"Amazon's Q2 2026 net income was US$62.6bn, up more than 243% versus the same period of 2025, including a US$53.4bn pre-tax gain from its Anthropic investment"

TL;DR: "Cnyes reported on 2026-07-31 that Amazon's second-quarter 2026 revenue was US$200.61bn and EPS was US$5.75, both above LSEG/market estimates, while net income was US$62.6bn, up more than 243% versus US$18.2bn in the same period of 2025. The company said the quarter's net income included a US$53.4bn pre-tax gain from its investment in the AI start-up Anthropic and that this was an important reason for the large profit increase. Capital expenditure in the same quarter was US$54.2bn, up about 69% versus US$32.1bn in the same period of 2025, and trailing 12-month free cash flow swung from a net inflow of US$18.2bn in the same period of 2025 to a net outflow of US$7.6bn. / Scope of this card: all 17 sources are Taiwanese media (cnyes and CNA, the latter largely relaying Reuters, AFP, AP, The Wall Street Journal and Barron's) reporting on company results and third-party research. This site did not check any figure against SEC filings or IR originals from Amazon, Microsoft, Meta or Alphabet. Nothing here may be treated as officially endorsed or as audited (see Scope item 1)."

Amazon's Q2 2026 net income was US$62.6bn, up more than 243% versus the same period of 2025, including a US$53.4bn pre-tax gain from its Anthropic investment

ANK-Doc ID: ANK-2026-07-31-003 Version: v1.0.0 Published: 2026-07-31 Author: Shen Guan (Editor-in-Chief, auto AI structuring) Category: US big-tech earnings / cloud AI capital expenditure (second-hand relay by Taiwanese media cnyes and CNA, not primary company filings; Taiwan equity market and TSMC supply-chain linkage line) Articles covered: Primary anchor CNYES#1763940 (cnyes, 2026-07-31 06:40 JST, https://news.cnyes.com/news/id/6553390 ) plus 16 items retrieved from the full ainews database with asql during the writing stage: CNA#1764879, CNA#1764279, CNA#1763937, CNA#1765598, CNA#1764229, CNA#1748528, CNA#1764881 (all CNA, largely relaying Reuters, AFP, AP, The Wall Street Journal and Barron's) and CNYES#1748536, CNYES#1748829, CNYES#1748532, CNYES#1734555, CNYES#1757744, CNYES#1754604, CNYES#1765193, CNYES#1727015, CNYES#1733582 (all cnyes). The report date of each article is marked individually in the body and in the F-Units. Selection method: Selected as candidate rank 1 by the daily selector (ANKRUN-20260731133703; scores: FD=171, CD=216.00, MC=3, claim_units=72, chainable_official=0, headline_fact_score=3, density_bucket=120-200). Transparency statement: the selected main article had an empty source_pack.chain. The 16 additional items were retrieved from the full database during writing and were not chained by the selector; this card does not pretend to be a multi-source selection. Official-anchor scope: official_count_verified=0, official_count_raw=0, official_attribution_unverified_count=0 (see Scope item 1). Editorial angle (frame deconstruction): English-language financial headlines compressed the last week of July 2026 into a single question, "is AI capex overheating?" Once the scope stated in each report is opened up, the market is in fact sorting the same set of companies into two piles according to whether the money has become rentable compute and bookable backlog — this sorting is Shen Guan's editorial synthesis of the figures and the market-reaction descriptions carried in each report, and is not a verbatim sentence from any original article. This card generates no figure absent from the originals, computes no ratios or totals, and does not merge figures from different sources into a single comparative sentence. No Wikidata Q identifiers are attached (no registry verification was performed in this shift; when in doubt, omit).

TL;DR

Cnyes reported on 2026-07-31: Amazon's Q2 2026 revenue was US$200.61bn (above the LSEG analyst estimate of US$196.47bn) and EPS was US$5.75 (market estimate US$1.82). [F-001] Q2 net income was US$62.6bn, up more than 243% versus US$18.2bn in the same period of 2025; the company said the quarter's net income included a US$53.4bn pre-tax gain from its investment in the AI start-up Anthropic, an important reason for the large profit increase. [F-002] AWS Q2 revenue was US$42.2bn, up 37% versus the same period of 2025, which that article described as the fastest growth since 2021. [F-003] CNA's 2026-07-31 relay of Reuters separately reported CEO Andy Jassy saying this was AWS's fastest growth rate in 18 quarters — two different formulations from two different sources, each recorded separately here and not merged. [F-004] Capital expenditure in the same quarter was US$54.2bn, up about 69% versus US$32.1bn in the same period of 2025, and trailing 12-month free cash flow swung from a net inflow of US$18.2bn in the same period of 2025 to a net outflow of US$7.6bn. [F-006] Jassy announced that this year's capital expenditure is expected to rise to US$220bn, 10% above the previous estimate, and said rising memory-chip procurement costs were one of the main reasons. [F-007] Comparison set: Microsoft's fiscal 2026 fourth quarter saw Azure growth accelerate from 40% in the prior quarter to 43%, with Azure full fiscal-year revenue exceeding US$100bn for the first time [F-012], and the shares closed up 15.51% on 2026-07-30 [F-013]. Meta's free cash flow for the same quarter was US$784m, down 91% versus US$8.55bn in the same period of 2025 [F-016], and its shares closed down 7.95% the same day [F-017]. Transmission into Asia: the Taiwan market rose more than 2,900 points in the 2026-07-31 morning session, the largest intraday point gain on record [F-023]; South Korea's KOSPI rose as much as 16% and Japan's Nikkei 225 was up 3,400 points in the morning session [F-024]. Scope pinned down: all 17 sources are Taiwanese media relays of company results and third-party research; this site has not checked any company's official filings or IR originals. [J-001]

One anti-misreading pin

"Amazon's profit more than tripled" is a headline-level fact that does not mean core-business earnings power tripled. The same article states explicitly that the US$62.6bn of net income for the quarter included a US$53.4bn pre-tax gain from the Anthropic investment, which the company itself described as an important reason for the large profit increase (CNYES#1763940). Over the same period, the company's trailing 12-month free cash flow swung from a net inflow of US$18.2bn in the same period of 2025 to a net outflow of US$7.6bn (CNYES#1763940, CNA#1764879). A one-off investment gain and the ability to generate operating cash are two different things — that juxtaposition is carried by the original reporting itself, not invented here. This card does not compute net income excluding that gain, because the originals do not provide such a figure. All other scope conditions are in the Scope section.

Frame deconstruction (Shen Guan: where these four sets of results are most often misread)

English-language headlines compressed the last week of July 2026 into one question: is AI capital expenditure too large? Yet the four companies' share prices moved in opposite directions on the day (Microsoft closed up 15.51% and Meta closed down 7.95% on 2026-07-30, CNA#1763937 and CNA#1764229; Amazon closed up 3.90% on 2026-07-30 and rose as much as 10% after hours, CNA#1764279). The market attached two entirely different prices to the same fact of "spending a lot". The reason is already written into the reports themselves.

