AWS Hits 39% Growth, Yet Capex Anxiety Grips Big Tech AI Spending

The question: All three big clouds accelerated this quarter, so why did Amazon and Alphabet sell off hard while Microsoft barely moved? Has the market's "more capex is better" view changed, and what does that mean for compute-chain names such as Marvell? | Prices as of 7:28 pm ET, Oct 29, 2026 (after hours).

Report date: Oct 30, 2026 | Sources: Amazon Q3 2026 earnings release (ir.aboutamazon.com), Alphabet Q3 2026 8-K (via Aya's Oct 29 report), Tencent Securities quotes, FMP consensus, X posts (grok-x), web search, Stratechery. Note: the SEC EDGAR and yfinance connectors were unavailable for this report, so Microsoft figures come from financial-news accounts on X and are labeled as such.

1. Same cloud acceleration, very different stock reactions

Start with Amazon. In the regular session on Oct 29, Amazon closed at $243.79, down 4.25%. That drop came before its earnings release and mainly reflected the previous night's Alphabet report. After the close, Amazon reported and fell another 6.85% after hours, about 10.8% below the prior day's close.

The report itself was not weak. AWS revenue was $46.0B, up 39%, its fastest growth in nearly five years. But Amazon's capex (spending on data centers and servers) reached $77.3B in the quarter, against expectations of about $57.5B, roughly a third more. Trailing-12-month free cash flow (operating cash flow minus capex) was –$42.4B.

Alphabet, which reported the day before, grew Google Cloud by 103%; GOOGL closed down 12.64% on Oct 29. Microsoft grew Azure by 48%; MSFT slipped only 0.16% the same day.

All three clouds beat expectations, yet the stock reactions ranged from flat to a plunge. The market is no longer asking whether AI demand is real. It is asking who pays with its own cash and how long the payback takes. That is also why GOOGL (heat 92), MRVL (86) and AMZN (80) took the top three spots on Aya's US heat ranking.

2. Background: cloud is a "build first, sell later" business

To follow this split, an ordinary investor first needs to understand how cash moves through a cloud business.

A cloud provider sells compute, and compute has to be built before it can be sold: land, power hookups, buildings and chips all come first. That spending is capex. Revenue arrives monthly and yearly only after the capacity goes live and customers start using it. Servers and buildings are not expensed all at once; their cost is spread over 5–6 years as depreciation. Contracts that customers have signed but that have not yet been recognized as revenue make up the backlog (Microsoft calls it commercial remaining performance obligation, or RPO).

So a cloud provider that is expanding fast shows three things at the same time: high cloud profits on the income statement, because most new assets are not yet depreciating; large outflows on the cash flow statement; and a growing backlog. All three giants are in this state this quarter, to different degrees:

Q3 side-by-side (US$ billions)

The row that best separates the three is free cash flow. Amazon burned about $34.9B in the quarter, Alphabet $9.1B, and Microsoft was still cash-positive. The stocks fell in roughly the same order.

3. Question one: can cloud acceleration cover capex and interest?

View: on the income statement, easily, and interest costs are nowhere near a problem. On the cash flow statement, not in the near term; the gap is being filled with debt. Whether the spending pays back depends on how fast the backlog turns into revenue.

Profits are strong. AWS operating income was $17.8B, up 56.1%, faster than its 39% revenue growth, and 63.3% of Amazon's total operating income. Amazon's overall operating margin of 13.9% was a record, per the release. Over the past five quarters AWS growth went 20%, 24%, 28%, 37%, 39%, four straight quarters of acceleration.

Both lines point up. Demand is not the problem; the problem is elsewhere.

Cash is deteriorating fast. Amazon's Q3 operating cash flow was $42.4B against capex of $77.3B, 1.82 times as much, and quarterly capex more than doubled (+120%) from a year earlier. TTM free cash flow fell from +$14.8B a year ago to –$42.4B:

The drop in the latest quarter (from –$7.6B to –$42.4B) was bigger than the previous four quarters combined. This is the direct cash flow effect of capex running far above expectations.

