Who Is Actually Paying AI's $3 Trillion Bill?

Creators lock in their holiday content calendars 90 days out, before most ecommerce brands finalize their commission strategy and way before Black Friday and October deal events.

Get ahead of the seasonal rush with The 90-Day Holiday Sprint, a practical guide for brands that want creators driving holiday demand while competitors are still recruiting:

Your 90-day countdown starts now.

Get the Free Guide

1.  Even the richest companies on earth cannot pay cash for this.

The data centers needed for the AI boom carry a price tag above $3 trillion, and as Bloomberg put it, not Amazon, not Microsoft, not Meta is prepared to foot that bill from their own cash. Equity investments in OpenAI and Anthropic do not come close either, and subsidies only stretch so far. So the money is coming from debt markets. All of them: blue-chip bonds, junk debt, private credit.

2.  The borrowing numbers are outside anyone's experience.

Morgan Stanley and Moody's put the capital expenditure at $3 trillion or more, while JPMorgan projects above $5 trillion including power supplies. Bank of America estimates AI-related companies raised at least $200 billion in debt last year, likely an undercount. One BofA executive's summary: the numbers are like nothing anyone in the business for 25 years has seen.

3.  Not everyone is borrowing at the same price.

The rates tell you how lenders rank the risk. xAI raised $5 billion in bonds and loans with fixed portions carrying a 12.5% coupon, and CoreWeave borrowed $3.75 billion at around 9% in the high-yield market. Meanwhile hyperscalers issue investment-grade paper at a fraction of that. Same boom, wildly different cost of money, which is the market's honest scorecard on who it thinks will still be standing.

Why it might be fine.  The strongest argument is that this is not the dot-com era. As Moody's John Medina notes, most of this investment is backed by established, profitable companies with strong existing revenue, not speculative startups with no customers. Hyperscalers also often build through joint ventures and special purpose vehicles specifically to avoid overloading their own balance sheets. Real demand, real revenue, structured financing.

Why it might not be.  The counterargument is concentration. JPMorgan credit strategists warn that bond portfolios, which historically tracked interest rates and banks, could become correlated with the operating fortunes of a handful of large technology companies. That matters because the Magnificent 7 already account for roughly a third of the S&P 500's value, so equity and now credit exposure point at the same few firms. Diversification gets harder exactly when you would want it most.

The honest uncertainty.  Notice that both cases cite the same data. The disagreement is not about the numbers, it is about whether AI revenue arrives fast enough to service the debt. Nobody knows that yet, and anyone who tells you they do is selling something. What is measurable is the trajectory: we covered Meta doubling compute by 2027, Alphabet lifting capex toward $205 billion, and Anthropic's roughly $1.25 billion monthly compute bill within the last month alone.

The takeaway.  For you as a user rather than an investor, the practical read is more useful than the forecast. All that borrowed capacity has to be sold, which is a large part of why token prices keep falling and free tiers keep improving. That is a real benefit today. The corresponding risk is dependency: if financing tightens, the cheapest providers are the ones most likely to raise prices or disappear. Build so you can switch, keep your prompts and data portable, and enjoy the discount while the money is flowing.

The record bond year.  Morgan Stanley expects $250 to $300 billion of debt issuance in 2026 from hyperscalers and their joint ventures alone, enough to push the whole investment-grade bond market to record volumes. AI is no longer just a tech story, it is a fixed-income story.

Convertibles are back.  Global convertible bond issuance hit a 24-year high of $167 billion in 2025, driven by AI demand, with CoreWeave issuing $2.25 billion at a 1.75% coupon. Cheap borrowing, paid for later in shareholder dilution.

Data centers as securities.  JPMorgan projects annual data center securitization could reach $30 to $40 billion in 2026 and 2027, up from about $27 billion in 2025. Server halls are being packaged into tradeable instruments, which is either financial maturity or a familiar warning sign depending on your reading of history.

Unease, but not restraint.  The line that captures the mood: for credit investors, AI is hard to resist even when it comes with a sense of unease. Everyone can see the risk. Nobody wants to sit out the round.

Stress-test your dependence on AI tools

You cannot control whether the financing holds. You can control how trapped you would be if your favourite tool doubled its price next year. This prompt maps your lock-in and gives you a portability plan, which is worth doing whether the boom continues or not.