The artificial intelligence buildout is increasingly a borrowing story, not just a technology story. As hyperscalers prepare to spend more than 1.3 trillion dollars in 2027, according to S&P Global Ratings, they are turning to debt markets to fund a programme that their current cash flow cannot cover, and the scale of that borrowing is now a risk factor in its own right.
Goldman Sachs estimates that about 400 billion dollars in investment grade bond issuance will be needed in 2027 to fund the buildout by five United States hyperscalers, against roughly 1.14 trillion dollars in capital expenditure. S&P expects all six of the largest hyperscalers to produce negative free operating cash flow in both 2026 and 2027, with recovery not projected until 2029. In other words, the companies building the future of AI are, for now, spending more cash than they generate from operations, and bridging the gap with lenders.
Balance sheets show how fast that gap has opened. Alphabet carried 98.2 billion dollars in long term debt as of June 30, a figure 568 percent higher than exactly four years earlier, before ChatGPT was released, according to research compiled by Vanguard. Meta carried 83.7 billion dollars in long term debt, having had no debt on its balance sheet four years earlier. Alphabet reported negative free cash flow of 5.9 billion dollars in the second quarter, while Meta’s free cash flow fell 91 percent year over year, and analysts expect both businesses to burn cash in 2027.
Leases, the hidden borrowing
Debt on the balance sheet understates total commitments, because hyperscalers also sign long term leases for data centre capacity. S&P projects Microsoft has more than 329 billion dollars in future lease obligations, ahead of Meta at 279 billion dollars. Those leases are a form of borrowing in economic terms, even if they are not counted as debt, since they commit future cash to landlords and infrastructure providers for years ahead.
The rating agency has not taken rating actions on the back of its report, and its central message is that strong current credit quality does not remove the financing question. Investment grade companies can borrow cheaply, and they are doing so, but the test will come if AI revenue grows more slowly than expected, or if bond markets demand higher yields for longer dated technology debt. A buildout funded by borrowing is sustainable only if the assets it buys start earning within the lease and bond maturities that fund them.
What investors should watch
Three indicators will tell whether the borrowing bet is working. First, AI revenue growth, against Goldman’s estimate that roughly 300 billion dollars in annual AI revenue is needed to break even. Second, free cash flow trends, especially whether Microsoft remains the only hyperscaler in positive territory in 2027, as S&P projects. Third, bond issuance volumes and yields, which will show whether debt markets remain willing to fund the buildout at current prices.
Bain and Company estimates that annual spending on AI infrastructure could reach 1.5 trillion dollars by 2031, according to reporting on its research. If that path holds, borrowing will remain a central feature of the AI economy for the rest of the decade, and the companies that manage their debt and lease maturities best will have an advantage that has little to do with model quality, and everything to do with financial discipline.
