AI runs on debt: $570 billion in issuance and $3 trillion off the balance sheet

Artificial intelligence is no longer paid for out of profits. The Chinese tech outlet TMTPost, citing Morgan Stanley estimates, reports that global AI-related debt issuance in 2026 is approaching $570 billion, twice the level of a year earlier.

Capital spending is larger still. The five biggest technology companies are set to spend between $725 and $769.2 billion on investment in 2026. Sequoia Capital puts global AI infrastructure spending at about $1.5 trillion, and notes that the profitability threshold for the whole industry has risen to $3 trillion.

The most troubling indicator is off-balance-sheet obligations. On the figures cited, they grew from $1.65 trillion in July 2026 to nearly $3 trillion in August. Cash flows show the effect. In the second quarter of 2026 Alphabet posted its first negative free cash flow, minus $5.9 billion, and Amazon also slipped below zero.

A second Chinese-language source describes the same shift: an interview with Bloomberg Economics' chief economist for Asia-Pacific, published by Yicai. In 2025, capital spending by technology companies in the top fifty largest firms in the world exceeded $500 billion and for the first time in history surpassed their cash reserves. In 2026 the financing moved to the bond market. Net borrowing by technology companies was twice what it had been a year earlier, far outstripping simultaneous stock market debuts in scale.

The Bloomberg Economics economist describes this as a shift from a "savings glut" to an "investment glut" driven by AI and by armaments. The macroeconomic effect is asymmetric. A growing supply of technology company bonds pushes up the long end of the yield curve, while monetary policy acts on the short end. In the same month the Fed raised its rate by 25 basis points, to a range of 3.75 to 4.00 percent, the first time in more than three years.

Regulators already see the risk. In September 2026 Japan's Financial Services Agency for the first time made AI data centre financing the target of a special audit. The agency focused on the exposure of the largest banks and insurers to projects in the United States. Typical projects are funded with leverage of up to 90 percent.

Macroeconomic conditions are no longer favourable. The Federal Reserve raised its rate by 25 basis points, to a range of 3.75 to 4.00 percent, for the first time in more than three years. Yields on two-year paper rose, and ten-year yields approached the 5 percent threshold. Infrastructure projects are typically funded with leverage of up to 90 percent, so every basis point changes the profitability calculation.

The open question is the obligations that do not appear on operators' balance sheets. They will decide whether the current cycle ends in a valuation correction or in lasting demand for computing capacity. Spending forecasts are impressive, but they do not set the limit of the boom. The profitability of a single project does.