The $690 Billion Question: How the AI Data-Centre Boom Became the Market's Engine , and Its Biggest Risk

For much of the past two years, a single force has done more than any other to lift global stock markets to record highs: the colossal wave of spending on artificial-intelligence infrastructure. What began as an arms race between a handful of technology giants has become the dominant story in finance, and understanding it is now essential for anyone trying to make sense of where markets go next.

A spending wave without precedent

The sheer scale of the money involved is difficult to comprehend. In 2026 alone, the five biggest spenders are on track to pour staggering sums into AI compute, data centres and networking. Amazon is projecting around 200 billion dollars in capital expenditure, Alphabet between 175 and 185 billion, Meta somewhere between 115 and 135 billion, Microsoft tracking toward 120 billion or more, and Oracle targeting roughly 50 billion.

Taken together, these five companies are set to spend somewhere in the region of 660 to 690 billion dollars on infrastructure this year, with the overwhelming majority directed at the hardware and buildings needed to train and run artificial intelligence. It is a figure that rivals the entire economic output of medium-sized nations, concentrated in the hands of a few corporations.

And this is no one-off surge. According to projections from Goldman Sachs, total spending by these so-called hyperscalers between 2025 and 2027 could reach around 1.15 trillion dollars, more than double the roughly 477 billion spent in the preceding three years. For the wider supply chain, from chip designers to power utilities, this represents a powerful, multi-year tailwind that few other sectors can match.

The circular financing puzzle

Yet as the numbers have grown, so too have the questions about how it is all being funded. The most striking concern centres on what analysts call circular financing, an arrangement in which the company selling the picks and shovels of the AI gold rush is also helping to bankroll the customers buying them, creating a loop that can flatter demand.

The clearest example involves the chip designer at the heart of the boom. Beyond the roughly 30 billion dollars it has already invested in OpenAI, Nvidia has reportedly been in discussions to finance as much as 350 billion dollars of that company's chip purchases. In effect, the supplier would be lending its biggest customer the money to buy its own products, a dynamic that makes headline demand harder to interpret.

Nor is this an isolated case. Google has agreed to backstop lease payments across several data-centre locations for Anthropic, an OpenAI rival, in an arrangement that helps the smaller firm secure what amounts to a loan of around 35 billion dollars. When the giants of the industry are underwriting one another's expansion, it becomes far more difficult to judge how much of the demand is genuinely organic.

The shadow on the balance sheet

There is also the matter of what does not yet appear in the headline accounts. In early 2026, the ratings agency Moody's reported that hyperscalers had signed approximately 662 billion dollars in data-centre lease commitments that had not yet formally commenced. Under prevailing accounting rules, these obligations can sit off the balance sheet until the leases begin.

The trouble is that the economic reality of these commitments is no less binding for the way they are treated in the accounts. Critics describe them as a kind of shadow liability, a vast set of future payments that investors may be underestimating simply because they are not yet fully visible in the standard financial statements the market relies upon to gauge risk.

The core worry that ties all of this together is a familiar one in the history of investment manias. Capital expenditure is growing far faster than the revenue and profit it is supposed to eventually generate. As long as the promise of future AI earnings holds, the spending looks visionary; if that promise slips, the same numbers could look reckless in hindsight.

Supercycle or bubble

This is the central tension facing investors. On one side stand the optimists, who see a genuine technological supercycle that will reward everyone positioned along the AI supply chain, from semiconductor designers to the data-centre operators and power utilities that keep the machines running. For them, today's spending is simply the price of building the future.

On the other side stand the sceptics, who point to the circular deals, the off-balance-sheet leases and the widening gap between spending and profit as classic warning signs. Whether the 690-billion-dollar bet of 2026 is remembered as the foundation of a new era or the peak of an overheated cycle will not be settled quickly, but for now it remains the single most important question hanging over the markets.