Interest is not the main issue. Q3 interest expense was $1.7B, up about 216% from $538M a year ago, but only about 6% of operating income. At Sep 30, long-term debt was $173.1B against $114.2B of cash and marketable securities, for net debt of about $58.9B. For a company earning what Amazon earns, that leverage is far from dangerous. The cost to watch is depreciation: $20.9B in the quarter, while most of the roughly $220B of 2026 capex has not started depreciating yet. AWS operating margin slipped from 39.4% in Q2 to 38.7%, and the Q4 operating income guidance midpoint of about $31B is roughly 6.3% below the $33.1B consensus. Both may be early signs of depreciation pressure, but one quarter of data cannot separate that from seasonality.

The payback case rests on the backlog. According to several media reports on the call, AWS backlog was $578B, up 16.5% from $496B in Q2 and about 3.1 times AWS's current annualized revenue of $184B. CEO Jassy said the "lion's share" of 2027 capacity is already reserved, projected AWS growth above 40% in both 2027 and 2028, and said 2027 capex will be meaningfully higher than 2026.

Reasoning (including this article's estimate): If Amazon's trailing operating cash flow keeps growing at last year's 29% pace, the next 12 months would bring about $217.1B, roughly equal to 2026 capex of about $220B. Since 2027 capex will be higher still, free cash flow will likely stay near zero or negative through 2027. This is a straight-line extrapolation, not a forecast. In other words, cloud acceleration already covers capex on the income statement, but on the cash flow statement that is at least a year away.

Counter-evidence and uncertainty: First, Amazon raised its 2026 capex plan from about $200B to about $220B in Q2, mainly because of higher memory prices, according to media reports. Part of the increase is price inflation, not more compute, and that money will not bring in matching revenue. Second, the multi-year, multi-gigawatt Trainium commitments reportedly come mostly from two AI labs, Anthropic and OpenAI. The more the backlog is concentrated in a few customers that themselves depend on outside funding, the more its quality depends on those customers' ability to raise money. That is this article's inference; the backlog's customer mix will only be visible in the 10-Q.

4. Question two: has the market's view of "more capex is better" changed?

View: yes, but not from "for" to "against." It has moved from "capex proves demand" to "capex depends on whose money it is and how fast it pays back." The market is pricing the three companies according to how their spending is funded.

Evidence 1: how the stocks reacted. All three clouds beat on growth, but capex surprised in different directions. Microsoft's quarterly capex of $39.4B came in below the roughly $46.6B estimate (the CFO said third-party capacity deliveries shifted to the second half of the fiscal year, and an accounting change also lowers reported capex). Its free cash flow stayed positive, and the stock was flat. Alphabet's free cash flow was negative and it said 2027 capex will increase significantly; the stock fell 12.64%. Amazon's capex beat estimates by 34.5%, and the stock fell 6.85% after hours. The chart below shows Oct 29 closing moves for the related names:

Note Nvidia, down just 0.8%. If the market were rejecting AI demand itself, the most direct seller of compute would not have held up this well.

Evidence 2: how valuations rank. On FMP consensus (all pre-earnings), Microsoft trades at about 26.6x fiscal 2027 EPS (fiscal year ending June 2027), Amazon at about 22.8x 2027 EPS (about 21.2x on the after-hours price), and Alphabet at about 18.6x. The steadier the free cash flow, the higher the multiple. The fiscal years do not line up exactly, so this is only a ranking.

Evidence 3: credit and rates point the same way. Bloomberg reported on X that the cost of insuring Alphabet's debt against default rose after the stock plunged. Microsoft launched its first bond sale since 2017 on Oct 29: at least $25B across 10 tranches with maturities of 2 to 40 years, with proceeds available for capex. Fidelity's Jurrien Timmer noted the SOFR curve still implies two more Fed hikes in the coming months, and that the P/E is 16% lower than a year ago. When rates are rising, borrowing to build AI costs more, and the market naturally tolerates negative free cash flow less.

