General commentary for a general audience, not personalised advice. Share prices are US closes on Friday 23 Oct 2026 (Yahoo Finance): Microsoft $524.51, Meta $642.62, Alphabet $320.21, Amazon $248.00. Company figures come from each company's latest results release: Microsoft's fiscal fourth quarter to 30 Jun 2026 (released 29 Jul), Alphabet's second quarter (22 Jul), Amazon's second quarter (30 Jul) and Meta's third quarter (21 Oct). Capex guidance is as stated on those calls, reported by Reuters and CNBC. Anything marked as "our calculation" or "our assumption" is ours, not the company's. Alphabet and Microsoft report after the US close on Wednesday 28 Oct and Amazon on Thursday 29 Oct, so several of these numbers will move this week.
Here is the tension. Every chip, memory and power name on our board is, in the end, being paid by four companies. Between them, Microsoft, Alphabet, Amazon and Meta have guided to roughly $725B to $745B of capital expenditure this year. (Capex, or capital expenditure, is money spent on long-lived assets like servers and buildings rather than on running costs.) That is roughly what their businesses bring in as cash. So the question is no longer whether AI demand is real. It is which of these four is getting its money back, and which is spending on faith and hoping the revenue shows up later.
Last week the market started asking that question out loud. Meta reported negative free cash flow on 21 Oct and the stock fell almost 7% the next day (CNBC, 22 Oct). The Nasdaq had its worst session since June that same Thursday on worries about AI spending (CNBC, 25 Oct). This week the other three have to answer it.
The spend versus the cash
Free cash flow is the cash a business has left after paying for its capex. It is the money that can actually go back to shareholders. Here is how the four line up, using the most recent reported periods.
Sources: Microsoft FY26 Q4 release and call (29 Jul 2026); Alphabet Q2 2026 release (22 Jul 2026); Amazon Q2 2026 release (30 Jul 2026) and CNBC for quarterly capex; Meta Q3 2026 release (21 Oct 2026) and Q4 2025 release for the trailing figure. Our calculations: Microsoft free cash flow is fiscal 2026 operating cash flow of $182.9B less $115.9B of property and equipment additions; Meta trailing figures add Q4 2025 (operating cash flow $36.2B, free cash flow $14.1B) to the first nine months of 2026. The last row compares calendar 2026 guidance with trailing cash flow, so the periods do not match exactly; treat it as a rough gauge, not a precise ratio.
That last row is the whole note in one line. Microsoft plans to spend a bit less than it brings in. Alphabet and Meta plan to spend slightly more. Amazon plans to spend about a third more than its operating cash flow. Remember too that Meta's quarter is one quarter newer than the others, so it is the only one where we have already seen the third-quarter damage.
Is the money coming back?
Spending is only half the story. The other half is whether new revenue and profit are arriving at a pace that justifies it. This is where the four split apart.
Microsoft has the cleanest receipt. Azure, its cloud business, grew 43% in the June quarter, the fastest in four years, and management guided to about 45% for the September quarter (Microsoft call, 29 Jul). Azure passed $100B of annual revenue for the first time. Its order book of signed but not yet delivered contracts (commercial remaining performance obligation) rose 84% to $678B, and Microsoft said the sequential growth came from customers other than AI model developers. The company still produced $19.6B of free cash flow in the quarter at a 45% operating margin. Management also said customer demand still exceeds the capacity it can bring online.
Alphabet has the most dramatic receipt. Google Cloud revenue jumped 82% to $24.8B, and its operating margin went from 20.7% to 35.6% in a year. Cloud backlog reached $514B, with just over half expected to turn into revenue within 24 months. Just as important, the fear that AI chatbots would eat Search has not shown up yet: Search revenue grew 17%. Group operating margin rose two points to 34%. The catch is cash. Capex of $44.9B pushed free cash flow to minus $5.9B in the quarter, a negative quarter, and guidance was raised to $195B to $205B.