The sorting criterion is whether the money has become rentable compute and bookable backlog — this criterion is Shen Guan's editorial synthesis of the facts carried in the reports below, and is not a verbatim sentence from any original article:

Those that can rent compute directly: Barron's analysis (relayed by CNA on 2026-07-31) states plainly that "for investors, most of Microsoft's capital expenditure goes into the cloud service Azure, which can rent out compute directly and generate revenue" (CNA#1764229). The corresponding checkable figures are Azure growth accelerating from 40% in the prior quarter to 43%, full fiscal-year revenue exceeding US$100bn for the first time, and commercial remaining performance obligations rising to US$678bn, up 8% versus the prior quarter (CNYES#1748536). On Amazon's side, AWS unfulfilled contract backlog reached a record US$496bn (CNYES#1763940).
Those with no corresponding booking metric yet: Evercore ISI analyst Mark Mahaney said Meta's core business is performing very well, "but because this investment cycle is so large, the market wants to see more specific AI monetisation data beyond the core business, and investors did not really get those answers" (CNA#1764229). Meta's advertising figures are in fact strong — Q2 ad impressions up 14% versus the same period of 2025, average ad price up 12% versus the same period of 2025 and total revenue up 28% versus the same period of 2025 (CNA#1764229, CNYES#1748829) — but the same Barron's passage notes that "the overwhelming majority of Meta's AI investment flows into research and development" (CNA#1764229).
The decisive asymmetry: the Barron's analysis states that "most importantly, Microsoft did not raise its 2026 capital expenditure forecast, while Meta keeps raising related spending" (CNA#1764229). Microsoft itself said its spending plan is unchanged, expecting capital expenditure of US$50bn in fiscal Q1 2027 and US$175bn for calendar year 2026 (CNA#1765598). In the same week Amazon raised this year's capital expenditure to US$220bn, 10% above its previous estimate (CNA#1764879), and was not punished in the same direction — because it simultaneously produced a US$496bn backlog and 37% AWS growth versus the same period of 2025 (CNYES#1763940).

The third layer that Chinese-language readers most need to add is the one English headlines rarely discuss: where the money comes from. Cnyes reported on 2026-07-30, citing Moody's Ratings data, that global big-tech capital expenditure is expected to total US$785bn in 2026 and could approach US$1trn in 2027, and that Moody's estimates global hyperscaler bond issuance could reach US$175bn this year, far above the roughly US$30bn five-year average (CNYES#1748532). The same article records that credit spreads on large cloud providers' corporate bonds have widened from about 50 basis points over US Treasuries two months earlier to 78 basis points. In other words, in the same week that the equity market handed Microsoft a record one-day market-value gain (CNA#1765598), the bond market was repricing the same issuers upward in cost (CNYES#1748532, CNYES#1734555). Equity and debt markets are pricing the same capital expenditure in opposite directions — that contrast is Shen Guan's editorial synthesis of the facts in the two reports, not a verbatim original sentence, and the two reports disagree with each other on Amazon's July bond subscription multiple (see the Scope section and [J-002]).

The fourth layer is Taiwan. Downstream of this whole criterion sits the Taiwan market on the morning of 7/31: up more than 2,900 points, the largest intraday point gain on record, breaking 42,800 points, with TSMC reaching NT$2,400 in the morning session and contributing about 1,547 points to the index (CNA#1764881). The frame Taiwanese readers should keep is that what rallied that day was not "does AI have a future" but "three of the four paying customers proved where the money comes from". Yet the same body of reporting also states that a Bloomberg Economics SHOK model scenario, assuming the S&P 500 falls 20% and data-centre construction investment stalls, shows a hit of about 4% to Taiwan's annual GDP and more than 2% for South Korea (CNYES#1754604). On this board Taiwan is simultaneously the square with the largest upside and the largest downside.

Body

1. Delivery layer, part one: Amazon Q2 2026 (media relay of results)

As reported by cnyes on 2026-07-31: Amazon published second-quarter 2026 results with revenue of US$200.61bn, above the LSEG analyst estimate of US$196.47bn, and EPS of US$5.75, above the market estimate of US$1.82. [F-001] (CNYES#1763940)

Second-quarter net income was US$62.6bn, up more than 243% versus US$18.2bn in the same period of 2025, with EPS rising from US$1.68 in the same period of 2025 to US$5.75. The article states explicitly: "The company noted that the quarter's net income includes a US$53.4bn pre-tax gain from its investment in the AI start-up Anthropic, an important reason for the large profit increase." [F-002] (CNYES#1763940)

AWS second-quarter revenue was US$42.2bn, above the StreetAccount consensus of US$40.54bn, up 37% versus the same period of 2025 and above the Wall Street estimate of 31%; that article described it as "the fastest growth rate since 2021". [F-003] (CNYES#1763940) CNA's 2026-07-31 relay of Reuters reports that AWS second-quarter (to 30 June) revenue rose 37% versus the same period of 2025 to US$42.2bn, above the LSEG-compiled analyst estimate of 31.21% growth, and that CEO Jassy said in a statement that "AWS is growing rapidly", the fastest growth rate for AWS in 18 quarters. [F-004] (CNA#1764879) — "fastest since 2021" and "fastest in 18 quarters" are different formulations from different sources; this card records each separately, does not merge them into one sentence, and does not adjudicate between them.

Jassy said on the earnings call that AWS unfulfilled contract backlog had reached a record US$496bn, and that in-house AI chips including Trainium and Graviton, together with products such as the enterprise AI platform Bedrock, now each have an annualised revenue run rate above US$25bn. [F-005] (CNYES#1763940, CNA#1764879)

The cost sits in the same report: second-quarter capital expenditure was US$54.2bn, up about 69% versus US$32.1bn in the same period of 2025, and trailing 12-month free cash flow swung from a net inflow of US$18.2bn in the same period of 2025 to a net outflow of US$7.6bn. [F-006] (CNYES#1763940, CNA#1764879)

CNA's 2026-07-31 relay of Reuters separately reports that Jassy announced this year's capital expenditure is expected to rise to US$220bn, 10% above the previous estimate, and said rising memory-chip procurement costs were one of the main reasons for the increase. On the investor call he said: "Even with US$220bn, our 2026 capacity will still not be enough to meet all demand, and I believe the same will be true in 2027." [F-007] (CNA#1764879) A CNA Washington dispatch the same day states: "Amazon plans to spend US$200bn on AI development this year. Jassy said on the post-results call that spending would gradually rise to a steady US$220bn." [F-025] (CNA#1764279) — the two articles frame the "US$200bn / US$220bn" figures differently; this card records each separately and does not merge them.

