Big tech AI spending now outruns cash flow

Big tech is now spending more on AI than its businesses bring in.

In the September quarter of 2026, Microsoft, Alphabet, Amazon and Meta together spent about US$202 billion building AI capacity, mostly data centres and chips. Over the same quarter, their businesses generated about US$176 billion in operating cash flow.

Here’s the catch. Spending at this scale has to earn a return eventually, and it’s too early to know whether it will.

Australia is tied to the build in two ways: the materials data centres use, and the data centres being planned here. In October, Australian AI data centre company Firmus withdrew its planned ASX float.

Let’s look at what changed, why it matters for Australia, and what to watch next.

What changed

The four largest US technology companies are now spending more on building AI capacity than their businesses generate in cash. Their September quarter filings, released on 28 and 29 October, show combined capital spending (what the filings call purchases of property and equipment) of about US$202 billion, against combined operating cash flow, the cash the businesses themselves generate, of about US$176 billion. Three months earlier, in the June quarter, capital spending of about US$165.0 billion was still just below operating cash flow of about US$171.8 billion. Alphabet’s release shows capital spending of US$51.4 billion against US$42.3 billion of operating cash. Amazon’s shows US$77.3 billion against US$42.4 billion. Microsoft remains the exception, with US$39.4 billion of capital spending against US$58.0 billion of operating cash.

Meta’s September quarter results show capital spending of US$33.9 billion against operating cash flow of US$32.8 billion, leaving free cash flow, what is left after capital spending, at minus US$2.12 billion. In the same quarter, three of the four (Alphabet, Amazon and Meta) spent more on the build than their operations brought in.

Most of this money goes into compute. Microsoft said roughly two thirds of its June quarter capital spending was for short-lived assets, primarily processor chips (CPUs and GPUs). Alphabet said about 60% of its technical infrastructure spending went to servers and 40% to data centres and networking equipment. Amazon said the majority of its 2026 capital spending supports AI and AWS.

Analysts had been watching for this crossing. Epoch AI said in June that combined cash capital spending across the largest cloud providers was on track to overtake operating cash flow around the third quarter of 2026. Across the wider market, Gartner forecasts worldwide AI spending of US$2.7 trillion in 2026, a broad measure that includes hardware, software, services and devices. Goldman Sachs Research puts AI-related investment, a narrower measure, at about US$1 trillion this year. The spending is still rising: Alphabet expects about US$200 billion of capital spending in 2026.

Why it matters

Capital spending is normally paid for out of the cash a business generates. When the build costs more than operations bring in, the difference has to come from somewhere else: cash already in the bank, borrowing or new shares. On 26 October 2026, Meta announced about US$35 billion of share sales to support its continued investment in AI infrastructure. When a company issues new shares, each existing share represents a smaller slice of the business. At Amazon, free cash flow over the twelve months to September was an outflow of US$42.4 billion, which it attributes primarily to higher purchases of property and equipment.

Borrowing carries its own risk. The Reserve Bank of Australia said in its October Financial Stability Review that optimism around AI is boosting asset prices, and that the long-term debt financing behind AI investment is becoming more opaque and more vulnerable to profits falling short of expectations.

The build also needs power. The International Energy Agency says global data centre electricity use has grown by around 12% a year since 2017, more than four times faster than total electricity use. In 2025, the IEA reported data centre demand rose 17%, against 3% growth in overall electricity demand.

One place to measure the return on this spending is cloud revenue. In the September quarter, Google Cloud revenue rose 103% to US$30.7 billion, Amazon’s AWS segment sales rose 39% to US$46.0 billion, and Microsoft Cloud revenue rose 30% to US$63.9 billion. None of the four reports a separate line for AI revenue, so investors are left comparing cloud growth with the size of the capital spending.

Who absorbs it

Shareholders of the large technology companies carry the first effect. If spending keeps outrunning operating cash, more of the gap may be funded with debt or new shares, as Meta has shown.

Australia is linked to the build in two ways. The first is through materials. Geoscience Australia reports that in 2024 Australia was the world’s top producer of lithium and the third largest producer of rare earths, and ranks in the top five for world economic resources of copper, lithium and rare earths. Data centres use these inputs: lithium-ion batteries for backup power, copper for electrical networks and cooling, and rare earth magnets in hard drives. BHP expects data centre copper demand to grow from about 500,000 tonnes a year to about 3 million tonnes by 2050.

The second is through local building. CommBank says Australia has around six gigawatts of potential data centre capacity in its pipeline, and estimates the build-out could total around A$150 billion by 2030.

That pipeline depends on investors being willing to fund it. On 9 October, Firmus, an Nvidia-backed AI data centre company, withdrew its planned ASX float, its first public share sale. It had planned to sell shares at A$11 each, valuing it at about US$30.6 billion (about A$43.7 billion). Its board said the terms on which the offer could proceed would not appropriately reflect the strength of the company’s business. Reuters reported the float was set to be the second-largest new share sale in Australia’s history, and quoted fund manager Jun Bei Liu of Ten Cap calling it an important reality check for the AI investment boom, but not the beginning of the end of the AI trade.

For the ASX itself, direct exposure is smaller than overseas. CommSec notes the ASX has held relatively few of the companies that have driven the global AI rally.

What to watch

The next readings are the four companies’ December quarter results, which usually arrive from late January next year.

The two lines to compare in each cash flow statement are purchases of property and equipment and net cash from operating activities. In the September quarter the combined figure for the four moved above their combined operating cash flow, as Epoch AI had forecast, so more of the build now depends on cash reserves, borrowing or new shares. The question for the next results is whether the gap keeps widening or starts to narrow.

The other measure is cloud revenue growth. If cloud revenue grows faster than capital spending, the gap begins to close. The build could then continue, and demand for the materials data centres use could hold up. Microsoft’s contracted future revenue rose 73% to US$677 billion in the September quarter. Meta has no cloud business, so its test is whether AI keeps lifting advertising revenue, which the company says it already does. If the gap widens, markets may reassess how long the investment cycle can last. The International Monetary Fund outlined a downside scenario in October in which a steep drop in AI investment could lead to a sharp fall in asset prices, cutting global GDP by 1.1% by 2028. In that case, demand for those resources could weaken.

Frequently asked questions