Amazon has a strong receipt but a big bill. AWS grew 37% to $42.2B, its fastest in 18 quarters, with operating income of $16.6B (a margin of about 39%). Backlog reached $496B and Andy Jassy said capacity will still fall short in 2026 and 2027. Advertising grew 26% to $19.8B. But Amazon is the one already in the red on a trailing basis: minus $7.6B of free cash flow over 12 months, against plus $18.2B a year earlier, and capex guidance went up from $200B to $220B, partly because memory chips got more expensive.
Meta is the hardest to read, because it has no cloud to rent out. Every dollar of AI capacity has to pay back through ads or through products that do not make money yet. And the ad business is doing a lot. Ad revenue rose 26%, adding about $13.0B year on year in a single quarter. That is more new dollars than Google Cloud (about $11.2B) or AWS (about $11.3B) added in their latest quarters (our calculation from the releases). Price per ad rose 14%, which is what better AI targeting looks like on an income statement. But costs rose faster. Operating margin was about 35% before the legal charge, down from 40%, free cash flow was minus $2.1B, and Meta said it will keep "investing aggressively" in 2027 without giving a number.
Depreciation: the bill that arrives later
Depreciation is how accounting spreads the cost of a server or building over its useful life. Capex hits cash flow today. Depreciation hits reported profits over the following years. That is why today's margins flatter all four: most of the 2026 spend has not yet reached the income statement.
You can already see the wave building. Over the latest year, depreciation grew faster than revenue at every one of them:
Sources: company cash flow statements in the releases listed above. Line items are not identical across companies, so compare the direction, not the decimals.
Two accounting choices matter here. Microsoft said in July it will spread the cost of its data center and office buildings over 25 years instead of 15, and that more future data center leases will be booked as operating leases rather than finance leases. Both lower reported capex and depreciation without changing the cash going out (Reuters and Microsoft call, 29 Jul). Microsoft also said about two thirds of its quarterly capex went on short-lived assets, mainly GPUs and CPUs. Servers are typically written off over about five or six years, not 25 (Amazon moved part of its server fleet from six years to five in 2025, per its filings). So the building change helps less than it sounds, and it is worth knowing it is there when you read Microsoft's margins on Wednesday.
What is priced in
A reverse DCF (discounted cash flow) runs the usual valuation backwards. Instead of forecasting cash flows and arriving at a price, it starts from today's price and asks what cash flow the market must be expecting. Our version is deliberately simple, and every input here is our assumption: a 9% required return, 3% growth forever after 2031, and free cash flow held flat at today's trailing level until then (zero for Amazon). We use enterprise value (market value of the shares plus debt minus cash) from Yahoo Finance at the 23 Oct close.
Our calculations. The 60% conversion assumption means capex eventually falls back to about 40% of operating cash flow. For reference, Microsoft converted about 53% in fiscal 2025 and about 37% in fiscal 2026, so 60% is a generous assumption that makes the bar look lower than it may be.
Read across the table and something odd appears. The market is asking Microsoft and Alphabet to grow cash flow at roughly 20% a year for five years and then hand most of it back. That is a high bar, but they are the two currently clearing it on the operating line. Amazon's bar is a little lower but it starts from a negative number. Meta's bar is the lowest by a wide margin. At about 18 times next year's expected earnings (Yahoo Finance consensus, 23 Oct), Meta is priced as if a good part of this spending never pays back. Microsoft trades at about 22 times, Alphabet about 21 and Amazon about 24.
That is the differentiated point. Ranking by who pays back best and ranking by what is cheapest give you nearly opposite orders. The market has already done the obvious sort. The question for a 2 to 5 year holder is whether the gaps are too wide.
The bear case, at full strength
The bear does not need AI to fail. It only needs the return on this particular wave of spending to come in lower than the market assumes.
Capex is now eating the whole cash pile. All four are spending roughly one year of operating cash flow. Two of them, Amazon and Meta, already show negative free cash flow, and Alphabet's last quarter was negative too. If 2027 budgets go up again, and Meta has already said they will, shareholder returns get squeezed. Meta bought back no stock in the first nine months of 2026, against $26.2B in the same period of 2025. It issued $24.9B of debt this year, its long-term debt rose to $80.9B from $58.7B at the end of 2025, and it announced a new non-voting class of shares it says can be used to raise capital. CNBC reported that Meta's five-year credit default swaps (a way of insuring its bonds against default) hit record highs after the results.
