Vol. IV, No. 82
Covering 26 January - 8 February 2026
Monday, February 9, 2026
Late edition · A Relentless publication
The Relentless Dispatch
All the fortnight that mattered, in technology and in the world, read next to what we were building at the time.
THE GULF

Shots in the Strait of Hormuz, and the last nuclear treaty quietly dies

As an American fleet closed on Iran, the two edged into open confrontation at sea: a downed drone, seized tankers, an aircraft carrier under threat. And on February 5, with almost no one watching, the final treaty limiting the world's two great nuclear arsenals simply expired.

The confrontation with Iran moved from prelude to skirmish this fortnight, in the water. On February 3, an American F-35 shot down an Iranian drone as it approached the carrier USS Abraham Lincoln, and Iranian gunboats attempted to seize an American tanker in the Strait of Hormuz, which continued under Navy escort; two days later Iran seized two foreign tankers it accused of smuggling fuel, and Washington sanctioned fourteen more. Each incident was, on its own, containable, the calibrated jostling of powers not yet at war. Together, with an American fleet closing and Tehran convulsed by protest and crackdown, they were the sound of a war being assembled, rung by rung, in the world's most important shipping lane.

And beneath the drumbeat, on February 5, a quieter and in some ways graver thing happened, almost unremarked: New START, the last remaining treaty limiting the strategic nuclear arsenals of the United States and Russia, expired, with no successor, no negotiation, nothing. For the first time since the depths of the Cold War, the two nations that between them hold most of the world's nuclear weapons are bound by no agreement at all on how many they may build or where they may put them. The end of arms control did not come with a summit or a signing or even a statement of much note; it came with a lapse, a date passing, a treaty simply allowed to die while the world's attention was on the Gulf, and the age of unconstrained nuclear competition, which the treaties of the last half-century had held at bay, resumed in silence.

MILAN

The Games open as the drums beat

The 2026 Winter Olympics opened on February 6 in Milan and Cortina d'Ampezzo, the familiar spectacle of the world's youth competing under one flag, staged this year against a backdrop of gathering war, an American fleet steaming toward Iran, a nuclear treaty freshly dead, an oil nation conquered weeks before. The Games proceeded, as they do, an insistent fortnight of sport and ceremony amid the darkening, and days earlier, at the Australian Open on February 1, the young Spaniard Carlos Alcaraz had completed a career Grand Slam by beating Novak Djokovic, the youngest man ever to hold all four major titles, a changing of the guard this paper has tracked to its completion.

IN BRIEF

The files, the layoffs, the mine, the music

The Justice Department released more than three million pages of the Epstein files at the end of January, a trove whose slow disgorging continued to shadow a president who had promised transparency and now found the transparency inconvenient. The Washington Post laid off roughly three hundred journalists, a third of its newsroom, another cut in a collapsing industry at the worst possible moment for the scrutiny of power. At least twenty-five people died in an illegal coal-mine explosion in Meghalaya, India. And at the Grammy Awards, Kendrick Lamar became the most-awarded hip-hop artist in history and Bad Bunny's became the first Spanish-language album to be named Album of the Year, small markers of a culture still capable of honouring the made thing amid the unmaking.

The Column

The race with no finish line

The largest technology companies will spend nearly seven hundred billion dollars on AI infrastructure this year, roughly double last year. The number missing from every announcement is the return. The structure that produces spending without a demonstrated payoff has a name in biology, and it is not reassuring.

By Daniel Reeve

The five largest American technology companies have committed to spending somewhere near six hundred and ninety billion dollars building artificial-intelligence infrastructure this year, nearly double last year's already staggering sum, and the number conspicuously absent from every triumphant announcement is the one that would justify it: the return, the demonstrated profit, the evidence that this spending will pay for itself rather than merely continue. The absence is not an oversight; it is the structure of the thing, and the structure, a headlong competitive spending that no participant can stop and none can show will pay, has a precise analogue in biology, and the analogue is not comforting.