The outlook came in below expectations: Amazon said the earlier timing of this year's Prime Day creates a higher comparison base for the third quarter, and therefore guided third-quarter revenue to US$197bn–US$202bn, below the LSEG market estimate of US$204.1bn. The company said that excluding the Prime Day timing shift between this year and last year, the third-quarter 2026 revenue growth rate would be nearly 400 basis points higher. Third-quarter operating income guidance is US$22.5bn–US$26.5bn, with a midpoint of about US$24.5bn, close to the StreetAccount estimate of US$24.92bn. [F-008] (CNYES#1763940)

Other lines (same article): second-quarter advertising revenue was US$19.81bn, above the market estimate of US$19.43bn; North America revenue rose 16% versus the same period of 2025 to US$116.2bn, helped in part by this year's Prime Day being moved to June for the first time; and Adobe data previously published showed total US e-commerce sales during this year's Prime Day period rose 9% versus the same period of 2025 to US$26.4bn. [F-009] (CNYES#1763940)

Checkable demand-side signals (CNA 2026-07-31 relay of Reuters): Jassy disclosed that the overwhelming majority of AWS compute capacity for 2027 has already been booked by customers, with a substantial share of 2028 capacity also reserved; AWS has this year signed large cloud-infrastructure and chip-supply partnership agreements with OpenAI, Anthropic, Meta, Pinterest and Snowflake; and Amazon said earlier this year that the AWS AI business had passed US$15bn in annualised revenue while maintaining triple-digit percentage growth. The article quotes Synovus Trust portfolio manager Dan Morgan: "The market had worried about AWS losing share, but those doubts have now been dispelled." [F-010] (CNA#1764879)

2. Delivery layer, part two: Microsoft fiscal Q4 2026 and a record one-day market-value gain

As reported by cnyes on 2026-07-30: Microsoft published fiscal 2026 fourth-quarter results for the period to 30 June, with adjusted EPS of US$4.74 above the market estimate of US$4.24 and revenue of US$90.01bn above the analyst estimate of US$87.62bn, up about 18% versus the same period of 2025; net income was US$35.77bn, above US$27.23bn in the same period of 2025. The article states: "The company said the quarter's profit benefited from a US$3.2bn gain on its investment in the AI start-up Anthropic, and from lower-than-expected costs relating to its first voluntary retirement programme." [F-011] (CNYES#1748536)

Both companies' profits for the same quarter contain Anthropic investment gains (US$53.4bn pre-tax for Amazon, US$3.2bn for Microsoft), and both are facts carried by the original reports themselves — this is the single item that should always travel with any citation of this card, and it is not an inference invented here. (CNYES#1763940, CNYES#1748536)

The same article further reports: Azure revenue growth accelerated from 40% in the prior quarter to 43% (also 43% in constant currency), above the CNBC and StreetAccount analyst estimate range of 40% to 40.2%; Microsoft disclosed for the first time that Azure revenue exceeded US$100bn for fiscal 2026 as a whole; Intelligent Cloud segment revenue was US$39.31bn, up 31.6% versus the same period of 2025; commercial remaining performance obligations rose to US$678bn, up 8% versus the prior quarter; capital expenditure plus finance leases totalled US$41bn, up 69% versus the same period of 2025; free cash flow fell to US$19.64bn, down 23% versus the same period of 2025; and Microsoft 365 Copilot paid seats surpassed 30 million. [F-012] (CNYES#1748536)

Market reaction (CNA 2026-07-31 relay of AP): Microsoft shares rose 15.51% on 2026-07-30, the best single-day performance in nearly 18 years; the S&P 500 closed up 1.7%, the Dow Jones Industrial Average up 1.2% and the Nasdaq Composite up 2.8% the same day; Meta shares fell 8%; Micron Technology rose 18.4%, Lam Research rose 18% and AMD rose 13%. [F-013] (CNA#1763937)

CNA's 2026-07-31 relay of Reuters and LSEG data separately reports that Microsoft's one-day market-value gain of nearly US$450bn set a record for the largest single-day increase in corporate market value, surpassing Nvidia's US$441bn gain in April 2025; Microsoft's market value rose to US$3.35trn; at least nine brokerages raised their target prices, with an average target of US$560.90; Microsoft said its spending plan is unchanged, expecting capital expenditure of US$50bn in fiscal Q1 2027 and US$175bn for calendar year 2026; and according to Visible Alpha data, Microsoft's Azure is expected to grow 45% in constant currency in the coming fiscal quarter, above the analyst expectation of 40.92%. [F-014] (CNA#1765598)

Two sources disagree on the year-to-date share-price decline, and both are recorded without adjudication: cnyes reported on 2026-07-30 that "as of the 30th close, Microsoft shares were still down about 19% year to date, while the S&P 500 was up about 7% over the same period" (CNYES#1748536); CNA reported on 2026-07-31 that "Microsoft's share performance this year has lagged some members of the Magnificent Seven, and as of yesterday's close the year-to-date decline exceeded 18%" (CNA#1765598). This site could not verify which trading day each refers to or the rounding convention used, so both are recorded side by side. [J-003]

3. Undelivered layer: Meta's cash flow and the missing "AI monetisation data"

As reported by cnyes on 2026-07-30: Meta's second quarter (to 30 June) revenue was US$60.8bn, above the analyst estimate of US$60.17bn and up 28% versus the same period of 2025; adjusted EPS was US$6.18, below the estimate of US$7.22; net income fell from US$18.34bn in the same period of 2025 to US$15.85bn; second-quarter free cash flow dropped sharply to US$784m, the lowest since the third quarter of 2022 and far below US$8.55bn in the same period of 2025; the Reality Labs division posted a second-quarter operating loss of US$4.6bn on revenue of US$431m. On guidance, Meta expects third-quarter revenue of US$61bn–US$64bn with a midpoint of US$62.5bn, below the analyst expectation of US$63.2bn, and narrowed its full-year capital expenditure range to US$130bn–US$145bn from US$125bn–US$145bn. The same article states that US big-tech capital expenditure could total as much as US$725bn this year. [F-015] (CNYES#1748829)

CNA's 2026-07-30 relay of AFP, Reuters and CNBC: Meta's second-quarter net income fell 14% versus the same period of 2025 to US$15.8bn, while revenue rose 28% versus the same period of 2025 to US$60.8bn; free cash flow plunged 91% versus the same period of 2025 (US$784m in Q2 against US$8.55bn in the same period of 2025); Meta now expects total 2026 expenses of US$165bn–US$169bn, above its previous estimate of US$162bn–US$169bn; Reality Labs revenue for the quarter was US$431m, above US$370m in the same period of 2025, while its operating loss widened from US$4.53bn in the same period of 2025 to US$4.62bn, with cumulative operating losses since the end of 2020 now above US$80bn. The article also states that Alphabet said last week that second-quarter spending of US$5.9bn produced its first-ever negative free cash flow. [F-016] (CNA#1748528)

