The Debt Wave: Big Tech's Trillion-Dollar Borrowing Spree
The most profitable companies in history just borrowed over a trillion dollars in twelve months — Amazon $62B, Meta $80B+, Broadcom $35B — to fund the AI buildout. The boom has moved from believers to creditors, and creditors are the constituency that ends cycles.
By MyAudioBooks.ai ·
Listen free: The Debt Wave: Big Tech's Trillion-Dollar Borrowing Spree
There is a number in the deal data that changes how you see the entire artificial-intelligence boom, and it is not a valuation. It is a borrowing total. From the trailing twelve months of completed private-market transactions — the same deal screen this channel used to trace the SpaceX stack — ninety-seven debt-type transactions totaling one point one four trillion dollars, and the names on them are not distressed borrowers or overleveraged roll-ups. They are the most profitable companies in human history. Amazon borrowed sixty-two billion dollars across two financings. Meta borrowed more than eighty billion across three, including a twenty-seven-billion-dollar joint-venture structure. Broadcom raised thirty-five billion in debt. Alphabet twenty-five. ByteDance twenty-nine point six. NVIDIA twenty-five. Oracle twenty-five. In the space of a single year, the companies with the strongest balance sheets in the history of business went, collectively, to the bond market for over a trillion dollars — and the reason they went tells you more about the AI era than any product launch, any model benchmark, or any keynote.
The AI boom, in its first phase, was funded the way the tech industry funds everything: out of operating cash flow, the river of money the platforms' existing businesses generate every quarter. The graphics cards, the first data centers, the model runs of twenty twenty-three and twenty twenty-four — the platforms paid cash, from earnings, without touching debt markets or diluting a share. That phase is over, and the deal data marks its end precisely. The capex of the second phase — the gigawatt campuses, the multi-year buildouts, the AI factories measured in power-plant units — has outgrown even those rivers. A single hyperscale AI campus now costs more than a year of some of these companies' entire capital expenditure from five years ago, and there are dozens of campuses under construction at once. The cash flows are still enormous. They are no longer enough. And so the most profitable companies on Earth did something almost none of them had needed to do in a generation: they went back to the bond market, at scale, in a wave that has quietly become one of the largest corporate borrowing episodes on record.
This is the story of the debt wave — why the smartest balance sheets in the world chose leverage when they did not strictly need it, how the bond market became the AI boom's final financier, and what changes when the technology revolution is funded by people who expect to be repaid.
At My Audio Books dot A I, you can create your own audiobooks from prompts, turn your documents into audio, all with one subscription, and store your items in your own personal library.
Start with the strangest question in the whole episode: why do companies sitting on tens of billions in cash borrow at all? The answers, in order of how much they matter, explain the entire shape of the wave. The first is the capex arithmetic, and it is simply bigger than anything the platforms have ever attempted. The AI buildout is not a software program; it is an industrial construction program — land, power, substations, buildings, cooling plants, networking, and the chips to fill them — and its defining feature is that it must all be built before it earns a dollar, at scales where a single year's program runs to the tens of billions per company. The platforms could fund that from cash. They would rather not, because cash is the strategic reserve for everything else — acquisitions, buybacks, the war for researchers paid like athletes — and because the next downturn, whenever it comes, is not the moment to be explaining to shareholders why the reserve was spent on concrete. The second answer is structural: much of the platforms' cash is offshore, earned in Ireland, Singapore, and the other jurisdictions where their international profits live, and repatriating it — bringing the money home across the tax border — to pour concrete in Louisiana has tax costs that borrowing in New York does not. The third answer is the bond market itself: investment-grade debt — the borrowing of companies rated as near-safe as governments — currently prices at spreads so tight that the after-tax cost of borrowing, for these companies, is functionally cheaper than the return their own cash earns. When the bond market offers you money that costs less than the value of keeping your own, you take the money. The fourth answer is the subtlest, and the deal data shows it directly in the structure of the biggest deals: some of the borrowing is deliberately built to stay off the balance sheet — joint ventures and special-purpose vehicles, like the twenty-seven-billion-dollar structure in Meta's name, in which the debt legally belongs to a project entity rather than to the platform itself, keeping the parent's leverage ratios pristine while the campuses get built. The borrowing is not a sign of weakness. It is a choice made from strength, and it is a choice with a customer: the bond market, which has decided it believes in AI.
