--- title: "Big tech borrowed for the AI build. Alibaba just sold shares instead" description: "US technology companies have issued at least $220 billion of debt this year for data centers. Alibaba raised $10.2 billion of equity on Sunday. The two instruments put the risk in different places, and that is the whole distinction." category: "Markets" category_url: https://boursel.com/category/markets author: "Sofia Marchetti" published: 2026-08-23T13:37:15.000Z updated: 2026-08-23T13:37:15.000Z canonical: https://boursel.com/article/big-tech-borrowed-for-the-ai-build-alibaba-just-sold-shares-instead tags: ["ai capex", "corporate finance", "debt", "equity", "analysis", "dilution"] --- # Big tech borrowed for the AI build. Alibaba just sold shares instead US technology companies have issued at least $220 billion of debt this year for data centers. Alibaba raised $10.2 billion of equity on Sunday. The two instruments put the risk in different places, and that is the whole distinction. *Analysis.* Alibaba's $10.2 billion share placement on Sunday is a small number against the scale of what is being built. Its interest is in the choice of instrument, because almost every other large AI buildout has been paid for with borrowed money. The contrast is worth setting out carefully, since the two options are usually discussed as if one were simply cheaper than the other. They are not substitutes with different price tags. They allocate the same risk to different people. ## What has been happening so far US technology companies have issued [at least $220 billion of debt this year to fund data centers, against $12.5 billion in the same stretch of 2025](/the-bond-market-has-started-charging-more-for-ai-even-to-the-safest-borrowers). That is roughly a seventeenfold increase in a single year. Amazon alone [returned to the bond market for at least $25 billion](/amazon-seeks-to-borrow-25-billion-as-investors-grow-wary-of-ai-debt). The market has begun to charge for it. Technology paper now trades wider than the investment-grade market it used to trade through, which is a reversal of a long-standing relationship, and demand for each successive deal has thinned. Alongside the borrowing sits a larger set of obligations that carry no interest coupon at all. A Wall Street Journal analysis of filings put nine companies at [roughly $3 trillion of AI commitments](/nine-companies-have-signed-3-trillion-dollars-of-ai-commitments-that-sit-off-the), about $1.9 trillion of purchase commitments and $1.2 trillion of leases signed but not started. Those are contractual promises to spend, disclosed in the notes rather than on the balance sheet, and they behave like fixed costs whether or not the revenue arrives. ## How the two instruments fail Debt is a fixed claim. The interest is owed on schedule and the principal is owed at maturity, and neither obligation cares what the asset turned out to be worth. If a data center earns its projected return, debt is the cheaper way to have built it, because the lenders' upside was capped at the coupon and everything above that accrued to shareholders. If it does not, the fixed claim is still there, and it is senior. The company refinances into a worse market, or sells something, or the equity absorbs the loss. Equity is a residual claim. New shareholders take a proportional slice of whatever the business turns out to be worth, good or bad, and there is nothing to repay. What it costs is permanent: the existing owners' share of all future profits is smaller, forever, in exchange for cash today. So the question a company answers when it picks one is not really about cost of capital. It is about how confident it is in the timing of the return, and how much fixed obligation it already carries. ## Where the timing question bites This is where the AI build differs from the ones it gets compared to. A data center's cash flows depend on demand for compute several years out, at prices nobody can currently observe, from customers whose own business models are unsettled. That is a wide distribution of outcomes. Fixed claims sit badly on wide distributions. They are efficient when the downside is bounded and unremarkable, and they are dangerous when it is not, because the obligation is rigid precisely where the asset is uncertain. That is the argument for funding this particular buildout with equity, and it holds regardless of what any individual company decides. There is a counter-argument and it is not weak: the companies doing the borrowing generate enormous operating cash flow from businesses that have nothing to do with AI, and that cash flow, not the data centers, is what services the debt. On that reading the fixed claim is safe because it is covered many times over by search advertising or retail or cloud revenue that already exists. Both can be true, and which one dominates depends on a company's own numbers rather than on a general principle. ## What to take from it Alibaba's placement dilutes its holders by roughly 4 percent and adds no fixed obligation. Amazon's bonds diluted nobody and added a schedule of payments. Neither is the correct answer, and any account that presents one as obviously smarter than the other has skipped the step where you look at what else the balance sheet is already carrying. The number worth watching is not how much is being raised. It is the ratio of fixed claims to the cash flow that does not depend on AI working. That figure is in the filings, it is not in the headlines, and it is what will separate the companies that can wait for the payback from the ones that cannot. *This is analysis, not investment advice.*