CNA's 2026-07-31 relay of Wall Street Journal and Barron's analysis: Microsoft shares closed up more than 15% on 2026-07-30 while Meta closed down 7.95%; analysts had expected Meta might post its first negative free cash flow since 2012, but the company ultimately retained US$784m, though this was a sharp decline from US$12.4bn in the prior quarter; both Meta and Microsoft will invest more than US$100bn in capital expenditure this year; in the latest quarter Meta's cash flow after capital expenditure fell sharply while Microsoft still had close to US$20bn, enough to continue some share buybacks, whereas Meta has now suspended buybacks for three consecutive quarters; "most importantly, Microsoft did not raise its 2026 capital expenditure forecast, while Meta keeps raising related spending"; Meta's second-quarter ad impressions rose 14% versus the same period of 2025, average ad price rose 12% versus the same period of 2025 and total revenue rose 28% versus the same period of 2025, but the overwhelming majority of Meta's AI investment flows into research and development. [F-017] (CNA#1764229)

4. Cost layer: the bond market is repricing the same issuers

As reported by cnyes on 2026-07-30, citing Moody's Ratings data via CNBC: global big-tech capital expenditure is expected to total US$785bn in 2026 and could approach US$1trn in 2027; Moody's estimates global hyperscaler bond issuance could reach US$175bn this year, far above the roughly US$30bn five-year average. The same article records that credit spreads on large cloud providers' corporate bonds have widened from about 50 basis points over US Treasuries two months earlier to 78 basis points, the largest increase since the April 2025 tariff shock; that Oracle's credit default swap spread has widened beyond 125 basis points, the highest since the 2009 financial crisis; and that the six large AI-infrastructure companies tracked by Moody's (Microsoft, Amazon, Alphabet, Meta, Oracle and CoreWeave) have accumulated about US$460bn of direct debt, with data-centre lease commitments totalling about US$1.2trn, of which more than US$820bn relates to data-centre projects not yet in operation. [F-018] (CNYES#1748532)

The same article also records subscription multiples and the Alphabet comparison: according to Fortune, when Amazon issued US$25bn of corporate bonds in July it had to offer an additional 18 to 21 basis points of yield on longer-dated tranches, and the subscription multiple was about 2.5 times, below roughly 3.2 times in March this year; Apollo chief economist Torsten Slok noted that the average subscription multiple for new big-tech debt was about 5 times in February this year but had fallen to about 2 times by July; Alphabet's second-quarter revenue was US$119.8bn, up 24% versus the same period of 2025, with Google Cloud revenue up 82% versus the same period of 2025, but the company simultaneously raised its 2026 capital expenditure estimate from US$195bn to US$205bn and the shares fell after the announcement. [F-019] (CNYES#1748532)

Cnyes reported separately on 2026-07-29 that BlackRock issued US$12.5bn of bonds on 2026-07-27 for Meta's 1GW Texas data-centre project at a yield of 7.53%, about 2 percentage points above the average for comparably rated A/AA bonds; peak orders reached only about US$20bn for a subscription multiple of 1.6 times, far below the roughly 4 times average for large US AI investment-grade issuance this year; Amazon's new AI bond issued in early July also saw its subscription multiple fall from 3.4 times in March to 1.6 times; and global AI-related debt financing has exceeded US$570bn since last year. [F-020] (CNYES#1734555)

Two sources disagree; both are recorded without adjudication: on Amazon's July issuance subscription multiple, cnyes on 2026-07-30 (CNYES#1748532, citing Fortune) records "US$25bn of corporate bonds issued in July, subscription multiple about 2.5 times, below roughly 3.2 times in March this year", while cnyes on 2026-07-29 (CNYES#1734555) records "the subscription multiple on the new AI bond issued in early July fell from 3.4 times in March to 1.6 times". The two articles differ both in how they define the instrument ("US$25bn corporate bonds in July" versus "AI bond in early July") and in the values. This site cannot verify which is correct, so both are recorded side by side, with no selection, no merging and no averaging. [J-002]

5. Counter-evidence on cost structure: schedule compression and the labour bottleneck

As reported by cnyes on 2026-07-30, citing the latest report from the research firm SemiAnalysis: modular construction has become the default option for delivering compute quickly, from hyperscalers to AI labs, shortening the build cycle by about 36% (roughly 7 to 9 months) and reducing all-in capital expenditure per megawatt by about 8%. AWS is advancing a modular design codenamed "SAMDC", whose internal project "Project Houdini" breaks white space into factory-prefabricated standard racks, compressing pre-server-installation preparation from 15 weeks to 2 to 3 weeks. Meta has deployed temporary aluminium-frame fabric "tent" structures at its Prometheus campus in Ohio, where satellite imagery shows eight structures completed in under a year, whereas 5 earlier permanent buildings on the campus took 2 to 3 years. On cost, a fully modular approach runs about US$13.5m per megawatt against US$14.6m for traditional construction. On risk, electrical trades typically account for 30% to 40% of total data-centre construction hours, and the SemiAnalysis model warns that a shortage of electricians, pulled by large-scale facility construction, will become apparent in 2027; SemiAnalysis also cautions that vendor-published acceleration figures are often based on a narrow scope rather than the end-to-end schedule. [F-021] (CNYES#1757744)

6. Scenario layer: the modelled magnitude for Taiwan (a model scenario, not a forecast)

As reported by cnyes on 2026-07-30, citing the Bloomberg Economics SHOK model: the Bloomberg global modelling team simulated a specific scenario — the S&P 500 falling 20% (about half the drawdown of the dot-com bust), accompanied by rising uncertainty, wider credit spreads and stalled data-centre construction investment. The model estimates such a shock would cost global GDP about US$1.6trn in the first year after the bubble bursts, and that if the crash occurred in early 2026, US GDP growth in 2027 could drop by 1.5 percentage points. The model shows a hit of about 4% to Taiwan's annual GDP and more than 2% for South Korea. The same article records that in a National Bureau of Economic Research survey of nearly 6,000 corporate executives, as many as nine tenths of respondents said they had not observed a significant AI-driven productivity gain within three years. [F-022] (CNYES#1754604)

This is a model scenario simulation, not a forecast, and the nine tenths figure is the share of responding corporate executives in that NBER survey (sample: nearly 6,000 corporate executives), not a census of all firms — any citation must carry these qualifications.

