Crosscurrents | World Without Us, Winners/​Losers of K‑Shaped Economy, Incumbents of 21st Century, Attention Economy

Sanjeev Krishnan | September 16, 2026

A World Without Us – the Sequel 

In 2007, Alan Weisman asked a simple question in The World Without Us: what would happen to the planet if every human vanished tomorrow? His answer was a quiet one. Within days, the subways flood because no one runs the pumps. Within years, the power plants go dark, the bridges rust, and the fields go to forest. The machine stopped because it needed us. Twenty years on, the sequel is already being written, and its premise is stranger than the original. We have not vanished. We have stepped out of the loops, but the machine has not stopped. 

Let’s start with the internet, the most measured system we have. In 2024, for the first time in a decade, automated traffic exceeded human traffic – 51% of all web activity was generated by bots, and that share is still rising. Most of what happens online is now machines talking to one another about things humans posted. The stakes are higher in markets. US equity volume hit a record 17.6 billion shares a day in 2025, and for the first time, more than half of it traded off-exchange, in venues no human ever watches, executed overwhelmingly by algorithms responding to one another. 

War, the oldest human activity, has crossed the same threshold fastest. Drones caused less than 10% of battlefield losses in Ukraine in 2022; by 2025, they accounted for an estimated 70 to 80%, and Ukraine reports that four out of every five strikes on Russian forces are now delivered by unmanned systems – roughly 820,000 targets in a single year. The soldier has not left the battlefield, but he has left the point of

contact. He is a pilot in a basement, increasingly supervising software that is learning to find its own targets – give an agent a goal and an obstacle, and it routes around, including around us. 

This is where the sequel departs from the original. In Weisman’s book, the systems failed because they depended on us: someone had to run the pump. The question Weisman asked – what would nature do without us – has become a different one: what will our systems do without us in the loop? In a world without us in the loop, the human who remains becomes the scarce feedstock, the operator who understands the system well enough to know when it has gone wrong. No interest rate has produced one, and no algorithm has yet replaced that operator. 

Societies that treat people as constraints rather than the cost will govern better; companies that keep a human where judgment matters rather than where volume is highest will fail less; and portfolios that underwrite what a machine-run world still needs from people, and what people still need from the physical world, will compound while the swarm sets the prices. Weisman’s original ended peacefully, with the forest reclaiming the city. The sequel will not be quiet. For the first time, the world without us has a say in how the story ends. 

The K‑Shaped Economy – Who Keeps and Loses Market Share 

Last month, Walmart reported its slowest US comparable sales growth in more than six years – 2.6% – and still gained market share. The two facts sit together only in a K‑shaped economy. Walmart’s lower-income core customer is pulling back; for two years, its share gains have come mostly from households earning more than $100,000 who have decided that the world’s largest discounter is where they shop. The letter K is usually drawn as an income chart, with one arm rising and the other falling. The most useful way to draw it is as a market share chart, because the K is sorting companies alongside households. 

The upper arm is powered by assets, not wages. The richest 10% of American households own roughly 87% of household stock market wealth, and at the end of 2025, a record 33% of total household wealth was held in equities – above the 2021 mania and the 2000 peak. An AI-led rally added $10 trillion to household portfolios in a single year. Those gains convert to consumption: Moody’s estimates the top 10% of earners account for roughly 46 to 49% of consumer spending, up from about 36% three decades ago, and their spending grew 62% between 2020 and 2025, far faster than any other group. The lower arm gets the feedstock inflation of our current reality – rent, utilities, insurance, groceries – arriving in the categories with the least room to absorb it, without the wealth effect. 

The same split is showing up in market share changes. Delta’s premium cabins now earn more than its main cabin – $6.9 billion against $6.85 billion in the June quarter – for the first time in the history of a US airline, and nearly all its planned seat growth is premium. Main cabin revenue fell 7% in December, while premium revenue rose 9%. At the other end of the runway, Spirit is in its second bankruptcy, and Allegiant is absorbing Sun Country. On the ground, Walmart is taking share from Target, which has posted same-store sales declines in recent quarters, and from dollar stores, whose core customers, according to

surveys, say they expect to sacrifice necessities this year. Even Dollar General’s best news is that it is now attracting higher-income shoppers – the same customers everyone else is chasing. 

The grocery aisle is where the K is most instructive, because it is the one place where both arms of the letter still meet. Walmart’s growth with affluent households has been driven by an upgrade in the goods themselves – a premium private label and healthier, more considered products – and surveys now find 

higher-income shoppers more willing to buy groceries at Walmart than at Whole Foods. Food is the category where a premium is measured in dollars rather than thousands, and where quality is a value judgment about the family dinner table rather than a status purchase. That is why premium everyday brands – in protein, in dairy, in the center of the plate – have been quietly gaining share through the K while higher-end discretionary categories swing with the stock market. 

