Know Thy AI Pessimists

Latest stories Even after midterms, Trump's still a problem for Seoul and Tokyo Chinese regulatory shadow hangs over Nigeria's OPay IPO failure America's AI backlash is Asia's design challenge

The Waverers' Watch. Some of the displacement is already visible, so the useful work lies in splitting it. Where governments try to restrict displacement, the waverer should check whether those policies have any genuine effect, whether the damage keeps growing anyway, and what else follows from it.

The Optimists' Watch. Optimists cannot afford to treat the social side of AI as background noise. The arguments will keep changing form. Because AI touches almost everything we do, policy events around it will only multiply in the coming years. An optimist who models the technology carefully and the politics lazily is holding half a thesis. The signal that should give an optimist pause is a backlash strong enough to slow adoption itself.

The Devotees' Path. For the fully convinced, the first question is whether the fear of displacement, and whatever policy it provokes, will actually slow AI down. If it will, what they do in policy recommendations or their own financial investments will differ from the countercase. Displacement that is real and beyond the reach of any regulator calls for adjustments in daily life: in skills, in careers, in what children study, and in how savings are held. The aim is to shield what one holds dear, because the law moves slower than the layoff.

Firewall fretters

The Camp. The Fretters worry about a single bad weekend. Cyberattacks are rising quietly, week after week. Those who follow them closely see the world drifting toward a breach large enough to cause real chaos. One such event could change overnight what societies decide to control and how. Unlike the Doom Collectors, the Fretters have a specific fear with a specific trigger. They just cannot name the date.

The Waverers' Watch. The waverer should notice who is best placed to stop the breach. Fighting AI increasingly takes AI. Defenders without the latest tools are already outgunned. At least once this year, limits meant to keep powerful models from attackers also kept them from defenders at the worst moment, while the attackers faced no limits at all. Rising attacks may lift demand for AI more than they dent it. The same tools are also shrinking the scope of older software and slower technology innovators.

The Optimists' Watch. Optimists should expect every major incident to bring more talk of regulation and more controls, including the risk of arbitrary shackles after a massive event.

The Devotees' Path. The fully convinced should follow that belief to its target. A major event would likely hurt ordinary life far more than the technology companies themselves. Banks, hospitals, utilities, and logistics run on old systems with thin defenses. The sharpest repricing could hit traditional sectors and whole economies through a sudden rise in their risk premium. The Fretter who sells AI stocks to prepare may be hedging the wrong half of the portfolio.

Spreadsheet scolds

The Camp. The Scolds concede that AI works. Their quarrel is with the arithmetic. Hundreds of billions go into chips, buildings, and power every year, while revenue from AI services remains a fraction. Hardware loses value fast. Cash piles shrink. The marginal builders borrow from private credit funds and through special vehicles. The camp's cousin draws circles on the revenue chart: money looping among the same few firms. We took that claim apart in an earlier essay. The Scolds will disagree with almost everything there.

The Waverers' Watch. The waverer should judge the spending the way an owner would, across an asset's working life: what customers pay, what running it costs, and what replacement will demand. A single year's AI revenue set against a single year's spending says little about either. Much of the payoff also hides inside older businesses, in sharper advertising, better search, and cheaper internal work. Lately, the chip and memory makers collecting the cash have been putting it back into owning and leasing hardware themselves. The circle drawers will file that under circularity. A landlord buying more buildings while the waiting list grows is the plainer reading.

The Optimists' Watch. Optimists should watch where the leverage sits. The largest operators will not pause their buildout over the free cash flow line unless the leverage ratios become too high for comfort. The weak links are elsewhere: smaller cloud operators, developers with land and power but little experience, and lenders who priced the risk for a world without delays. Spreads, covenants, and refinancing terms at these names will crack long before anything at the giants does. A fully booked machine can still earn too little for the price paid to own it.