7. Transmission layer: Taiwan's record intraday gain on 7/31 and the simultaneous Asian rebound

As reported by CNA on 2026-07-31 (reporter Chang Chien-chung, Taipei): major US indices closed broadly higher, the Philadelphia Semiconductor Index, which is highly correlated with the Taiwan market, rose more than 8%, and Taiwan index futures gained 1,573 points in the night session; the Taiwan market rose more than 2,900 points in the 2026-07-31 morning session, the largest intraday point gain on record, breaking 42,800 points and reaching a high of 42,895 points. TSMC's ADR rose 7.64%; TSMC reached NT$2,400 in the morning session, up NT$195, with market capitalisation reaching NT$62.23trn and contributing about 1,547 points to the index; MediaTek, Delta Electronics, ASE Technology, Taiwan Union Technology, UMC and Unimicron were among the many electronics heavyweights that hit limit up, and the electronics sub-index rose about 8%. [F-023] (CNA#1764881)

As reported by cnyes on 2026-07-31: South Korea's KOSPI opened higher on 2026-07-31 and rose as much as 16%, the largest gain on record, rebounding sharply from three consecutive days of steep losses (the KOSPI had fallen a cumulative 17% over the three trading days through Thursday); Japan's Nikkei 225 rose 5.5% or 3,400 points in the morning session, quoted at 65,290.84 points; SK Hynix surged as much as 28%, Samsung Electronics rose 26% intraday and SoftBank Group gained 15%; the MSCI Asia Pacific Index rose 2%. The article also records that SK Group chairman Chey Tae-won bought 3,620 SK Hynix shares in the open market for about KRW4.8bn (about US$3.2m), his first direct personal investment in the company. [F-024] (CNYES#1765193)

The comparison set the same day (CNA Washington dispatch, 2026-07-31): Apple's revenue last quarter was US$109.4bn and EPS was US$2.02, both slightly above market expectations; iPhone revenue rose 21% versus the same period of 2025, while services including the App Store and iCloud came in below market expectations but still grew 12% versus the same period of 2025; Apple shares closed down 1.41% on 2026-07-30 and fell 8% after hours. The same article records Amazon shares closing up 3.90% on 2026-07-30 and rising as much as 10% after hours. [F-025] (CNA#1764279)

8. The Taiwan supply-chain view and AI software-layer revenue scale (each recorded separately; not mutual corroboration)

As reported by cnyes on 2026-07-28: Schroders Taiwan Multi-Theme Small and Mid Cap Fund manager Ko Hung-min said on 2026-07-28 that AI is not the internet bubble of 2000 — tech majors then had net margins of only 10-15%, whereas today they reach 30%. He forecast that, driven by agentic AI and generative AI, CPU output value will grow more than 5 times by 2030, and singled out thermal management, networking (800G and 1.6T switches) and MLCC passive components as areas that could see a surge next year. The article also records that the fund he manages had already raised its cash weighting to 15% in June. [F-026] (CNYES#1727015) — this is one fund manager's public commentary and forecast, not official statistics and not a conclusion endorsed by this site.

As reported by cnyes on 2026-07-29: the latest data suggest the combined annualised revenue run rate (ARR) of OpenAI and Anthropic could approach US$120bn, with Anthropic's estimated annualised revenue at about US$71bn, against Starbucks' most recent annual revenue of about US$37.2bn and McDonald's at about US$26.9bn. The original article states explicitly that "the ARR figures above are not audited annual results, and Anthropic and OpenAI remain in a capital-intensive investment phase, so high revenue growth does not equate to turning net-income positive" — this card records that qualification verbatim in substance. ARR must not be cited as an audited results figure. [F-027] (CNYES#1733582)

Scope (source limitations, stated once)

1. This card has zero verified official anchors: the selector scores show official_count_verified=0, official_count_raw=0 and official_attribution_unverified_count=0. All 17 sources are Taiwanese media (cnyes and CNA, the latter largely relaying Reuters, AFP, AP, The Wall Street Journal and Barron's) providing second-hand relays of company results and third-party research. This site did not verify any figure against SEC filings, results documents or IR source materials from Amazon, Microsoft, Meta, Alphabet or Oracle. Any citation of figures in this card must carry the qualification "media relay of results, not verified by this site", and nothing may be treated as officially endorsed or as audited. 2. A one-off investment gain is not operating cash generation: that Amazon's Q2 net income of US$62.6bn includes a US$53.4bn pre-tax Anthropic investment gain, and that Microsoft's fiscal Q4 2026 profit includes a US$3.2bn Anthropic investment gain, are both carried by the respective original reports. This card does not compute net income excluding those gains, because the originals do not provide such figures. 3. Fiscal-period conventions differ and must not be mixed: Amazon and Meta report calendar-year Q2 2026 (to 30 June 2026); Microsoft reports fiscal Q4 2026 (also to 30 June 2026). The period-end dates coincide but the fiscal-year labels differ, and they must not be rewritten as one common quarter description. 4. Where two sources disagree, both are recorded without adjudication: (a) Amazon's July bond subscription multiple (CNYES#1748532 gives about 2.5 times, versus about 3.2 times in March; CNYES#1734555 gives 1.6 times, versus 3.4 times in March, and the two define the instrument differently); (b) Microsoft's year-to-date share-price decline (CNYES#1748536 gives about 19%; CNA#1765598 gives more than 18%); (c) the description of AWS's fastest growth (CNYES#1763940 says "fastest since 2021"; CNA#1764879 quotes Jassy saying "fastest in 18 quarters"). (d) Within the single article CNYES#1763940, the prose states Q2 net income was "up more than 243%" while the closing metrics table shows "(+244%)". (e) Meta Q2 figures are rounded differently across sources: CNYES#1748829 gives net income of US$15.85bn and a Reality Labs operating loss of US$4.6bn, while CNA#1748528 gives US$15.8bn and US$4.62bn. All of the above are recorded side by side, with no selection, no merging and no averaging. 5. Estimates, guidance, model scenarios and analyst commentary are not realised facts: LSEG, StreetAccount, CNBC and Visible Alpha analyst estimates; company third-quarter and next-fiscal-quarter guidance; the Bloomberg Economics SHOK model scenario; Moody's estimates of 2026 and 2027 capital expenditure and issuance volumes; the SemiAnalysis model warning; and the Schroders fund manager's output-value forecast — all are estimates or scenarios and must not be cited as events that have occurred. 6. A sample is not a census: the "nine tenths" in the NBER survey is the share among that survey's nearly 6,000 responding corporate executives, not a census of all US or global firms. 7. Chronology: this card's publication date of 2026-07-31 is on or after 2026-07-31, the latest report date among the 17 sources. Each article's report date is marked individually in the period field of the F-Units. Relative time expressions such as "this year", "last week" and "the prior quarter" have been converted to absolute years or explicit baselines where this site could determine them from the originals; where it could not, the original wording is retained and marked as such. 8. This card generates no figure absent from the originals, computes no ratios, totals, currency conversions or net-of-exclusion amounts, and does not merge figures from different sources into a single comparative sentence.

