Eventually, the bill comes due. It always does – though in Silicon Valley it arrives dressed as a footnote, slips past the auditors, and takes a seat at the table long before anyone thinks to ask who invited it.
The room, as ever, is crowded. Microsoft sits at the head, cufflinks glinting, insisting the evening is comfortably within budget. Meta lingers near the bar, spending like a man who has recently discovered religion and intends to tithe in GPUs. OpenAI, the prodigy everyone indulges, orders another round of compute without once glancing at the menu. And in the corner, rather unfashionably solvent, Oracle nurses a drink and pays its debts in plain sight, which in this company feels almost vulgar.
One must admire the choreography. The balance sheets, at first glance, are models of restraint. Microsoft’s debt-to-equity ratio hums along at a polite sub-one figure. Meta’s is practically ascetic, the sort of number that suggests a monkish devotion to prudence. Oracle, by contrast, looks like it has been gambling again—its ratio lurching above five, all its borrowings laid out like empty glasses after a long night. If one were naïve, one might conclude that the old guard is reckless and the new aristocracy positively virtuous.
But then, virtue has always depended on what one chooses to count.
The modern technology balance sheet has developed a certain coyness. Debt, in the traditional sense – loans, bonds, the sort of obligations that arrive with interest rates and unpleasant reminders – has fallen out of fashion. In its place: commitments. Agreements. Partnerships. Words that sound like they belong in a wedding announcement rather than a creditor’s ledger.
Microsoft, for instance, has obligingly disclosed more than $150 billion in future contractual obligations, a sum so large it begins to feel abstract, like the national debt or the concept of time. These commitments include leases for data centres, capital expenditures for infrastructure, and, most delicately, the vast computational appetite of its relationship with OpenAI. Azure, one is told, will provide the compute. Over many years. At considerable expense. But expense, you see, is not quite the same as debt – at least not when one is writing the invitation list.
And OpenAI itself has perfected the art of dining on someone else’s tab while promising to become the restaurant. Its obligations are not framed as borrowings but as dependencies: on Microsoft’s cloud, on continued funding, on the assumption that tomorrow’s revenues will politely exceed today’s appetites. It is a financial structure that relies less on leverage than on belief, which is to say it is perfectly suited to its era.
Meta, never one to be outdone in spectacle, has embarked on its own infrastructure crusade. Tens of billions in annual capital expenditure, much of it devoted to AI data centres and the silicon that powers them. The official balance sheet remains trim, like a Dame squeezed into a corset. Debt is low. Ratios are flattering. And yet, buried in the filings, there they are: lease obligations, purchase commitments, promises to buy servers and chips in quantities that would make a small nation blush. Not debt, precisely. Just… inevitabilities.
It is here that the cloud, that most ethereal of inventions, reveals its rather solid underpinnings. Compute is not a metaphor. It is steel and concrete and electricity, paid for in instalments that stretch obligingly into the future. The genius, in the sense of a carnival barker stumbling upon snake oil, has been to recast these obligations as operating necessities rather than financial liabilities. One does not borrow; one subscribes. One does not incur debt; one scales.
The recent courtships make the pattern clearer still. Google’s multibillion-dollar dalliance with Anthropic is not merely an investment but a mutual entanglement: capital on one side, cloud commitments on the other, each binding the future a little tighter. The arrangement ensures that Anthropic will consume Google’s compute for years to come, while Google secures a tenant who cannot easily leave. It is less a transaction than a long-term cohabitation, with all the financial implications that polite society prefers not to discuss.
Then there is Nvidia, the indispensable arms dealer to this particular race. Access to its GPUs has become a currency in its own right, exchanged through prepayments, supply guarantees, and strategic alignments that look, from a certain angle, very much like borrowing against the future. When compute itself becomes scarce, the obligation to secure it begins to resemble debt, even if no one is gauche enough to call it that.
Against this backdrop, Oracle’s apparent profligacy takes on a different hue. Its leverage is explicit, its borrowings recorded with a kind of old-fashioned honesty. It has taken on real debt to finance acquisitions and expansion, and the numbers reflect it without apology. There are no elaborate disguises, no romantic language of partnership. Just liabilities, sitting squarely where one can see them.
Analysts, those tireless chaperones of financial propriety, have begun to murmur that the new aristocracy’s balance sheets may be understating the true extent of their obligations. Off-balance-sheet commitments, they note, behave economically like debt. They bind future cash flows, constrain flexibility, and, in less forgiving climates, demand to be paid all the same. Adjust for these, and the neat hierarchy of leverage begins to wobble. Microsoft’s modest ratios look less modest. Meta’s restraint appears more theatrical. The gap between them and Oracle narrows, perhaps uncomfortably.
But theatre, after all, is the point. Silicon Valley has always understood that perception is a kind of capital, and nowhere is that more evident than in how it presents its finances. A clean balance sheet reassures investors. It suggests control, discipline, and the absence of unpleasant surprises. Footnotes, by contrast, are where one hides the bodies.
The risk is not that these companies cannot pay. For now, they are flush with cash, buoyed by revenues and, more importantly, by expectation and the greed of investors. The risk is that the expectations themselves are doing a great deal of the accounting. AI, we are assured, will justify the spending. The demand will come. The models will scale. The future will be generous.
It usually is. Until it isn’t.
Because the bill, in the end, is indifferent to narrative. It does not care whether an obligation was called a lease or a loan, a partnership or a purchase commitment. It arrives, as bills do, with numbers attached and a due date that refuses to be charmed. And when it does, one suspects the room will look rather different – less like a salon of visionaries, more like a gathering of very clever people wondering, quietly, who exactly ordered all this.
Leave a Reply