US$700 billion of AI spending now has to show a receipt

Friday’s tech rout was a verdict on capital discipline, not on AI as a technology. Amazon missed expectations and promised to spend even more. Intel sank nearly 30% on a weak outlook, and the Nasdaq fell 2.9% to 25,437.00, capping an ugly week. The four hyperscalers, Amazon, Microsoft, Alphabet and Meta, are heading for more than US$700 billion in capital investment this year, and TechCrunch puts the figure above US$740 billion.

That is no longer being read as confidence. It is being read as risk.

Big spending used to be the good news

The sentiment shift happened inside a single earnings season. TechCrunch put it bluntly, noting that investors once rewarded big outlays as a sign of faith in AI’s growth, but that enthusiasm now “seems to be giving way to concerns about negative free cash flow”, mounting debt and whether the investment will ever pay back.

The warning signs were there all year. In February 2026, CNBC reported that combined free cash flow for the four had already dropped from US$237 billion in 2024 to US$200 billion in 2025, with Alphabet’s long-term debt quadrupling to US$46.5 billion. A week later, UBS estimated hyperscaler capex would consume almost 100% of operating cash flow in 2026, against a 10-year average of 40%.

The cash flow maths stopped adding up

By July, the gap was impossible to ignore. Epoch AI estimated capex growing at roughly 70% a year against operating cash flow growth of 23%, putting spending on track to overtake cash generation around the third quarter. That is precisely where the market snapped. Amazon’s trailing 12-month free cash flow had already shrunk to just US$1.2 billion, from about US$26 billion a year earlier.

The scale has been extraordinary. In April 2026, The New York Times reported first-quarter capex of US$130.65 billion across the four, up 71% year-on-year and more than triple the inflation-adjusted cost of the Manhattan Project. In February, Bridgewater co-CIO Greg Jensen warned the boom had entered “a more dangerous phase”, increasingly reliant on outside capital.

Chips that age faster than the loans

The deeper problem is what the money buys. In April 2026, Research Affiliates partner Chris Brightman argued that “big spending is not the same as capital accumulation”. GPUs stay economically competitive for roughly three years, meaning around a third of the installed base must be replaced every year just to stand still. That is maintenance spending dressed up as growth capital, and it changes how any sensible investor should value it.

Proof still gets paid

This is not a blanket rejection of AI. In August, Microsoft shares jumped 15% after Azure growth accelerated to 43% and Copilot passed 30 million paid seats. Show returns, and the market rewards you.

But the broader revenue picture is thin. Sales in the current AI cycle have risen about 175%, against a seven-fold increase during the dot-com boom. Spending is running well ahead of what customers are actually paying for.

That is why Friday matters. Nobody needed a regulator, a select committee inquiry or a new oversight body. Capital markets priced the risk and punished the company asking for more money without more proof. That is the system working exactly as it should.

What this means for your AI budget

For New Zealand, the link is direct. Growth and balanced KiwiSaver funds carry heavy US tech exposure through global index holdings, so a week like this one lands in household balances. NZX growth stocks tend to trade in sympathy with Nasdaq sentiment too.

The more practical lesson sits with operators. The cheap AI era was subsidised. RNZ reported in August that Tesla, Uber, Meta and Amazon were already restricting employee use of AI agents, while OpenAI was spending around US$60 billion a year on compute against roughly US$13 billion of revenue. That gap has been a gift to downstream users, and it is being withdrawn. Providers under pressure to earn a return on US$700 billion will pass costs through.

Any NZ business signing off an “AI transformation” budget on enthusiasm rather than measurable productivity gains should expect higher tool and API costs, and should demand the same evidence Wall Street now demands of Amazon.

The next test comes with fourth-quarter results and 2027 guidance early next year. If the hyperscalers keep raising capex without matching revenue, expect the discount to deepen. If they can show the cash coming back, Microsoft-style, the boom survives on firmer ground. Either way, the era of being paid simply for spending is over, and that is a healthy correction for everyone.

Sources

• NZ Herald: Markets lashed by tech rout as earnings fan AI fears (2026-10-30)
• TechCrunch: Cracks are emerging in the AI spending boom (2026-10-30)
• CNBC: Tech AI spending approaches $700 billion in 2026, cash taking big hit (2026-02-06)
• CNBC: The Tech Download: Can hyperscalers justify their huge AI capex? (2026-02-13)
• Investment News NZ: Hoper-scalers: why AI cash-splash risks capital belly-flop (2026-04-12)
• NZ Herald: AI boom enters new phase as investors demand proof it is working (2026-08-04)
• RNZ: Companies must adapt to rising AI costs, tech commentator says (2026-08-09)