Market share is flowing to companies that serve the upper arm directly, to companies with the scale to win the upper arm’s trade-down while serving the lower arm at cost, and to the brands that sit on the shelf between them: premium products in everyday categories that the upper arm buys by habit and the lower arm buys by choice. Losing share is the undifferentiated middle – the cheap-chic retailer, the mid-priced brand, the budget airline, the commodity product with no reason to be chosen – businesses built for a mass-affluent customer that is being pulled apart. In systems terms, the middle is a monoculture: one product, one price, one customer, and the customer has left. Delta’s front cabin fills and empties with the S&P; the better product in the grocery basket is bought in good years and defended in bad ones. Who keeps market share in a K‑shaped economy is a question at the top of the letter. If the letter is redrawn, the advantage goes to whoever earned the customer’s trust in the aisle where both arms still shop. 

The Incumbents of the 21st Century – the Pearl River Delta versus Midtown Manhattan and the Conversion of Future Cashflow 

Later this week, China’s Xi Jinping and President Trump will meet in Washington for a summit that will once again put the economic and geopolitical order of the 21st century on display. I assume the press coverage will focus on tariffs, Taiwan, or perhaps a more coordinated approach to global issues, from artificial intelligence to the Strait of Hormuz. Analysts will parse body language and declare who won or 

lost, while largely missing what the two superpowers of our era actually represent. 

One side controls the factory of the world, an agglomeration of physical production so dense that tariffs and de-risking have failed to loosen its grip on many manufacturing-related industries. The other sits atop the deepest and most inventive capital markets in human history – a machinery for converting future cash flow into present capital. Neither advantage can be de-risked, sanctioned, or conceded. 

China’s incumbency is physical: the Pearl River Delta and its extensions in the Yangtze and Shenzhen are built on more than cheap manufacturing. They combine a supplier ecosystem, talent, tooling, and process know-how that the world cannot easily escape – battery chemistry, solar, shipbuilding, and rare earth processing among them. America’s incumbency is financial: only American capital markets could invent the mortgage-backed security, the venture capital ecosystem, the high-yield bond, or even music

royalty finance. One country learned to manufacture anything; the other learned to finance anything, often with ​“fit for purpose” financial innovation. 

Both are network advantages that compound over time. And while the two superpowers may be in competition, their incumbencies are not – each is the missing half of the other. 

That mutual dependency is now being tested from both directions. The West has committed to rebuilding what the Delta already has – fabs, grids, ports, and the rest – a generational, balance-sheet-intensive endeavor to recapture the physical economy it abandoned. Financing that buildout at the scale is accelerating will require exactly America’s historic ​“fit for purpose” genius: for every new economic reality, America has engineered a bespoke financial instrument – bonds for railroads, venture equity for silicon, securitization for mortgages, royalties for molecules and music. China faces the mirror problem: a factory of the world that must learn to fund itself without the capital-market depth it never built. 

The future of inflation, supply chains and geopolitics rests on this race and unproven capacity — The negotiation of the 21st century, therefore, will not happen at any one summit or photo op. It is happening as incumbencies that took generations to build are being refinanced and reconfigured – the factory of the world learning how to fund itself, and American capital markets relearning how to build physical infrastructure not seen since the immediate post WW II era. 

After the Attention Economy – Turning Electrons from Dumb to Smart 

The attention economy is usually described as a destination. It makes more sense as a transition, bridging an economy that ran electricity through production and one that will run it through AI agents. Every customer the grid has ever served has had one thing in common: someone else decided when it could be turned down. The customer arriving now can decide for itself, and the power system being built to serve it does not yet know that. 

For a century, electricity was a measure of output. A kilowatt-hour meant a widget, a square foot, a ton of steel; demand rose with the factory whistle and fell when the lights went off; and after 2009, consumption went flat for fifteen years while GDP kept growing. The attention economy broke that link. 

US demand set a record in 2025 and will set two more, and the load driving it produces nothing in the old sense: seven hours a day on screens, two hundred phone checks, half a trillion dollars wagered since 2018, prediction markets up eleven-fold in a year, models that answer at 3 a.m. Next year, the commercial sector – the category that includes data centers – will use more electricity than all households in America combined, for the first time on record. Kilowatt-hours per unit of GDP fell for decades; kilowatt-hours per hour of leisure are rising. Electricity is decoupled from production and coupled to people. 

