A bank chief executive said this week that the AI investment cycle is expanding capital. Read that phrasing slowly, because the verb is doing all the work. Not returning capital. Not harvesting it. Expanding, which means pouring more in, building more, committing further out on a curve nobody has seen the end of. Coming from David Solomon at Goldman Sachs, the head of one of the most institutionally cautious shops on Wall Street, a line about AI investment expanding capital is not throat-clearing. It is a description of which phase we are standing in, delivered by someone with an unusually good view of the balance sheets.
And here is the thing I want to argue, because it runs against the entire reflex of the moment: knowing the phase is now the easy part, and it is worth almost nothing. The market wants a verdict. Bubble or boom. Overbuilt or underbuilt. Every headline sorts itself into one of those two bins, because a single answer is easy to trade, easy to defend on a panel, and easy to argue about at dinner. The trouble is that a single answer is almost always the wrong shape for the question being asked. A capital cycle of this size does not resolve into one word. It has layers moving at different speeds, and most of those layers are contradicting each other at any given moment.
The Fox Reads More than One Gauge
There is a well-worn distinction between two ways of forecasting, and it has aged better than most frameworks. One kind of thinker reaches for a single large idea and bends every incoming fact until it fits. The other reaches for many ideas, pulls information from many unrelated places, and then does the unglamorous work of adding it all up into one provisional conclusion. The second kind is reliably better at seeing what is coming, and not because any individual model in their head is smarter than the first thinker’s one big model. The advantage comes from aggregation itself. Errors in independent estimates tend to cancel, and the person holding six half-right views ends up closer to the truth than the person holding one confident view all the way to the end.
Applied to the AI capital cycle, that means the honest position is not yes and it is not no. It is several things held at once without forcing them into agreement. The spending is real, and some meaningful portion of it will be wasted. The demand for compute is genuine, and the revenue required to justify today’s buildout is not yet fully there. The infrastructure being poured into the ground will outlast most of the companies pouring it, the way railroad track outlived the majority of the railroads that laid it and went on generating value for owners who never took the original risk. All of those can be true in the same week, and in fact they usually are. The person who insists on collapsing them into one position is not being decisive. They are being lazy in a way that feels, from the inside, exactly like conviction.
What a bank chief saying “expanding capital” actually tells you is narrow and genuinely useful. The people closest to the money are still deploying rather than defending. That is one real data point and it should update your view. It is not the whole instrument cluster. It sits alongside the quieter readings that never make a headline: whether the returns are showing up on the schedule that was promised, whether the second-order businesses built on top of the models can charge enough for what they do to cover what it costs them to do it. A fox notes the line, weighs it against the rest, and keeps reading. A hedgehog turns it into a thesis and stops reading, which is the expensive part.
The Harder Skill Is Not Analysis
Here is the part almost nobody wants to talk about, because it is considerably less flattering than being right. In a cycle that is visibly expanding capital, the scarce skill is not figuring out what is happening. Plenty of people can read the gauges, and the readings are widely shared and mostly free. The scarce skill is doing nothing when nothing is the correct move, and doing something deliberately small when every instinct in the body is demanding something large.
There is a body of thinking about self-mastery that separates two capacities which look identical from the outside and behave nothing alike under pressure. The first is the ability to point your actions at your actual goals. The second is the ability to keep your emotional weather from taking the wheel. A boom stresses both simultaneously, and it stresses the second one far harder. Feelings were not designed for capital cycles. They were built for much older and more immediate problems, and they fire loudest at precisely the moments when they carry the least information: the fear of being left behind near the top, the urge to capitulate near the bottom, the certainty that this time the thing will not come back. The emotion arrives feeling like a signal. It is almost never a signal.
So the market’s real test during an expansion phase is rarely intellectual. It is behavioral, and it is administered daily. Can you hold a position while the number moves against you for reasons that have nothing whatsoever to do with your thesis. Can you decline to add on the green days, when adding feels like the only sane response available to a reasonable person. Can you sit with the specific discomfort of being partly right and somewhat early, which is the normal and permanent condition of anyone who bought something real before the crowd agreed that it was real. Analysis is what gets you into the position. Self-regulation is the only thing that lets you keep it long enough to be paid for it.
What the Phrase Actually Asks of You
“Expanding capital” is a statement about the world. It is also, quietly, a statement about you. It says the easy part, deciding that the AI cycle is significant, is finished. Everyone knows that now; it is in every deck and every earnings call. Solomon is not telling the room anything the room has not already priced. He is confirming the phase, and confirming the phase moves the entire game from having an opinion to holding a discipline.
The opinions are cheap and largely identical at this point. Whether you can act on one, hold it through noise that has nothing to do with it, and resist the constant pull to make the position bigger than your actual conviction warrants is what separates outcomes from here forward. That is not a research problem and no amount of additional reading solves it. The capital is expanding whether or not you are prepared for what that does to your judgment. The only variable genuinely under your control is the one between your ears, and it happens to be the exact variable a boom is designed to break.
Sit still better than the next person and you do not need to be smarter than them. In a cycle like this one, the calm hand beats the quick one, and it is not close.
Leave a Reply