Biologists call it the Red Queen, after the character in Lewis Carroll who must run as fast as she can simply to stay in the same place. The concept describes evolutionary arms races in which competitors are locked into ever-escalating investment, not because the escalation improves their absolute position but because any competitor who fails to escalate falls behind the others who do, so that everyone must run faster and faster merely to maintain their relative standing, expending enormous and ever-growing resources to stay in the same competitive place. The peacock's absurd and costly tail, the arms race between predator and prey, are Red Queen dynamics: escalations that consume vast resources and confer no lasting advantage, because every competitor escalates in lockstep, and yet no individual can stop, because to stop while others run is to lose. This is precisely the structure of the AI build-out. Each company spends not because it has demonstrated the spending will pay but because its rivals are spending, and to spend less than they is to fall behind, so that the collective total escalates toward seven hundred billion dollars driven not by any calculated return but by the pure competitive logic that no one can afford to be the one who ran slower.

The counterargument is the one the participants themselves would offer, and it is not empty: that the race is real and the stakes are real, that whoever leads in artificial intelligence may command a genuinely dominant position, and that in a winner-take-most contest the enormous spending is rational insurance against the catastrophe of losing, so that the Red Queen framing, with its implication of futility, understates the real prize at the finish. This has force, and it is why the spending is not simply irrational. But notice what the framing exposes: the assumption of a finish line, a moment when the race ends and the leader collects a decisive, durable prize, and it is precisely this finish line that the missing return number calls into question. A Red Queen race is defined by the absence of such a resolution, by the fact that the escalation is permanent, that there is no point at which anyone can stop running and enjoy the winnings, because the competitors keep coming and the running never ends. If the AI race has a true finish line, a point of durable dominance, then the spending may be justified; but no one has demonstrated that it does, and the pattern so far, each advance quickly matched, each lead eroded, each new model soon equalled, looks far more like a Red Queen race with no terminus than like a sprint to a decisive prize. The spending assumes a finish line that the evidence does not show, and a race with no finish line is not an investment but a treadmill.

And that is the question the missing number finally poses, beneath all the announcements: whether this is a race that someone wins, or a race that everyone is simply condemned to keep running, spending more each year to stay in the same relative place, until the spending exhausts the runners or the whole contest is revealed to have had no finish line at all. The Red Queen dynamic is stable precisely because no individual can defect; each company, rationally, must keep spending as long as the others do, and so the collective total climbs, seven hundred billion this year and more next, driven by a logic that guarantees the running without guaranteeing the prize. The participants call it a race to the future, and perhaps it is; but a race is defined by its finish line, and the number that would prove there is one, the return, the payoff, the durable advantage that ends the running, is the number no one will name, because to name it would be to admit that after nearly seven hundred billion dollars, doubled again, no one can yet point to where the running stops and the winning begins. This paper has asked for that number for three years. It is still missing, and the spending resting on its absence has just doubled once more, and the Queen runs on.

Field Notes
A Relentless build, told plainly

Building the tool that decides where the data centres go

The front page is an economy staking its future on a build-out of unprecedented scale. This is about a specific, unglamorous piece of that build-out I am working on now: the problem of deciding where, on a strained and finite grid, the enormous new machines can actually be built. No client is named.

By Alosh Paul

The AI build-out that this paper keeps returning to, the $690 billion of capital spending, the data centres the size of towns, is usually discussed as a story of chips and models and money. I am working, this year, on a much more physical piece of it, and it has taught me where the real difficulty of the whole enterprise actually lives, which is not in the silicon but in the ground. The problem I am building software to address is deceptively simple to state and brutally hard to solve: given that you want to build an enormous data centre, one that will draw as much power as a small city, where, exactly, can you actually put it? And the answer, it turns out, is a question not of real estate but of the electrical grid, and it is the hardest kind of problem there is, because it is a problem of constraints that all bind at once.