Coupling to people carries one non-negotiable property: the person is waiting. A feed that loads in four seconds is dead; a companion that answers in the morning is not a companion. So attention demanded power the way a hospital does – firm, instant, every hour – and the grid is being rebuilt to that specification. Three closed nuclear reactors are returning to service under 20- and 25-year contracts with Microsoft and Google, backed by $4.4 billion in federal loans. PJM’s capacity price rose from $28.92 to

$333.44 per megawatt-day in three auctions, with data centers responsible for 63% of the first jump. All of it rests on the assumption that the load cannot be moved. 

Exogenous curtailment 

Curtailment simply means a customer using less power than they otherwise would, for a while. To the extent the grid has ever managed demand, it has done so from the outside – which is what ​“exogenous” means here. Demand response, across its hundred-year history, works this way: the utility sees stress coming, sends a signal, and a smelter or a Bitcoin mine turns down in exchange for a payment, a few dozen hours a year, treating the interruption as a cost. Even the most advanced version – Google’s 1 GW of data center demand response across five utilities, Emerald AI cutting an Oracle facility’s draw by 25% at peak – is exogenous. The utility decides; the data center complies. 

Duke’s Nicholas Institute showed what that alone is worth: 76 GW of new load, about 10% of national peak, fits on the existing grid if it accepts curtailment 0.25% of the year; 98 GW at 0.5%; 126 GW at 1%. Those are large numbers, but this remains the old model with a better-behaved customer. 

Endogenous curtailment 

Agentic AI changes who decides, and that is what ​“endogenous” means: the decision to use less comes from inside the load, not from the utility. An agent that plans, calls tools, and runs while no one watches consumes five to thirty times the compute of a chat exchange, but for much of that work, nobody needs 

the answer this second – and the agent knows it. It can read a price. It can see that PJM is short at 6 p.m. and long at 3 a.m., and that ERCOT is curtailing solar at noon. It can act on that without a utility program, a signal, or a payment, because minimizing the cost of its own compute is already part of its objective. Curtailment is no longer imposed on the load; it is chosen by the load, continuously, as a byproduct of doing its job well. 

The nearest everyday analogy is the difference between a thermostat that the utility can override on a hot afternoon and a tenant who watches electricity prices and decides for herself when to run the dishwasher – except the tenant is software, and it makes that decision thousands of times an hour. 

This changes the nature of demand response. Exogenous curtailment is rare, binary, and reluctant. Endogenous curtailment is constant, granular, and willing: thousands of small decisions an hour, load sliding toward wherever electrons are cheapest, nobody compensating anyone because nobody was forced. Every customer in the history of the grid has been a fixed quantity that supply had to chase. An agentic workload optimizing its own power bill is the first load that chases supply back, less like demand, more like a battery that does useful work while it charges. 

Agency still needs a firm core: trading, defense, anything with a person or a market waiting behaves like attention and will run on the reactors and gas turbines now being built. But the growing remainder does not need to be served firmly or managed. Nuclear and gas for the heartbeat; the metabolism regulates itself. 

What nobody has modeled

Endogenous curtailment appears in no planning document. Load forecasts treat demand as given; capacity markets clear against a peak assumed to be immovable; minimum-billing tariffs fine a customer for consuming less than it contracted for, which is to say they fine endogenous curtailment. Firms holding 25 years of firm supply have every reason to run it flat rather than let their agents shop. And the workload mix remains unresolved: inference, which serves requests in real time, is forecast to overtake training by about 2030. If agentic AI turns out to be mostly agents serving people at conversational speed, it is attention without a bedtime, and the firm buildout was right. 

There is also a risk no one has priced: a million agents chasing the cheapest kilowatt-hour behave like a herd, and synchronized demand swings at software speed are an instability the old, clumsy model never produced. What changes with agentic AI is who decides when. Policy still pulls on the top of the stack, and portfolios are still sized to baseload, because both assume the next customer is like the last one. The defining property of the next customer is that, for the first time, it is smart enough to curtail itself – and no one has yet built a grid for a load that can think. 

Endogenous versus Exogenous curtailment is an experiment in modern intelligence. Electricity is just the first commodity to test this dynamic, but calories, molecules, and freight are next. For a century, intelligence sat mainly on the supply side of every commodity market. The century ahead belongs to whoever figures out what happens when it shifts to the demand side as well.

—

The views and opinions expressed are those of the S2G team members identified and do not necessarily reflect the views of S2G. This content is for informational purposes only, should not be taken as legal, business, tax or investment advice, or be used to evaluate any investment or security, and is not directed at any investor or potential investor in any investment vehicle sponsored by S2G. Investing involves risk, including the risk of loss. Specific companies mentioned in this podcast are for educational purposes and should not be construed as an endorsement of any kind. Please note that S2G may maintain investments in some of the companies discussed on this podcast. Any past performance discussed is not indicative of future results. For more important information, please see s2gin​vest​ments​.com/​d​i​s​c​l​o​sures.