This is not entirely new. Stratechery wrote on Feb 10, 2026 that it was more nervous about Amazon's capex than Alphabet's, so the market has been telling capex stories apart for some time. What is new is the scale: free cash flow at two giants turned negative at the same time, and both have already said they will spend more in 2027.

Counter-evidence: First, Amazon's drop cannot be pinned entirely on capex. The same report showed EPS of $1.89 against $1.97 expected, ad revenue of $19.6B about 6.1% below expectations (the company blamed an earlier Prime Day), and soft Q4 profit guidance. By size of the miss, the 34.5% capex overshoot stands out most, but the share of the stock move each item caused cannot be separated. Second, the sell side has not turned: per FMP, before earnings 84 analysts rated Amazon Buy, 9 Hold and 1 Sell. Third, on Aya's heat ranking, 72% of self-labeled GOOGL posts were bullish (only 18 posts), the opposite of the bearish tone among large accounts.

5. Question three: what does this mean for Marvell and the compute supply chain?

View: MRVL's 13.24% drop on Oct 29 was mainly about its own M&A reports, not hyperscaler capex. On their own, the capex numbers are good news for suppliers. The real risks sit one layer down: funding constraints and changes in what the money is spent on.

First, get cause and effect right. Multiple X posts, citing Bloomberg and Reuters, said Marvell is in advanced talks to buy optical-components maker Lumentum. Posts put the target's value at about $83B, and the market is worried about dilution from a large stock deal. No terms have been disclosed, and this article could not verify them. Lumentum itself rose 0.61%. According to X posts, MRVL started falling after hours on Oct 28 on this report, before Amazon's earnings. Reading MRVL's drop as a sign that AI compute demand has peaked therefore gets the causality wrong.

The capex figures themselves point to more orders. The three companies spent a combined $168.1B on capex in Q3. Amazon says it will spend meaningfully more in 2027, Alphabet says significantly more, and Microsoft's CFO says spending will speed up in the second half of the fiscal year. All three say demand exceeds supply. For companies selling chips, optical modules and networking gear, that means a growing order book.

Second-layer risks worth watching:

1. Where the money goes. Per @financialjuice, Microsoft's CFO said about 75% of this quarter's capex went to short-lived assets such as CPUs and GPUs, with memory and storage costs rising; Amazon also cited memory prices as the main reason for raising capex. This article's inference: with budgets set, the more that memory price increases absorb, the less incremental spending goes to other parts of the stack. Suppliers will not benefit evenly.
2. Custom chips. Anthropic and OpenAI have made multi-gigawatt commitments to Trainium, and Amazon is discussing offering Trainium outside AWS. Some X users list MRVL as a Trainium beneficiary, but this article could not verify its share.
3. Funding constraints. If the market keeps punishing negative free cash flow, the odds that the giants slow spending go up. For now, all three management teams are signaling the opposite, so this remains a possibility rather than a trend.

MRVL also trades at about 72.3x trailing earnings, which makes it more sensitive to event shocks. In Hong Kong, Zhongji Innolight fell 9.9% and YOFC fell 9.6%. X discussion cited both US AI-capex sentiment and weak Q3 results from China-listed CPO names, and this article cannot separate the two effects.

6. What the market is betting on

Putting the sections together, the market's current pricing comes down to one sentence: it believes in the demand but is unsure about the payback timeline, so it is charging the "spend first" discount to the companies with the tightest cash flow.

At the after-hours price of about $227, Amazon trades at about 21.2x 2027 EPS, with a market value of about $2.45T. Notably, pre-earnings consensus has Amazon's 2027 EPS ($10.69) below 2026 ($12.95), and the same is true for Alphabet ($15.26 in 2027 vs. $20.39 in 2026). There are two possible reasons: 2026 includes one-time items such as investment gains, or analysts expect depreciation to depress 2027 profits. FMP does not break this down, so this article cannot say how much each contributes. Either way, the market has already accepted that 2027 profit will not grow in step with revenue.