The Devotees' Path. Suppose the return never comes. The Purists' consolation applies here with more force, since the spending still built grids, fabs, and skills. The sharper question is who bears the loss. Most of it would land on shareholders of cash-rich companies. Private credit funds would take the next share. Their investors signed up for exactly this risk. That is capitalism doing its job. The exposure of deposit-taking banks, by all accounts, appears small. That leaves little room for the kind of systemic contagion seen in earlier financial crises. The believer's real watch is whether that ever changes.

Ego auditors

The Camp. The Scolds blame the arithmetic. This camp has found a culprit. It sits in a boardroom. In its telling, data center spending is decided by a handful of hyperscalers and neocloud operators, funded by greater-fool-believing parts of financial markets and fueled by ego. Every profit at the chip and memory makers hangs on those few signatures. The camp rarely looks past them to the people actually using AI. The story rarely addresses how much of the spending rests on real consumption by companies and households. When financial stress gets called the next choke point, this confusion is usually where it starts.

The Waverers' Watch. The waverer should separate the layers of decision. Users come first. Whatever surveys or outsiders say about them, hundreds of millions of people and companies now use AI. That number keeps growing in everything we track. Watching that demand matters more than second-guessing the intermediaries. A hyperscaler can misread demand for a while. If the demand is real and supply is short, the intermediaries will have genuine reasons to continue doing what they have been doing.

The Optimists' Watch. Optimists should remember that a cycle needs no oversupply. Fear alone can freeze financial markets. A freeze can force a downturn, whatever the eventual demand. Strong demand may only keep it shallow. Spending cuts announced while usage is still climbing would be the sign that this kind of cycle has arrived.

The Devotees' Path. Those convinced that everything hinges on a few big spenders will want them restrained by regulators or by shareholders demanding discipline. They should consider who fills the space. Competitors are not standing still. A rival with a patient owner, or with a state as its owner, will happily take the share a disciplined company gives up. The technology's potential is understood everywhere now. Throttling one country's spenders hands the field to companies and countries that face no such brakes.

Grid NIMBYs

The Camp. From the boardroom, the argument moves to the neighborhood. The NIMBYs accept that AI works and that people want it. They simply want it built somewhere else. Theirs is NIMBYism with a power cord. For them, local electricity bills climb, impacting the cost of living, because of data centers. For some, there are other environmental concerns. The most committed NIMBYs oppose the buildout everywhere, from the county next door to the plans for data centers in orbit.

The Waverers' Watch. The waverer should track where the blocked capacity goes. Data centers are less rooted than factories. Governments in Central Asia, the Middle East, and North Asia are eager to take the load, often with cheap power, fast permits, and land already cleared. Every project refused at home is one someone else can approve within months. The real question is how much of the work can travel, and how quickly the receiving countries can switch it on. The local fight may be won while the global buildout carries on unbothered.

The Optimists' Watch. Optimists should note what the NIMBYs rarely mention. Demand has never been part of their argument. New hardware fills up almost as soon as it is switched on. The real risk lies in the timing. A data center canceled in one country takes years to reappear in another. In the gap, markets may read every cancellation as proof of a slowdown. Optimists should watch the extent of any displacements against the ongoing work to estimate the true demand and supply balance at other choke points.

The Devotees' Path. Suppose the NIMBYs win everywhere. Demand keeps climbing while supply stays stuck behind choke points that were already tight. The shortage would show up as a new kind of inflation: rationed model access, rising compute prices, and waiting lists for intelligence. Public opinion would then turn. Sustainable investing went from virtue to villain within a few years. A shortage of AI could flip the mood on data centers even faster.

Assumption artists

The Camp. Some pessimists need no single thesis. They live off the fog. How many data centers will be built, when, where, and which of them can move? How short is memory? How tight is packaging? How much efficiency will the next model or chip squeeze out? Who will finance any of it? Nobody knows, which suits this camp perfectly. Every gap becomes a number. Every number becomes a scare. We hear that most of the world's tokens are paid for with borrowed money, sometimes more than 90% of them. We hear that a credit freeze would shrink the buildout to a tenth of its plans. Neither figure comes with a source. The numbers change daily. The fear stays constant.