Depreciation will catch up with margins. The table above shows depreciation growing 31% to 44% while revenue grows 18% to 26%. That gap closes in one of two ways: revenue speeds up or margins come down. Useful-life choices can soften the timing but not the cost.
Some of the money goes round in a circle. The hyperscalers invest in AI labs, the labs buy compute from the hyperscalers, and the investments get marked up. In the June quarter Amazon booked $53.4B of non-operating income, mainly from its Anthropic stake. Alphabet booked a $98.0B net gain, mainly unrealised gains on equity securities. Microsoft booked a $3.2B gain on Anthropic. None of that is fake, but it is not cloud revenue either. If a large AI lab struggles to raise its next round, backlog, revenue and those investment gains could all wobble at the same time.
Memory inflation means each dollar buys less. Amazon raised capex partly because memory prices rose, and Microsoft mentioned higher component pricing in its capex number. Some of the higher spending is paying more for the same capacity rather than building more of it.
What would change our mind
Thesis-breakers: cloud growth slowing sharply while capex guidance still rises (Azure well below 40%, Google Cloud below 50%, AWS falling back toward 30%); backlog shrinking quarter on quarter; or a big AI lab customer cutting a commitment. Any of those would say the capacity is being built ahead of paying demand.
Expression changes only: a single negative free cash flow quarter at Alphabet or Microsoft caused by timing of chip deliveries; a margin dip from depreciation while revenue keeps accelerating; or Meta showing a clear new revenue line from its AI products (for example paid tiers or business messaging) that changes how fast it can pay back. Those would move the ranking, not break the story.
The ranking
We rank these by who is getting paid back best on the evidence so far: incremental revenue and profit per dollar of capex, plus how much of the spending they can fund themselves. That is not the same as which stock has the best expected return.
- Microsoft. Azure is accelerating into the low to mid 40s, it is the only one still clearly producing free cash flow quarter after quarter, margins are the highest of the four, and 2026 capex sits just under its cash generation. The asterisk is accounting: the building life change makes its reported numbers look a bit better than the cash.
- Alphabet. The best incremental story (Cloud up 82% with margins near 36%, Search still growing 17%), held back to second only because its spending now runs slightly ahead of its cash and the June quarter was free cash flow negative.
- Amazon. AWS is clearly paying back again at 37% growth and a margin near 39%, but the group is already cash negative over 12 months and has the largest budget relative to its cash flow.
- Meta. The ad engine is genuinely getting better with AI, but it has no outside customers renting its compute, free cash flow has turned negative, it is borrowing and creating a new share class to fund the plan, and 2027 will be bigger again. This is the one most clearly spending on faith. It is also the one where the price already expects the least.
What we are watching this week
In order of how much each tells us:
- Microsoft (Wed 28 Oct): Azure growth against the roughly 45% guide, and whether capex lands near the "above $50B" it guided for the September quarter. A beat on Azure with capex in line is the cleanest possible payback signal.
- Alphabet (Wed 28 Oct): free cash flow. After minus $5.9B in the June quarter, a second negative quarter with 2027 capex guided up would put Alphabet in the same conversation as Meta. Also watch the Cloud margin and the backlog.
- Amazon (Thu 29 Oct): trailing free cash flow and any first word on 2027 capex. AWS growth staying in the high 30s would support the spending; trailing free cash flow sliding further below zero would test patience.
- All three: depreciation growth versus revenue growth, and any change to useful-life assumptions.
- Meta (late January): the 2027 capex number it declined to give last week, and whether any AI product starts to show up as its own revenue line.
Bottom line
AI demand is real, and the receipts at Microsoft, Alphabet and Amazon prove the cloud businesses are being paid for the capacity they build. But all four are now spending roughly everything they earn, and depreciation will lag into margins for years. On payback we rank Microsoft, then Alphabet, then Amazon, then Meta. The risk to that view is that the market has already priced the order: Meta's bar is the lowest, so a single quarter of visible AI revenue there could matter more than another strong Azure print. This week's three reports will tell us whether the spending gap is closing or widening.