The Price of Calm

Low volatility is not the absence of risk – it is risk moved somewhere it no longer has to be counted. This episode argues that the AI build-out is the largest version of that trade ever built.

It starts with the 2018 collapse of XIV, the note that sold calm and returned 585% until it lost 96% in a single day, and traces carry – the family of trades that earn a steady income for assuming stability holds, from the trader selling a put to the company buying back its stock to the central banks that have spent thirty years promising disorder will not be allowed to persist. Then it asks where that logic shows most clearly today, and lands on the financing edge of the AI build-out: the labs' compute commitments, the off-balance-sheet "shadow borrowing," and the circular flows between Nvidia, OpenAI, and CoreWeave.

The other side gets its hearing – this may be infrastructure, not carry, the way railways and fiber once were. So the piece ends where most commentary won't: a written-down level at which the author is wrong.

With thanks to Ahmed Husain, whose newsletter and podcast The Curious Mind first put The Rise of Carry in my hands.

A personal editorial view, read aloud – not investment advice.

Transcript

SPEAKER_00 0:00

The Price of Calm on Volatility Suppression, the AI build-out, and the trouble with being right once. Prefer to listen? There's an audio version of this essay above. About 12 minutes. In early February 2018, an exchange traded note called XIV was worth about $1.9 billion. Within a single trading session, it lost 96% of its value and was wound down. The name was VIX spelled backward in case anyone missed the joke. For most of a decade, it had done one simple thing, which was to sell calm, to pay you for assuming tomorrow would look like today. For two years it returned 585%, and the danger printed on its label kept being wrong in exactly the profitable direction. The people who owned it were not all fools. Many knew precisely what it did. That is the part I keep returning to. Each is paid to assume disruption will not arrive soon. Each is broadly short volatility. The mechanism is seductive because it works. Low volatility is not the absence of risk. It is a risk that has been moved somewhere, so it no longer has to be counted. Capital crowds in, prices rise, measured volatility falls, the models permit more leverage, and prices rise again. The calm is not merely the backdrop to the trade, it is partly produced by it. From the inside, none of this feels reckless. It feels like discipline, the returns are real, and after a while the trade stops looking like a trade at all. It becomes the way sensible people behave. Harry is also a habit the whole system has learned. Central banks do not blow bubbles in the cartoon sense, they respond, for understandable reasons, to serious stress by cutting rates and putting the floor under prices. Do it once and it is a rescue. Do it for 30 years and it becomes a promise nobody had to write down. Investors learn that if disorder grows large enough, the authorities arrive. Gains in the calm stay private. Losses in the crisis are socialized. When carry unwound in March 2020, help came within days. When the Yen Kerry trade began to snap in the summer of 2024 and the VIX briefly traded above 60, a single reassuring signal from Tokyo turned the market within days. The lesson lands every time. Dorder will not be permitted to persist, and each time it lands, the next position is built a little larger because the thing that made it truly dangerous, a fall allowed to keep falling, has been quietly removed from the model. So where does this logic show most clearly today? I think it is the AI build out, but I want to be careful with the claim. AI is not itself a carry trait. I use it every day. The demand is real, the revenue is real. By late May, Anthropic was reportedly running at roughly a forty seven billion dollar annual run rate against about 9 billion at the end of 2025. OpenAI is reported to be mere 24 billion. The growth is extraordinary and people are paying gladly. That is the engine of the trade, and the engine is genuinely impressive. The part that worries me is the financing edge, where longated physical infrastructure is funded on the assumption that demand, capital, power, chips, and model economics all arrive on time. To produce those tokens, the industry is committing to spend on a scale with few peacetime peers. McKinse puts the global data center build near $7 trillion by 2030, with the largest American hyperscalers spending on the order of $700 billion this year alone. The obvious retort is, so what? Microsoft, Google, Amazon, and Meta carry among the strongest balance sheets in corporate history. True, and important, which is why the fragility is not at the center. It lives at the edge. The first edge is the labs. OpenAI's compute commitments have been reported near one trillion of deals this year. Its own chief executive has framed them at about 1.4 trillion over eight years, a figure since guided down toward roughly sixty billion dollars of spend by 2030. That may be the right