ally with the people who hold the world together when the weather turns. That looks like tipping for care, reading contracts, preferring dull cash flows, buying a small share of the electrons that keep the lights up on my block, and answering texts from neighbors who need a hand more than an opinion.
I do not know how the curve will look from space. I do know what feels like wisdom at street level: celebrate abundance, fight for guarantees, keep a human calendar, and make the last scarcities, land, energy, verification, attention, sit in the same room as wonder. When a data center raises the cost of quiet, insist that it lower the cost of power, automatically and measurably. When a platform takes your hinge, take it back in the morning and spend it on someone who cannot invoice you. When a policy offers speed without receipts, tell it no; when a founder offers receipts without theater, tell them yes.
Near dusk I walk back to the fence. The hum is steadier now, a new weather we will learn to predict. A maintenance crew waves as they swap a panel; the greenhouse vents a breath of warm air onto tomato plants the size of rumor. On my phone, the model has finished drafting a paragraph I will not use. It is competent and empty in the way free things often are. I close the screen and listen to two sounds that did not exist together here until this season: birds and the lightest version of thunder. The boy with the basketball misses his first shot in the driveway, then makes three in a row. Practice, then proof.
I can live with that pairing. The world will keep getting cheaper where it streams and dearer where it stands. The right response is not to mourn the flood or worship it, but to build levees where they matter, share the uplands, and keep a ledger we are not ashamed to show our children. Wonder with a governor. Abundance with receipts. A yes strong enough to survive its price.
If you want more of this, less gloss, more consequence, subscribe. I write a couple times a month about the last scarcities and the choices that hold a day together. No spam, no performative urgency, just receipts and workable moves.