Home / Learn / AI capital spending explained
Learn

The $500bn AI spending question

Five of the largest companies in the world reported strong results in the same week and moved in opposite directions. The dividing line was not how much they were spending — it was whether they could show what the spending was producing.

By Nathan Wickham-Hurd · Founder, Oak Growth · First-class Economics & Finance, MBA · Last reviewed August 2026

The short answer

The largest technology companies have committed to roughly half a trillion dollars of capital spending in a single year, and the market has stopped rewarding the spending itself. It now rewards evidence that the spending produces revenue. That single shift explains why five megacaps reported strong results in the same week and moved in opposite directions.

Context, August 2026. Across late July 2026 results: Microsoft guided to roughly $175bn of calendar-2026 capital spending, Alphabet to $195–205bn (raised from $180–190bn a week earlier) and Meta to $130–145bn — about $500–525bn between three companies. Amazon gave no forward guide but its trailing twelve-month figure was around $173bn. Microsoft rose about 8% and Amazon about 10%; Meta and Alphabet fell sharply and Apple slipped, in the same week. Market conditions change; the framework below is meant to outlast them.

The sorting rule the market applied

It wasn't the size of the spending. Every one of these companies raised its number. What separated them was whether they could attach revenue to it.

CompanyWhat it showedReaction
Microsoft“Azure demand is greater than supply”Rose ~8%
AmazonFirst $200bn quarter, sales up ~20%Rose ~10%
MetaCapex $31.1bn, free cash flow $784mFell sharply
AlphabetCapex guide raised againFell
AppleChina and Services both missedFell ~3–4%

Being capacity-constrained turned out to be the best thing a company could say. It means the spending is chasing demand that already exists rather than demand management hopes will appear.

Why this is a valuation question, not a technology one

Capital spending doesn't reduce profit immediately — it's depreciated over years. But it consumes cash today. A discounted cash flow values free cash flow: operating cash flow minus capital spending. So a company spending everything it earns has, for valuation purposes, produced very little for its owners this year regardless of how good the revenue line looks.

Free cash flow = operating cash flow − capital spending Meta Q2 2026: $31.9bn − $31.1bn = $784m

That's why the market can look at 28% revenue growth and sell the shares anyway. Both reactions are internally consistent; they're just measuring different things.

The three questions that decide whether this ends well

1. Does the revenue arrive before the depreciation does?

Data centres get written down over a defined life. If revenue ramps faster than depreciation, margins expand later. If it doesn't, earnings get hit for years after the cash has already gone. This is the crux, and nobody knows the answer yet.

2. Is the demand end-customer or circular?

A meaningful share of AI infrastructure demand comes from other AI companies, some with financial relationships to their suppliers. Spending that circulates between a small group of firms looks like growth until one of them stops. Work out how much of a company's demand comes from outside that circle.

3. Who is funding it?

Internally generated cash is one thing. Debt is another — and the market has already seen very large bond issues from companies in this space. Debt-funded capital spending raises the bar: the assets now have to clear an interest cost as well as their own depreciation.

What history suggests, without predicting

Large infrastructure buildouts have a recognisable pattern. Railways, fibre optics, cloud computing: the technology usually turns out to matter enormously, the capacity gets built faster than demand arrives, and the first wave of investors frequently does badly even though the thesis was right. Being correct about the technology and correct about the shares are separate achievements.

That's not a forecast that this ends the same way. Cloud computing is the counter-example — the buildout was enormous and the returns were real for the companies that led it. The honest position is that nobody can tell you which pattern this is, which is exactly why the margin of safety matters more than usual here.

What a value investor actually does

Value each company on the free cash flow you believe it generates in a normal year — after the buildout, with depreciation running through the accounts. Be explicit about when you assume capital spending plateaus, because that single input moves the answer more than any other. And check who's paying for it: a company funding this from operating cash has far more room to be wrong than one funding it with debt.

Then insist on a discount to your number. If the assumption required for today's price is that everything goes right, you're not investing in an infrastructure buildout — you're underwriting it.

See what the spending leaves behind

Oak Growth values roughly 1,000 companies on discounted cash flow across eight markets, with free cash flow, debt and returns on capital scored on every company.

Explore Oak Growth

Common questions

How much are big tech companies spending on AI?

Forward guidance for 2026 from three companies alone — Microsoft at roughly $175bn for the calendar year, Alphabet at $195–205bn and Meta at $130–145bn — totals around $500–525bn. Amazon gave no forward guide but its trailing twelve-month capital spending was around $173bn.

Why did some big tech stocks rise and others fall on similar results?

The market stopped rewarding the spending and started rewarding evidence it produces revenue. Microsoft rose after stating Azure demand exceeded supply, and Amazon rose on its first $200bn quarter. Meta and Alphabet fell because capital spending rose without an equivalent revenue link, and Meta's free cash flow collapsed to $784m.

Is AI capital spending a bubble?

Nobody can say reliably. Large infrastructure buildouts have historically tended to overbuild capacity relative to near-term demand, with early investors faring badly even when the technology mattered enormously — railways and fibre optics both followed that pattern. Cloud computing is the counter-example where returns were real. Which pattern this follows is unknowable in advance.

Why does capital spending hurt a share price if profits are fine?

Because valuations discount free cash flow rather than reported profit. Capital spending is depreciated over years in the accounts but consumes cash immediately, so a company spending nearly everything it earns produces very little for shareholders that year even with strong revenue growth.

Also see: Are AI stocks overvalued? → · Is AI a bubble? → · Free cash flow explained →