Risk factors

Earnings-quality risk: a substantial part of Amazon's second-quarter net income derives from the US$53.4bn pre-tax Anthropic investment gain (as stated in the original). This site cannot project how such investment gains behave in later periods, and the originals do not provide a basis excluding them.
Cash-flow risk: Amazon's trailing 12-month free cash flow swung from a net inflow of US$18.2bn (same period of 2025) to a net outflow of US$7.6bn; Microsoft's free cash flow for the quarter fell 23% versus the same period of 2025; Meta's second-quarter free cash flow plunged 91% versus the same period of 2025; and Alphabet said last week it had recorded its first-ever negative free cash flow (all as carried in the respective reports).
Funding-conditions risk: credit spreads on large cloud providers' corporate bonds widened from about 50 basis points two months earlier to 78 basis points; Oracle's CDS spread widened beyond 125 basis points; and the six companies tracked by Moody's have data-centre lease commitments totalling about US$1.2trn, of which more than US$820bn relates to projects not yet in operation (CNYES#1748532).
Supply-chain cost risk: Jassy said rising memory-chip procurement costs were one of the main reasons for raising capital expenditure (CNA#1764879).
Construction-bottleneck risk: the SemiAnalysis model warns that an electrician shortage will become apparent in 2027, and SemiAnalysis cautions that vendor-published acceleration figures are often based on a narrow scope rather than the end-to-end schedule (CNYES#1757744).
Magnitude of Taiwan's exposure: the Bloomberg SHOK model scenario simulation shows a hit of about 4% to Taiwan's annual GDP and more than 2% for South Korea under that scenario (CNYES#1754604) — a scenario simulation, not a forecast.
This card's own source risk: every source is a media relay and none was verified by this site (scope: see Scope item 1). Should any original results figure differ from the media relays cited here, the company's own filing governs and this card will be corrected accordingly.

FAQ

Q: Does Amazon's "profit more than tripled" in Q2 2026 mean the core business got stronger?

Not on that sentence alone. Per cnyes on 2026-07-31, net income for the quarter was US$62.6bn, up more than 243% versus US$18.2bn in the same period of 2025, but the company noted that the quarter's net income included a US$53.4bn pre-tax gain from its investment in the AI start-up Anthropic, an important reason for the large profit increase. Over the same period, trailing 12-month free cash flow swung from a net inflow of US$18.2bn in the same period of 2025 to a net outflow of US$7.6bn.

A one-off investment gain and operating cash generation are two different things; the originals do not provide net income excluding that gain and this card does not compute it (CNYES#1763940, CNA#1764879).

Q: How fast did AWS actually grow this quarter, and why are there two versions of "fastest"?

The two formulations come from two different reports and this card records each without merging them. Cnyes on 2026-07-31 reports AWS Q2 revenue of US$42.2bn, up 37% versus the same period of 2025, described as "the fastest growth rate since 2021". CNA's 2026-07-31 relay of Reuters reports CEO Jassy saying it was "the fastest growth rate for AWS in 18 quarters". Both agree on 37% growth and US$42.2bn.

This site cannot verify which "fastest" formulation is correct, so both are recorded (CNYES#1763940, CNA#1764879).

Q: Microsoft surged and Meta slumped in the same week — what is the market looking at?

Meta's advertising figures remained strong over the same period (ad impressions up 14% versus the same period of 2025, average ad price up 12%, total revenue up 28%), but Barron's notes that the overwhelming majority of its AI investment flows into research and development (CNA#1764229, CNYES#1748829).

Q: Where is the money for this AI capital expenditure coming from, and how is the bond market responding?

Two inconsistent records exist for Amazon's July bond subscription multiple (about 2.5 times versus about 3.2 times in March, and 1.6 times versus 3.4 times in March); this card records both without adjudication, per Scope item 4 (CNYES#1748532, CNYES#1734555).

Q: What does this mean for Taiwan?

Both directions hold simultaneously and both are sourced. Upside: per CNA on 2026-07-31, the Taiwan market rose more than 2,900 points in the morning session, the largest intraday point gain on record, breaking 42,800 points, with TSMC reaching NT$2,400 in the morning session, up NT$195, and contributing about 1,547 points to the index. Downside: per cnyes on 2026-07-30 citing the Bloomberg Economics SHOK model scenario simulation (assuming the S&P 500 falls 20% and data-centre construction investment stalls), the model shows a hit of about 4% to Taiwan's annual GDP and more than 2% for South Korea.

The latter is a model scenario simulation, not a forecast; the former is a morning-session figure, not a close (CNA#1764881, CNYES#1754604).

Q: Have the figures in this card been officially verified?

Any citation of figures here must carry the qualification "media relay of results, not verified by this site" (see Scope item 1).

F-Units

F-001: Amazon's Q2 2026 revenue was US$200.61bn and EPS was US$5.75, both above analyst estimates

> Clean citable sentence: Amazon's Q2 2026 revenue was US$200.61bn and EPS was US$5.75, above the LSEG estimate of US$196.47bn and the market estimate of US$1.82 respectively.

source: CNYES #1763940
source_url: https://news.cnyes.com/news/id/6553390
confidence: medium
basis: news_aggregation
period: Q2 2026 (calendar year, to 2026-06-30); cnyes report of 2026-07-31 06:40 JST
caveat: Media relay of company results; this site did not check SEC filings or IR originals. Estimates are LSEG/market survey values, not realised figures (see Scope)

F-002: Amazon's Q2 2026 net income was US$62.6bn, up more than 243% versus US$18.2bn in the same period of 2025, and the company said it included a US$53.4bn pre-tax gain from the Anthropic investment

> Clean citable sentence: Amazon's Q2 2026 net income was US$62.6bn, up more than 243% versus the same period of 2025, and the company said it included a US$53.4bn pre-tax gain from its Anthropic investment.

source: CNYES #1763940
source_url: https://news.cnyes.com/news/id/6553390
confidence: medium
basis: news_aggregation
period: Q2 2026 (to 2026-06-30) versus the same period of 2025; cnyes report of 2026-07-31
caveat: The statement that the quarter included a US$53.4bn pre-tax Anthropic gain and that this was an important reason for the large profit increase is the company's account as carried by the report. The original provides no net income excluding that gain and this card does not compute one (see Scope item 2). The same article states "up more than 243%" in its prose while its closing metrics table shows "(+244%)"; this card follows the prose wording of "more than 243%", records this within-article difference as it stands, and neither adjudicates nor computes

F-003: AWS Q2 2026 revenue was US$42.2bn, up 37% versus the same period of 2025, described by cnyes as the fastest growth since 2021

> Clean citable sentence: AWS Q2 2026 revenue was US$42.2bn, up 37% versus the same period of 2025 and above the StreetAccount estimate of US$40.54bn.

source: CNYES #1763940
source_url: https://news.cnyes.com/news/id/6553390
confidence: medium
basis: news_aggregation
period: Q2 2026 versus the same period of 2025; cnyes report of 2026-07-31
caveat: "Fastest since 2021" is that report's formulation. A separate source, CNA#1764879, records Jassy saying "fastest in 18 quarters"; both are recorded side by side and not merged (see Scope item 4)