Now the wave's shape, because it is not uniform — it has tiers, and the tiers tell you how the risk is being distributed. At the top, the platforms borrow in their own names at the finest rates: Amazon's sixty-two billion, Alphabet's twenty-five, Oracle's twenty-five — general corporate purpose bonds that buy the fleet, the campus, the power contracts, everything, with full faith and credit behind them. The second tier is the structured tier, and it is where the genuinely new financing is happening: the project-finance structures in which a data-center campus is owned by a special-purpose vehicle, funded by its own debt against its own long-term lease to the platform, so that the platform gets the building and the lender gets a mortgage on the AI age. The twenty-seven-billion-dollar joint venture in the deal data is the emblem of this tier — the largest single structure of its kind this year — and the cleverness of it deserves a sentence of honest admiration: the platform commits to rent the campus for decades, the vehicle borrows against that commitment, the debt never touches the platform's ratios, and the whole arrangement works precisely as long as the rent keeps being worth paying. The third tier is the refinancing tier, the quiet tell: NVIDIA's twenty-five billion in debt refinancing — replacing older loans with new ones on better terms — the revolving facilities being upsized across the sector, the lenders being asked to extend and expand — the plumbing of the credit system being widened in anticipation of the next phase, before anyone can quite say what the phase will cost.
It is worth naming who the lenders actually are, because the buyer list is the wave's most revealing document. The platforms' bonds are being absorbed by the investment-grade complex: the giant asset managers running the bond indices, the insurance companies matching thirty-year promises with thirty-year paper, the pension funds whose mandates require them to own the safest corporate debt in the world, and — in the structured tier — the private-credit giants this channel profiled last month, who have discovered that an AI data center with a twenty-year platform lease is the closest thing the new economy offers to a power plant. The platforms are not borrowing from gamblers. They are borrowing from the institutions that hold the retirement savings of the developed world, and those institutions are bidding against each other to lend.
At My Audio Books dot A I, you can listen to this story and thousands of others that explore the hidden science and mechanics behind the headlines.
What does it mean that the boom is now bond-funded? Three things, and each one changes the frame. The first is a change in who judges the boom. Equity investors are dreamers by construction — they buy upside and forgive delays, and their patience with an AI story is measured in years of narrative. Bond investors are the opposite: they buy the downside, they are paid to be repaid, and they price that expectation with a cold precision that no earnings call can spin. The AI boom now answers, every quarter, to a constituency that does not care about the vision — it cares about the cash flows that service the debt. The bond market has become the buildout's first honest auditor. The second change is duration — the mismatch between the life of the asset and the life of the loan. A data center campus has a physical life of fifteen to twenty years, but the chips inside it may be economically obsolete in three to five, and the loans funding them run ten, twenty, thirty years. The bond market is lending twenty-year money against a three-year asset cycle wrapped in a fifteen-year building, and the question nobody can answer yet is what those buildings are worth in year seven, when the chips inside them are two generations behind. The lenders are not pricing that question today. They will have to eventually. The third change is systemic reach, and this is where the debt wave touches everyone who owns a bond fund: the wave is large enough that the AI buildout is now embedded in the investment-grade bond indices that pensions, insurers, and ordinary retirement accounts hold as their safe allocation. The equity boom put AI in everyone's stock portfolio whether they chose it or not. The debt wave is putting AI in everyone's bond portfolio too — the boring, careful, supposed-to-be-safe part. When the boom stumbles, if it stumbles, the reverberation will not stop at the growth funds. It will arrive in the funds that retirees were told not to worry about.
The strongest case against the alarm — the case that the debt wave is healthy, and it is strong, because these are genuinely the strongest borrowers on Earth — deserves a full hearing. The platforms' leverage, even after the wave, is modest by any industrial standard: their net debt relative to their cash generation is a fraction of what ordinary blue-chip companies carry comfortably, and their interest payments are covered by operating income not twice or three times but many times over. They are borrowing because it is optimal, not because they are desperate — because cheap long-term money against long-lived assets is textbook corporate finance, and the offshore-cash logic makes it nearly free. The bond market is not being fooled; it is lending to companies with the deepest cash moats in business history, against data centers those companies desperately need and will fill with their own products. The structured tier is not hiding risk; it is allocating it to the investors who want exactly that risk — infrastructure debt with long leases — and keeping the parents' balance sheets clean for actual storms. And the capex is not speculative in the classic sense: the platforms are selling AI capacity as fast as they can build it, their cloud revenues are growing at rates that justify the buildout, and the gap between supply and demand is the reason the campuses are being built at all. On this reading, the debt wave is what a mature, well-financed technology buildout looks like: the largest, safest companies in the world using the cheapest capital available to build the infrastructure of the next era. It is not a bubble. It is a construction program, financed the way construction should be.