Is that expectation reasonable? Revenue visibility is quite high: AWS backlog is about 3.1 times annualized revenue, and Microsoft's commercial RPO is about 2.65 times Microsoft Cloud's annualized revenue. Costs are certain: higher 2027 capex, with depreciation rising after a lag. Three things are genuinely uncertain: how much of the backlog sits with a few AI labs; whether rates keep rising, making debt-funded building more expensive; and how much of the budget memory and other component price increases will absorb. Until those are answered, it is broadly consistent for the market to pay more for cash-positive Microsoft and less for Amazon, which has the largest cash gap. That said, Microsoft's commercial RPO of $677B came in below the roughly $705.6B estimate and roughly flat with about $678B last quarter. If that stall continues, Microsoft's premium will be tested too.

7. What to watch next

• Oct 30, 8:30 am ET: US Q3 Employment Cost Index (ECI). A hot print would strengthen rate-hike expectations and raise the cost of borrowing to build AI, which hurts the cash-negative names most.
• Next few days: the Amazon and Alphabet 10-Qs. For Amazon, watch lease and purchase commitments and how concentrated the backlog is by customer; for Alphabet, paid-click data. If Amazon's backlog is heavily concentrated in one or two AI labs, the backlog-quality issue from Section 3 becomes concrete.
• Pricing and spreads on Microsoft's $25B bond, and Alphabet's credit default swaps (CDS). Wider spreads would mean creditors also want a higher return for AI spending, a clearer sign of tighter funding conditions than stock moves.
• Whether Marvell and Lumentum formally announce a deal, and how it is paid for. If it is mostly stock, MRVL's drop is mainly a dilution story; if the deal falls apart, the stock will trade more on the compute chain's own orders.
• Nvidia's next quarterly report (historically late November; check the company's announcement for the date). This is the most direct test of whether the extra hyperscaler spending turns into upstream orders.
• Late Jan to early Feb 2027: Q4 reports from all three, plus 2027 capex guidance. The test is whether capex grows clearly faster than backlog and operating cash flow. If it does, worries will deepen. If backlog growth keeps pace and operating margins hold despite rising depreciation, the "invest now, earn it back later" case gets stronger.

8. Conclusion

This earnings season answered whether AI demand is real: AWS grew 39%, Azure 48% and Google Cloud 103%, with combined backlogs above $1.7T. It also raised a new question. Amazon's $77.3B of quarterly capex was 1.82 times its operating cash flow, TTM free cash flow is –$42.4B, and 2027 spending will be higher still. The market is not rejecting capex; it is repricing it by who can afford it and how long the payback takes. That is why cash-positive Microsoft barely fell and the companies with the biggest cash gaps fell the most. MRVL's same-day plunge was mostly about its M&A reports and should not be read as a sign that compute demand has peaked. What the compute chain really needs to watch is whether the hyperscalers' funding costs and spending mix change how fast they spend.

Appendix: Data and Sources

A1. Key financials (US$ billions unless noted)

AWS over five quarters: revenue $33.0B / $35.6B / $37.6B / $42.2B / $46.0B; growth 20% / 24% / 28% / 37% / 39%; operating margin 34.6% / 35.0% / 37.7% / 39.4% / 38.7%. Amazon TTM: operating cash flow $168.3B (+29%), net capex $210.7B (+82%), FCF –$42.4B. At Sep 30: cash $55.8B, marketable securities $58.4B, long-term debt $173.1B.

