725 billion for AI. Who pays the bill?

Within nine days the four biggest buyers of artificial intelligence infrastructure in the world reported results. Alphabet on 22 July, Microsoft and Meta on 29 July, Apple and Amazon on 30 July. The number that stayed in everyone's mind is not a profit. It is a cost: roughly 725 billion dollars of capital spending during 2026, up from about 410 billion in 2025.

Where the money goes

Amazon leads with about 200 billion and raised its own 2026 guidance towards 220. Microsoft follows with about 190 and guided to 255 to 260 billion for its fiscal 2027. Alphabet lifted its ceiling to 175 to 205. Meta raised its range twice and now sits at 125 to 145.

These amounts are not marketing budgets and they are not hiring. They are buildings, chips, cooling and electricity. They are paid once and written off over years. For a sense of scale, the combined figure exceeds the annual output of most European economies, and four companies are spending it inside twelve months.

Why the spending does not show up in profit

This is where it gets interesting for anyone who reads financial statements. Capital spending does not pass through the profit of the year in full. It passes gradually, as depreciation. It does pass through cash in full, immediately.

The effect showed up in the quarter. Meta's free cash flow fell 91% and landed at 784 million dollars. Amazon's, on a trailing twelve month basis, turned negative at roughly minus 7.6 billion. These companies remain profitable. It is the cash left over after the investment that has almost disappeared.

That distinction is not an accounting nicety. Spare cash is what pays dividends, buybacks and acquisitions. When it dries up, a management team runs out of options, even while the profit line still looks healthy.

The question that dominated the earnings calls

It was not how much they will spend. It was whether operating cash flow can cover it without borrowing.

The answer decides between two very different outcomes. In the first, the infrastructure produces revenue over the coming years and the spending proves to be an investment. In the second, the same amount becomes depreciation with nothing behind it, which is a permanent drag on a decade of profits. Neither version is visible in a single quarter, and that is exactly where analysis stops and patience begins.

Where else that money lands

Spending on this scale does not stay inside four companies. It passes to chip makers, to construction firms, to manufacturers of cooling equipment and, above all, to electricity suppliers. A data centre is first of all a power customer with steady, large and predictable consumption.

That is why investment in artificial intelligence now shows up in sectors nobody calls technology. Anyone holding a broad index owns all of them already, without having to guess which supplier benefits most. It is one of the least spectacular but most useful properties of diversification: you do not need to find the winner, you only need to avoid sitting outside the chain.

Why this matters even if you hold none of these shares

If you own an ETF that tracks a global index, you own all of them. The United States accounts for roughly two thirds of an index such as the FTSE All-World, and the largest positions inside that share are exactly these companies.

This is not a flaw in the index. It is how market capitalisation weighting works: the bigger a company becomes, the larger the slice of the portfolio it occupies. The same mechanism that produced the returns of recent years is what makes today's concentration visible. The ATLAS guide explains how an index is built and why weighting is never a neutral choice, while the comparison of major indices shows how different a picture each one gives of the same market.

To see how much weight each name carries in your own fund, the Logifin ETF Hub has a profile for every ETF and the index behind it.

What to take away

• Capital spending hits cash first and profit later. That is how a profitable company can report free cash flow close to zero.
• The 725 billion figure is a plan, not money already spent. Guidance gets revised, and during 2026 it was revised upwards more than once.
• A global index does not protect you from concentration. It gives you what the market holds, at the weights the market has today.