And the strongest case for watching the debt rather than the headlines is written in the history of every capital cycle that ever ended. Capital expenditure cycles do not end when the story is disproved; they end when the returns on the last dollar invested fall below the cost of that dollar — and the debt wave matters precisely because it makes that measurement unavoidable. Equity can be patient with narrative; debt service cannot. The day the AI buildout's incremental revenue stops covering its incremental cost of capital, the bond market will be the first to know, because it is the bond market that is owed. The structured tier has its own fragility: the off-balance-sheet vehicles work as long as the platforms' long-term commitments to them hold — commitments that are themselves bets on AI demand lasting twenty years, signed by companies whose planning horizon is five. The asset-cycle mismatch is the quiet one, and the history here is older than telecom. Every great capital cycle of the modern era was financed, at its peak, by the bond market's finest names: the railroads of the eighteen seventies, whose bonds were the investment-grade paper of their age, funded twice as much track as the traffic could carry and took a generation of investors down with the paperwork; the electrification bonds of the nineteen twenties, sold against utility holding companies that were models of structured finance until the structure was the story; and the telecom buildout of the late nineteen nineties, which left the world with dark fiber that was physically fine and economically dead for a decade, and the companies that borrowed to lay it are not the companies that survived it. The pattern is not that the technology failed — the railroads carried the freight, the utilities lit the cities, the fiber carries this article. It is that the borrowing was timed to the euphoria, and the euphoria never called to say when it was leaving. And the systemic reach is the one that should concentrate every careful mind: when the safe part of every retirement portfolio in America contains the debt of the AI boom, the AI boom's risk is no longer a tech-sector question. It is a fixed-income question, and fixed-income questions do not stay in their sector.
Three developments would disprove or confirm which reading of the debt wave is right, and each is observable in the same data. First, the spreads: the borrowing costs the platforms pay on their next offerings are the bond market's live confidence rating on the buildout — if spreads widen materially while the borrowing continues, the auditor is losing faith; if they stay tight, the bond market's belief is holding. Second, the structured market: if the off-balance-sheet structures keep pricing at or near the parents' own rates, the lenders are underwriting the platforms rather than the projects — the first time one of those vehicles prices at a real discount to its sponsor, the structure's illusion breaks, and the risk transfer is revealed as risk retained. Third, the utilization numbers: the buildout's own companies will eventually have to report how full the campuses are running — the day the utilization data shows the new capacity arriving faster than the revenue to fill it, the capex cycle has turned, and the debt service clock starts ticking in a room the equity market cannot see into.
It is worth saying what this article has not claimed. It has not claimed the platforms are overleveraged; by their own metrics, they are underleveraged, and the article says so at length. It has not claimed the AI buildout will fail; the demand evidence is real, and the bull case here is stated at full strength. It has not claimed the bond market is naive; the lenders are the most skeptical audience in finance, and their continued appetite is itself evidence. And it has not claimed the structures are deceptive; they are disclosed, rated, and analyzed openly — the article's question about them is about risk allocation, not concealment. The claim here is narrower and more consequential: the AI boom has moved from being funded by believers to being funded by creditors, and creditors are the constituency that ends cycles. The equity market's faith moved the boom's first phase. The bond market's math will decide its second.
Which returns to the one point one four trillion, and the strangeness of the sentence it makes you say out loud: the richest companies in history are now also some of the largest borrowers in the world, and they are borrowing to build a machine that has not yet proven it can pay them back at the rate the money costs. That is not a warning of collapse, and this article has refused, carefully, to become one. It is a change of regime. The phase where the boom answered to dreamers is over. The phase where it answers to accountants has begun — and the accountants, unlike the dreamers, can count exactly how much time is left on the clock. The debt wave is not the end of the AI story. It is the moment the story got a lender, and lenders always read to the last page.
At My Audio Books dot A I, you can create fiction, non-fiction, and turn your documents into audio, all stored in one place with a single subscription — plus get instant access to thousands of audiobooks and deep-dive investigations. Learn more today at My Audio Books dot A I.