A2. Derived calculations

A3. Timeline (ET)

• Feb 10, 2026: Stratechery says it is more nervous about Amazon's capex than Alphabet's.
• Late Jul 2026: Amazon Q2: AWS +37%, backlog $496B, 2026 capex raised to ~$220B.
• Oct 28, after hours: Alphabet and Microsoft report; Marvell–Lumentum talks are reported and MRVL falls after hours.
• Oct 29: GOOGL −12.64%, MRVL −13.24%, AMZN −4.25%; Microsoft launches a bond sale of at least $25B; Amazon reports after the close and falls 6.85% after hours.
• Oct 30, 8:30 am: US Q3 ECI.
• Late Jan–early Feb 2027: Q4 results from all three and 2027 capex guidance (dates not announced).

A4. Aya heat data (US heat ranking, Oct 29, 2026 close)

• GOOGL: #1, heat 92 (discussion 91, trading 97, options 80), 5x relative volume; 72% bullish / 28% bearish (18 posts).
• MRVL: #2, heat 86 (discussion 84, trading 82, options 100), 2.3x relative volume; 47% / 53% (19 posts).
• AMZN: #3, heat 80 (discussion 89, trading 85, options 40), 2.1x relative volume, board after-hours −6.5%; 48% / 52% (25 posts).
• MSFT: #13, heat 57, 2x relative volume; 43% / 57% (21 posts).
• Note: for Amazon's after-hours move this article uses the 7:28 pm ET snapshot (−6.85%), which differs slightly from the board's snapshot (−6.5%).

A5. Market views

• @wallstengine (X, Oct 29): AWS $46.0B (est. $45.4B); capex $77.3B (est. $57.47B); TTM FCF −$42.4B; quoting Jassy: "projected revenue growth rates of over 40% both in 2027 and in 2028."
• @CNBC (X, Oct 29): "Amazon posts negative free cash flow of $42.4 billion, holds 2026 capex unchanged."
• @financialjuice (X, Oct 28): Microsoft's CFO said about 75% of capex went to short-lived assets such as CPUs and GPUs, and that spending will speed up in the second half of the fiscal year.
• @unusual_whales / @wallstengine (X, Oct 29): Microsoft bond sale of at least $25B, its first since 2017; 10 tranches, maturities of 2–40 years.
• @S_NewsRoomCOM (X, Oct 29): "The market is charging for the financing, not the lasers" (MRVL; unverified).
• @TimmerFidelity (X, Oct 29): P/E 16% lower than a year ago; SOFR curve implies two more hikes.
• @VulcanMK5 (X, Oct 29): "The $220B capex bill is a real concern but Amazon's earnings power is substantially stronger."

A6. Sources

1. Amazon Q3 2026 earnings release, Oct 29, 2026: https://ir.aboutamazon.com/news-release/news-release-details/2026/Amazon-com-Announces-Third-Quarter-Results/default.aspx
2. Alphabet Q3 2026 8-K (EX-99.1), Oct 28, 2026 (via Aya's Oct 29 report): https://www.sec.gov/Archives/edgar/data/1652044/000165204426000080/googexhibit991q32026.htm
3. Tencent Securities quotes (eastmoney.quote), 4:00 pm ET Oct 29, 2026
4. FMP consensus (AMZN / GOOGL / MSFT), retrieved Oct 29, 2026 ET
5. X posts (grok-x): @wallstengine, @financialjuice, @CNBC, @unusual_whales, @tomi_usstocks, @S_NewsRoomCOM, @AryaFintech and others, Oct 28–29, 2026
6. Web search: FourWeekMBA, cryptobriefing, huggingnews (AWS backlog of $578B and 2027 capex comments), Oct 29, 2026; TechTimes (Q2 backlog of $496B), Jul 31, 2026; mezha.net / The Star (Microsoft lease accounting change)
7. Aya Market Update (Oct 30, 2026): Bloomberg report on Alphabet CDS; ECI schedule
8. Stratechery, Feb 10, 2026, "Amazon earnings, capex concerns, commodity AI"

For informational and research discussion purposes only. Not investment advice.

🎙️ Podcast script (Chinese, ~20 min, 泓观阳略): read it here