bet, but it is a bet that the future arrives on schedule, and the future does not have to fail for the financing to hurt. It only has to arrive late. The second edge is the structure. The early build came largely from retained earnings. That phase appears to be ending, and more spending now runs through debt, leases, special purpose vehicles, and joint ventures, what the Bank for International Settlements has described as shadow borrowing, raised against chips and buildings while kept off the parents' main balance sheet. The lenders on the other side are often private credit funds and insurers, precisely the institutions that like steady coupons in a low spread world, and that are asked to hold the risk that looks boring until it suddenly is not. The third edge is circularity and time. Nvidia has agreed to invest up to 100 billion in OpenAI, which will spend vast sums on NVIDIA chips. Nvidia backs Core Weave, which buys NVIDIA chips and rents capacity back to OpenAI. Money leaves one balance sheet and returns as another company's revenue, in a real boom, a virtuous circle, in a carry regime, a system manufacturing proof for itself. And the GPUs at the center may be depreciated over five or six years while the debt behind the buildings assumes a useful life of twenty or thirty. Each of these assumptions may be reasonable. Together they begin to look like a sale of volatility. It rhymes with 2007, not as prophecy, but because the structure is familiar. Clever vehicles, large obligations, off-balance cheek comfort, and serious people persuading themselves that risk has been engineered into something safe. The strongest objection is that I am using the wrong lens. On that reading, the build out is not carry at all, but the early infrastructure of a general purpose technology, and the infrastructure always looks wasteful while it is being built. The railways bankrupted many of their first investors and then carried a century of growth. Electrification and late 1990s fiber demanded capital far ahead of visible revenue. That objection is serious, and it may be right, but AI can be transformative and still destroy capital for whoever finances it first. The internet was real in 1999, the fiber was useful, and many of the companies were not. The question is not whether AI matters. It is who owns the stranded capital if it matters, but the margins, timing, or bargaining power land somewhere other than expected. I also have to be honest about myself. By temperament, I am a buyer of protection, drawn to the side of the trade that bleeds quietly and waits for the day the dam breaks. That instinct has been right before, and being right is the most dangerous thing that can happen to it. A man vindicated once can spend a decade paying a premium for the encore. So I have to ask whether carry is the right analysis, or merely the lens a congenital bear reached for this week. I would mock that in someone else. I should not excuse it in myself. Here is the test I have settled on. If this is infrastructure, the column should survive a shock. If it is carry, the calm depends on the shock not lasting. So I am not watching the level of the NASDAQ, AI revenues or capital spending in isolation. I am watching whether the column still repairs itself. That means I owe the reader a number, because without one, this is a mood dressed as analysis. The early warning is probably not a crash, but a deal that does not clear. The first piece of AI paper, an asset backed tranch, an SPV refinancing, a hyperscalar bond, a data center loan, that is pulled, repriced hard, or left to roll and cannot, or the first volatility spike that takes more than a week to be bought back, where the system that healed 2018 and 2024 in days suddenly needs longer. I would call the break confirmed if the VIX holds above 30 for 10 trading sessions or more, while investment grade credit spreads widen alongside it. Volatility alone heals. Volatility and credit together, refusing to mend, are a different thing. And here is the part the sellers of Calm rarely publish, the level at which I am wrong. If the big AI financing's clear, if open AI and anthropic reach public markets without disturbing credit, if spreads sit near their lows through the end of the year and every volatility spike is bought back within a week, then the regime is intact. This was infrastructure rather than carry, and I should fold the position and say so plainly. The sellers of XIV believe the calm was simply the world as it was. The people writing the checks for data centers may believe something similar, with better arguments and far more at stake. They may also be right. I am not standing above them. I have made their mistake in the opposite direction more than once. The only useful discipline I know is to write down in advance the price at which I would admit it. Poem is not free. We have just become very good at forgetting where the bill is sitting. GS, with thanks to my big friend Ahmed Hussein, whose newsletter, The Curious Mind, is one of my favorite weekly reads. He always recommended great books to me, and The Rise of Kerry was no exception.