If the AI Boom Breaks, It Will Break Through Credit, Not Code

If the AI Boom Breaks, It Will Break Through Credit, Not Code

The common fear about an AI correction is that the technology will disappoint. TAN's position is different. If a correction comes, the damage will spread through the way AI is being financed, not through any failure of the models. For India, the direct risk is small, but the indirect risk is real and its compute plans should be built accordingly.

The reasoning Spending is running far ahead of revenue.

Leading US hyperscalers expect AI capital expenditure of about $725 billion in 2026, against $410 billion in 2025. Revenue is not keeping pace. OpenAI indicated its September annualised revenue was nearer $50 billion than the roughly $70 billion signalled earlier and Anthropic recorded an operating loss of about $8 billion in 2025.

The way this spending is funded matters most. Hyperscalers once paid for expansion from their own cash. Increasingly, they rely on debt, including private credit, which is opaque and lightly supervised. Goldman Sachs reported, as cited in the op-ed, that most hyperscaler bonds issued in 2026 had fallen below their issue price by end-August.

Those bonds and loans sit with banks, insurers and pension funds. A fall in AI valuations would therefore reach savers and lenders who never chose to bet on AI. The IMF made the same point in its October 2026 Global Financial Stability Report.

It warned that markets remain resilient largely because of optimism about AI, but that "resilience is not safety". It identified highly leveraged non-bank financial institutions as the channel through which a correction could spread.

The strongest counter-argument JPMorgan's strategists argue that current valuations are justified.

Their case is that elevated multiples anticipate above-trend earnings growth and a genuine capex cycle, with global hyperscaler capex projected to reach $1 trillion by 2027. On this view, demand is real and today's spending will produce tomorrow's profits.

Why the

position still holds The counter-argument may be right about demand and still miss the risk. Even if AI eventually earns its returns, the timing matters. Debt has fixed repayment schedules and revenue that arrives a few years late can still break a lender. Firmus shows that investors are already questioning valuations: it was a well-backed infrastructure company, yet it could not list at the price it sought.

What it means for India India's direct exposure is limited.

It holds only about 3 percent of global data-centre capacity and its main AI programme, the IndiaAI Mission, provides over 38,000 GPUs at subsidised rates. A correction would not hit Indian balance sheets the way it would hit US ones.

The second-order risk is different. If foreign investors pull back, the capital India expects for data centres and compute could become scarce and expensive. India's plans should therefore not assume that boom-era foreign money will continue to flow.

Public compute, domestic financing and phased capacity building are safer foundations.