The Waverers' Watch. The waverer should doubt every data point. The more frightening a claim, the more checks they need. Most are invalid or unprovable in a world where nobody can arrive at what data center capacity is planned or should be expected in the US next year, with the bottom-up ranges varying by over 100%. The rest move too fast to hold.

The Optimists' Watch. Optimists need one tool: the choke point ladder. The buildout runs through many choke points at once. The tightest sets the pace for all the others. Without a doubt, they will also suffer from a lack of forward-looking data, but there could be a lot in the actual information on prices, plans, and buyer-seller contracts without the aggregates for what is coming next.

The Devotees' Path. Those certain that everything is imbalanced will be right about one or the other bad outcome. Knowing what kind of imbalance could still be important, as long as they are not a Doomer discussed in the first section. That could be the reason for them to verify the data they come across.

Good-enough gang

The Camp. The scaling-law pessimists have mostly gone quiet. Two years of models that refused to plateau saw to that. Their heirs have moved the argument from the lab to the desk. The models may keep getting smarter, they concede, but nobody needs them to. Most companies cannot absorb what last year's models already offer. Change is outrunning the people meant to use it. Demand for frontier intelligence will saturate. Model makers will then grind each other down on price.

The Waverers' Watch. The waverer should look at what actually happens to old models. We covered this in the section titled“Nobody Sells iPhone 3 Currently” in The World of Zero Visibility. Models are retired almost as fast as new ones arrive. Whether last year's model was good enough is beside the point. Newer models, open or closed, take most of the use and most of the compute. How intensely they are used sets total demand. A retired model, frozen on old information and old methods, may linger in some corner. It barely moves the overall compute bill.

The Optimists' Watch. Optimists should track whether model makers keep finding new products, services, and methods. If the newest models end up confined to exotic applications, saturation will reshape what model makers sell. Demand would feel it, though not at once. The opposite can happen just as fast. The agentic tools launched in the past two months show how a single new use can send demand for the latest sharply higher, the way it has been happening with smartphones. As of now, the replacement cycle remains short, but it may not forever.

The Devotees' Path. Those fully convinced that smarter models are unnecessary should ask who wins in their world. Cheap, open, often years-old models would gain the upper hand. That matters well beyond hardware. Much of the open-source frontier comes from makers some governments would rather keep out. A world settling for good enough would run increasingly on their models and dated technologies, with economic and geopolitical consequences the Gang rarely prices in.

Profit pallbearers

The Camp. The Pallbearers are half convinced. On today's numbers, many AI stocks look like some of the best value in the market. They simply refuse to trust the numbers. The E in the P/E is borrowed from a cycle that can turn overnight. If capacity runs ahead of demand, earnings collapse and take the bargain with them.

The Waverers' Watch. The waverer should start with the choke point ladder. Many choke points are tight enough to keep supply short even if a good share of planned data centers is canceled or delayed. This year's shortages came with a smaller buildout than next year's. Corporate behavior matters even more. Companies that have tasted pricing power rarely give it away. Software makers have held their prices for years with no capacity limit at all. Hardware makers are unlikely to slash theirs because demand dips a little.

The Optimists' Watch. Optimists should read the ladder from the other side. A tighter rung upstream can tip the next segment into oversupply almost overnight.

The Devotees' Path. Those certain the cycle will turn should ask what is still a commodity. We have written increasingly about markets once treated as interchangeable that no longer are. Memory made to one customer's design behaves less like a commodity every year. Commodity cycles punish commodities. The devoted pessimist should track where commoditization has stopped, because the old cycle may never return there.

Gravity faithful

The Camp. The Pallbearers at least look at earnings. The Faithful do not bother. For them, a sector that has grown far more important than the rest of the market deserves to shrink back, whatever its ratios say. They see risks gathering in bond markets, geopolitics, politics, and policy. When markets finally price those risks, the sector that rose the most is expected to fall the most. In this decade, that sector is AI.