They moved three trillion dollars of debt into the footnotes, kept the leverage off the balance sheet, and called it innovation. A Wall Street Journal analysis of recent securities filings finds nine major technology companies holding roughly $3 trillion in off-balance-sheet commitments tied to AI infrastructure: $1.9 trillion in purchase obligations and $1.2 trillion in uncommenced-lease promises — about four times the level disclosed a year earlier. Three trillion dollars, hidden in plain sight, in the books of the firms that ask the rest of us to trust them with the future.
The optimists will tell you the optimistic case honestly, and I want to give them their due. Demand for AI tools is real. Chips and power and land are constrained. Securing supply in a tight market is what a prudent operator does. The off-balance-sheet structures are standard project finance — the same instruments that built pipelines and rail lines for a century. Locking in power-purchase agreements across multi-year horizons is the same thing the railroads did when they locked in land grants, the same thing utilities did when they locked in fuel contracts across decades. If the demand forecast is right, the revenue will materialize, and the leverage will look like a clever bet well-timed. A man who thinks the future will be bigger than the present will build for it. That is not, in itself, a sin.
Here is the sin. Three trillion dollars of obligations have been engineered to stay out of the headline numbers. Morgan Stanley’s accounting analysts wrote in April that the trend is “making it increasingly difficult for investors to assess companies’ total potential leverage.” That is the polite version of what the firms have done: they have made it harder for the people whose money they hold to know what they own. The further the claim gets from the asset, the harder it is to see the exposure, and the easier it is to keep borrowing. That is the architecture of the thing.
I have seen this before. I used to trade paper claims on corn and beans — futures, options, the synthetic instruments that let a man in a tower bet on a crop he would never see. The same hands that built the derivatives markets learned to dress the position so the risk looked small. A futures position is a line on a blotter until the market moves and the margin call arrives; a purchase commitment is a line in the footnotes until the delivery date arrives and the bill comes due. From the leveraged-buyout desks of the eighties I learned that contingent obligations are never as contingent as they look on the day they are signed. The buyout guys bought companies with other people’s money and called the debt “non-recourse.” They were wrong every time the company went sideways. The S&Ls in the eighties parked their bad loans in off-balance-sheet vehicles and called them “asset-backed.” They were wrong every time interest rates moved. The banks in two thousand six sold mortgage-backed securities built on loans they knew would default and called them “AAA.” They were wrong until the music stopped. The footnotes are where you put the thing you do not want the reader to see. Three trillion dollars of footnotes is not a rounding error. It is a confession.
Look at the structure Meta built in Louisiana. They call the campus Hyperion. It is roughly the size of seventeen hundred football fields. Blue Owl Capital owns most of a joint venture that owns the campus. A holding company called Beignet Investor — the name is the kind of thing accountants laugh about over drinks — sold twenty-seven billion dollars of bonds to build it. Meta is the minority partner and the tenant. They agreed to a four-year initial lease with options to renew for up to twenty years. They guaranteed the bondholders they would make them whole if they did not stay the full two decades. They told investors they did not view the guarantee as probable, so they did not record any related liability. A twenty-year promise on a building the size of a small city, hidden because it is not yet “probable.” I have read a great many prospectuses in my years on the desk. That sentence is the kind that ends careers.
Alphabet’s purchase commitments and contractual obligations jumped to eight hundred and eleven billion dollars as of the end of June — up from three hundred and thirty-two billion dollars three months earlier, a hundred and forty-four percent in ninety days. The company said the commitments “primarily relate to technical infrastructure and inventory” and to agreements to secure energy for data center usage, with some obligations running out to twenty fifty-four. They did not detail why the obligations increased so substantially over the quarter. When a single company’s off-balance-sheet obligations jump by almost five hundred billion dollars in ninety days and the company declines to explain why, that is not disclosure. That is the dog that did not bark. Nvidia — the company on whose chips this whole apparatus is built — committed to twenty-seven billion dollars in equity investments between April and the end of its fiscal year, and has separately arranged a half-trillion dollars in financing with Wall Street firms to keep the supply chain liquid. The supplier of the picks and shovels is itself levered up to its chin strap.