Here is what makes it hard, and why it is a genuinely interesting piece of engineering rather than a lookup. To site one of these machines you must satisfy, simultaneously, a dozen constraints that pull against each other: there must be enough power available on the grid at that point, which is rarer than anyone outside the industry imagines, because the grid was not built for loads like these; there must be the ability to actually deliver that power, the transmission lines and substations, which take years to build; there must be water for cooling, and land, and fibre, and a community that will accept the thing, and a regulatory path, and a price that works. Each of these is a constraint, and the sitable locations are the vanishingly small set of places where all of them are satisfied at once, and finding those places, in a country where the grid is already strained and the good sites are already taken, is a search through an enormous space of possibilities under constraints that interact in complicated ways. That is a real computational and analytical problem, and it is the actual bottleneck of the AI build-out, far more than the chips everyone talks about.

What I have come to understand, building this, is that the whole trillion-dollar AI story rests on a physical foundation that almost no one is looking at, and that the foundation is much narrower than the money assumes. The capital is nearly infinite; the chips, eventually, will be made; but the places where you can actually put a small city's worth of new electrical load, on a grid that is already near its limits, are few, and finding and sequencing them is the constraint that will actually determine how fast, and whether, the build-out happens. The tool I am building is, at bottom, an attempt to see that narrow foundation clearly, to map the small set of the possible against the vast ambition of the money, and the deepest lesson of the work is the one this paper has been pressing for three years: that beneath the digital abstraction of AI lies a stubbornly physical reality of wires and water and power, and that the physical reality, not the money or the models, is where the whole thing will be decided. The people pouring $690 billion into this are betting, mostly without knowing it, on the answer to a question I am spending my days trying to compute: where, on this finite and straining grid, the future can actually be built.

The Ledger
AI
The $690 billion capex commitment for 2026, double 2025, is the fortnight's defining economic fact, and the return that would justify it remains unnamed. The concentration deepens; the question sharpens; the bubble debate moves from the fringe to the front page.
Data centers & power
The Field Notes above make the point the money obscures: the binding constraint on the whole build-out is not capital but the physical grid, the narrow set of places a small city's worth of new load can actually go. The wire, again, is the frontier.
Rates
The death of New START and the Gulf skirmishes add war and nuclear risk to an already fraught picture; markets weigh an AI bet of unprecedented size against a world sliding toward conflict. The Fed watches, its tools irrelevant to either.
Real estate
US 30-year mortgage near 6.1 percent. The data-centre build-out is quietly reshaping land and power markets across whole regions, a real-estate story of a new and strange kind.
India tech
India, with its own strained grid and its own AI ambitions, faces the siting problem this issue's Field Notes describe in an acute form; the physical constraints on the build-out are, for a power-short country, even sharper, and the frugal, constraint-driven approach may prove the only viable one.
What we called wrong
Nothing to retract. The doubling of the AI capex confirms this paper's long concern about concentration, and the quiet death of New START is the fulfilment, in the gravest register, of the "arms control is lapsing" warnings this Ledger has carried for two years.
The Back Page

The Games go on

They lit the cauldron in the Italian Alps this fortnight, and for all the darkness gathering around them, an American fleet steaming toward a war, a nuclear treaty freshly dead, the Games went on, as they always improbably do. There is a case for cynicism about the Olympics, the cost, the corruption, the nationalism, and this paper has made it before. But there is also the thing the Games stubbornly are, in a fortnight like this one: a fortnight in which the world agrees, against all the evidence of its own behaviour, to settle some of its rivalries by skiing and skating rather than by killing, to send its young people to compete under rules everyone accepts, to let the losers shake hands. It is a fragile and partial and easily-mocked truce, and it does not stop the fleets. But watched from a world sliding toward war, there is something in the sight of the nations gathering to race down a mountain rather than to seize one that this paper, which spends its life chronicling the failure to do exactly that, cannot quite bring itself to sneer at. The fleet steams toward Iran. The treaty is dead. And in the Alps, for a fortnight, the young of the quarrelling world run and jump and fly, and it is not nothing, and it may be, this year, the best of what we are still capable of.