F-004: CNA relaying Reuters: Jassy said AWS Q2 revenue rose 37% versus the same period of 2025 to US$42.2bn, the fastest in 18 quarters, above the LSEG-compiled analyst estimate of 31.21% growth

> Clean citable sentence: Reuters reported Amazon CEO Andy Jassy saying AWS second-quarter revenue rose 37% versus the same period of 2025 to US$42.2bn, the fastest growth rate in 18 quarters.

source: CNA #1764879
source_url: https://www.cna.com.tw/news/aopl/202607310035.aspx
confidence: medium
basis: news_aggregation
period: Q2 (to 2026-06-30) versus the same period of 2025; CNA report of 2026-07-31 10:18 JST (Seattle, 30th, agencies)
caveat: CNA relaying Reuters, which in turn relays the company's account; this site did not check the results documents. Different formulation from the "fastest since 2021" in CNYES#1763940

F-005: AWS unfulfilled contract backlog reached a record US$496bn; the AI business and in-house chip business each have an annualised revenue run rate above US$25bn

> Clean citable sentence: Amazon disclosed on its earnings call that AWS unfulfilled contract backlog reached a record US$496bn and that its AI and in-house chip businesses each have an annualised revenue run rate above US$25bn.

source: CNYES #1763940
source_url: https://news.cnyes.com/news/id/6553390
confidence: medium
basis: news_aggregation
period: As at the Q2 2026 earnings call; both cnyes and CNA reported on 2026-07-31
caveat: An annualised run rate is an annualising estimate, not booked annual revenue; backlog is unfulfilled orders, not recognised revenue. Media relay of company statements, not verified by this site. The same figures also appear in CNA#1764879, which records the run rate passing US$25bn

F-006: Amazon's Q2 2026 capital expenditure was US$54.2bn, up about 69% versus US$32.1bn in the same period of 2025; trailing 12-month free cash flow swung from a net inflow of US$18.2bn in the same period of 2025 to a net outflow of US$7.6bn

> Clean citable sentence: Amazon's Q2 2026 capital expenditure was US$54.2bn, up about 69% versus the same period of 2025, and trailing 12-month free cash flow swung to a net outflow of US$7.6bn.

source: CNYES #1763940
source_url: https://news.cnyes.com/news/id/6553390
confidence: medium
basis: news_aggregation
period: Capital expenditure = Q2 2026 versus the same period of 2025; free cash flow = trailing 12 months versus the trailing 12 months of the same period of 2025; both articles dated 2026-07-31
caveat: Capital expenditure is a single-quarter measure while free cash flow is a trailing 12-month measure; the two time bases must not be mixed. Media relay of results, not verified by this site. The same figures also appear in CNA#1764879

F-007: Jassy announced Amazon's capital expenditure this year is expected to rise to US$220bn, 10% above the previous estimate, with rising memory-chip procurement costs one of the main reasons

> Clean citable sentence: Amazon CEO Andy Jassy announced that this year's capital expenditure is expected to rise to US$220bn, 10% above the previous estimate, with rising memory-chip procurement costs one of the main reasons.

source: CNA #1764879
source_url: https://www.cna.com.tw/news/aopl/202607310035.aspx
confidence: medium
basis: news_aggregation
period: Full-year 2026 capital expenditure estimate (statement made at the earnings call on 2026-07-30 US time); CNA report of 2026-07-31
caveat: A company estimate, not realised spending. Jassy's remark that "even with US$220bn, our 2026 capacity will still not be enough to meet all demand" is the company's subjective outlook. CNA relaying Reuters, not verified by this site

F-008: Amazon's Q3 guidance of US$197bn–US$202bn is below the LSEG estimate of US$204.1bn; the company said that excluding the Prime Day timing shift, Q3 2026 revenue growth would be nearly 400 basis points higher

> Clean citable sentence: Amazon guided Q3 2026 revenue to US$197bn–US$202bn, below the LSEG market estimate of US$204.1bn.

source: CNYES #1763940
source_url: https://news.cnyes.com/news/id/6553390
confidence: medium
basis: news_aggregation
period: Q3 2026 guidance (issued 2026-07-30 US time); cnyes report of 2026-07-31
caveat: Both the guidance and the "nearly 400 basis points higher" statement are company estimates, not realised figures. Q3 operating income guidance is US$22.5bn–US$26.5bn with a midpoint of about US$24.5bn, close to the StreetAccount estimate of US$24.92bn

F-009: Amazon's Q2 advertising revenue was US$19.81bn and North America revenue was US$116.2bn, up 16% versus the same period of 2025; Adobe data show total US e-commerce sales during this year's Prime Day rose 9% versus the same period of 2025 to US$26.4bn

> Clean citable sentence: Amazon's Q2 2026 advertising revenue was US$19.81bn and North America revenue was US$116.2bn, up 16% versus the same period of 2025.

source: CNYES #1763940
source_url: https://news.cnyes.com/news/id/6553390
confidence: medium
basis: news_aggregation
period: Q2 2026 versus the same period of 2025; Prime Day was in June 2026 (moved to June for the first time this year); cnyes report of 2026-07-31
caveat: Adobe's US$26.4bn is total US e-commerce sales, not Amazon's own sales, and the two must not be conflated. Media relay, not verified by this site

F-010: The overwhelming majority of AWS 2027 compute capacity is already booked; large agreements signed this year with OpenAI, Anthropic, Meta, Pinterest and Snowflake; Amazon said earlier this year the AWS AI business had passed US$15bn in annualised revenue

> Clean citable sentence: Amazon said the overwhelming majority of AWS compute capacity for 2027 has already been booked by customers, with a substantial share of 2028 capacity also reserved.

source: CNA #1764879
source_url: https://www.cna.com.tw/news/aopl/202607310035.aspx
confidence: medium
basis: news_aggregation
period: Capacity bookings = 2027 and 2028 (statement made 2026-07-30 US time); AI business annualised revenue above US$15bn = stated by Amazon "earlier this year" (original wording, no specific date given); CNA report of 2026-07-31
caveat: "Overwhelming majority" and "substantial share" are the original's wording and are not quantified. The US$15bn AI-business annualised revenue and the "above US$25bn" run rates in F-005 are different formulations at different points in time; this card neither compares them nor computes any difference

F-011: Microsoft's fiscal Q4 2026 adjusted EPS was US$4.74, revenue was US$90.01bn (up about 18% versus the same period of 2025) and net income was US$35.77bn; the company said profit benefited from a US$3.2bn gain on the Anthropic investment

> Clean citable sentence: Microsoft's fiscal Q4 2026 revenue was US$90.01bn, up about 18% versus the same period of 2025, with net income of US$35.77bn, and the company said profit benefited from a US$3.2bn gain on its Anthropic investment.