Burke prosecuted the East India Company because a corporation that held dominion over a continent had no business hiding its books. Brandeis prosecuted the Money Trust because a handful of banks that held dominion over the credit of a nation had no business running their affairs in the dark. The conservative tradition — the one I was raised on, the one I still keep on the shelf — does not love business because business is efficient. It loves honest business because honest business is the kind a free country can be built on. What these nine firms have done is the opposite. The leverage is real. The risk is real. The exposure is concentrated in the hands of a few firms whose executives will never set foot in the Louisiana parishes where their data centers rise, whose bondholders will never see the servers they financed, whose employees will bear the layoffs if the demand forecast fails. They have built a structure in which the answerability has been engineered out.
I know this particular sin by its older name. A firm with ten million in capital could control a hundred million in corn. The leverage made the position look bigger than the firm. When the price moved against the bet, the firm went under, and the brokers — not the partners who placed the bet — paid the clearinghouse default. The losses got distributed. The gains got concentrated. The same men placed the next bet. That is what concentrated capital does when the discipline of transparency is removed: it loads the risk onto people who did not choose to bear it, and it lets the men who placed the bet walk away whole.
I remember when the rural electric cooperatives built the lines that electrified Adams County. My grandfather told me about it. The private utilities said it could not be done — too few customers per mile, the density did not justify the capital. The cooperatives did it anyway, with low-interest loans through the Rural Electrification Administration and member-owned structures that put the bill and the balance sheet on the same page. The co-op that serves my county still operates that way: every dollar of debt is on the books, every member can read the audit, the board is elected by the people whose power it carries. Adams-Columbia Electric Cooperative is the largest rural electric co-op in Wisconsin, and it is owned by the people it serves. The model is older than the AI buildout. It is what conservatism, properly understood, was supposed to defend: an infrastructure answer that centralizes nothing, that keeps the obligation transparent, that lets the people who use the system own it.
The men who built Hyperion could have built it that way. They chose not to. They chose the structure that hid the leverage, the holding company that hid the debt, the bond sale that hid the obligation from the people whose savings funded it. They chose the architecture of concealment because the architecture of concealment is the only architecture that lets a speculative bet reach three trillion dollars without the cost of capital adjusting to reflect the risk. That is not a market. It is the rentier, doing what the rentier has always done: borrowing against a future the rentier does not own, hiding the obligation from those who do, and walking away from the wreckage when the forecast fails. The Meta capex plan is the same bet, in the same posture, with the same disclosure.
I am not anti-market. I am anti-rentier. I am not anti-AI. I am anti-concealment. The men who built this bet could have built it on balance sheets they were willing to show their lenders. They could have kept the leverage where the cost of capital could discipline the position. They could have built smaller, slower, and in member-owned structures the way the rural electric cooperatives ran the lines that electrified my county in the nineteen-thirties. They chose the opposite. They chose the structure that maximizes the bet and minimizes the answerability. That is what the conservative movement, in its decades-long fusion with concentrated capital, has stopped being able to name. It is a sin against the very prudence the movement once claimed to defend.
The bill will come due. If the demand holds, the leverage will retire quietly and the men who placed it will be called visionary. If it does not, the holding companies will not pay, the bondholders will take their losses, and the counties that granted the tax abatements will be left with the shells of the buildings and the memory of a two-year construction crew where they were promised a permanent workforce. I have seen this before. I have watched the paper claims eat the real crops. I have watched the speculative leverage hollow a town. I have watched the men who placed the bet leave, and the people who were never asked to weigh in pay. There is another way, and it was built in this county in the nineteen-thirties, when the rural electric cooperatives strung the lines that lit every farm and every church on the sand plain. Three trillion dollars of disclosed co-op debt would have sat on the books where every member could read it, and the men who borrowed it would have had to answer for it at the next annual meeting. What concealment costs the rest of us is the power to say no: the men who hide the leverage never have to sit in the room where a member asks whether the bet is worth it. The footnote is not a technicality. It is the confession.