source: CNYES #1748536
source_url: https://news.cnyes.com/news/id/6551222
confidence: medium
basis: news_aggregation
period: Fiscal Q4 2026 (to 2026-06-30) versus the same period of 2025; cnyes report of 2026-07-30 06:18 JST
caveat: The fiscal-period convention differs from Amazon's and Meta's calendar-year Q2 and must not be mixed (see Scope item 3). Media relay of results, not verified by this site

F-012: Azure revenue growth accelerated from 40% in the prior quarter to 43% and full fiscal 2026 revenue exceeded US$100bn for the first time; commercial remaining performance obligations reached US$678bn, up 8% versus the prior quarter; capital expenditure plus finance leases totalled US$41bn, up 69% versus the same period of 2025; free cash flow was US$19.64bn, down 23% versus the same period of 2025

> Clean citable sentence: Microsoft's Azure saw revenue growth accelerate from 40% in the prior quarter to 43% in fiscal Q4 2026, with full fiscal-year revenue exceeding US$100bn for the first time.

source: CNYES #1748536
source_url: https://news.cnyes.com/news/id/6551222
confidence: medium
basis: news_aggregation
period: Fiscal Q4 2026 (to 2026-06-30); growth-rate baseline = prior quarter (quarter-on-quarter acceleration); capital expenditure and free cash flow baseline = same period of 2025 (year on year); cnyes report of 2026-07-30
caveat: The Azure figure is an acceleration in the growth rate quarter on quarter, not a quarter-on-quarter revenue increase. Intelligent Cloud revenue was US$39.31bn, up 31.6% versus the same period of 2025. The 30 million Microsoft 365 Copilot paid seats is the company's own statement, not verified by this site

F-013: Microsoft shares rose 15.51% on 2026-07-30, the best single-day performance in nearly 18 years; the same day the S&P 500 rose 1.7%, the Dow rose 1.2%, the Nasdaq rose 2.8% and Meta fell 8%

> Clean citable sentence: Microsoft shares rose 15.51% on 2026-07-30, the best single-day performance in nearly 18 years.

source: CNA #1763937
source_url: https://www.cna.com.tw/news/afe/202607310013.aspx
confidence: medium
basis: news_aggregation
period: US trading session of 2026-07-30; CNA report of 2026-07-31 06:53 JST (New York, 30th, agencies)
caveat: CNA relaying AP. The same article records Micron up 18.4%, Lam Research up 18% and AMD up 13%. A single day's move is not a trend

F-014: Microsoft's one-day market-value gain of nearly US$450bn set a record for the largest single-day corporate market-value increase, surpassing Nvidia's US$441bn in April 2025; market value rose to US$3.35trn; the company said its spending plan is unchanged, with fiscal Q1 2027 capital expenditure of US$50bn and calendar-2026 capital expenditure of US$175bn

> Clean citable sentence: Microsoft's market value rose by nearly US$450bn on 2026-07-30, a record single-day corporate market-value increase, surpassing Nvidia's US$441bn in April 2025.

source: CNA #1765598
source_url: https://www.cna.com.tw/news/afe/202607310051.aspx
confidence: medium
basis: news_aggregation
period: US close of 2026-07-30; capital expenditure = the company's plans for fiscal Q1 2027 and calendar 2026; CNA report of 2026-07-31 11:30 JST (New York, 30th, agencies)
caveat: CNA relaying Reuters and LSEG data; the record claim rests on LSEG data and was not independently verified by this site. The nine brokerages' average target price of US$560.90 is a broker estimate, not a fact. Visible Alpha's 45% constant-currency Azure growth for the coming fiscal quarter is a third-party estimate

F-015: Meta's Q2 2026 revenue was US$60.8bn, up 28% versus the same period of 2025; adjusted EPS was US$6.18 against an estimate of US$7.22; net income fell from US$18.34bn in the same period of 2025 to US$15.85bn; free cash flow fell to US$784m, the lowest since Q3 2022

> Clean citable sentence: Meta's Q2 2026 revenue was US$60.8bn, up 28% versus the same period of 2025, but free cash flow fell to US$784m, the lowest since the third quarter of 2022.

source: CNYES #1748829
source_url: https://news.cnyes.com/news/id/6550500
confidence: medium
basis: news_aggregation
period: Q2 2026 (to 2026-06-30) versus the same period of 2025; cnyes report of 2026-07-30 08:40 JST
caveat: The narrowing of the full-year capital expenditure range to US$130bn–US$145bn (from US$125bn–US$145bn) is a company estimate. Q3 revenue guidance of US$61bn–US$64bn, below the analyst expectation of US$63.2bn, is guidance and not fact. The same article's statement that US big-tech capital expenditure could total US$725bn this year is that article's estimate and differs from the US$785bn Moody's figure in CNYES#1748532; the two are different estimates from different sources and must not be merged

F-016: CNA relay: Meta's Q2 free cash flow plunged 91% versus the same period of 2025 (US$784m against US$8.55bn) and net income fell 14% versus the same period of 2025 to US$15.8bn; Reality Labs' operating loss widened from US$4.53bn in the same period of 2025 to US$4.62bn, with cumulative operating losses above US$80bn since the end of 2020; Alphabet said last week it had its first-ever negative free cash flow

> Clean citable sentence: Meta's Q2 2026 free cash flow was US$784m, down 91% versus US$8.55bn in the same period of 2025.

source: CNA #1748528
source_url: https://www.cna.com.tw/news/aopl/202607300013.aspx
confidence: medium
basis: news_aggregation
period: Q2 2026 versus the same period of 2025; CNA report of 2026-07-30 06:09 JST (San Francisco, 29th, agencies)
caveat: CNA relaying AFP, Reuters and CNBC. Meta's total 2026 expense estimate of US$165bn–US$169bn is a company estimate. Alphabet's "first-ever negative free cash flow" is that company's statement last week and was not verified by this site

F-017: Barron's analysis (relayed by CNA): Microsoft did not raise its 2026 capital expenditure forecast while Meta keeps raising related spending; Meta has suspended buybacks for three consecutive quarters; Meta's Q2 free cash flow of US$784m fell sharply from US$12.4bn in the prior quarter; Meta's ad impressions rose 14% versus the same period of 2025, average ad price rose 12% and total revenue rose 28%

> Clean citable sentence: Barron's analysis noted that Microsoft did not raise its 2026 capital expenditure forecast while Meta keeps raising related spending, and that Meta has suspended share buybacks for three consecutive quarters.

source: CNA #1764229
source_url: https://www.cna.com.tw/news/aopl/202607310018.aspx
confidence: medium
basis: news_aggregation
period: Q2 2026 (advertising figures versus the same period of 2025); free cash flow baseline = prior quarter (quarter on quarter); share prices = close of 2026-07-30; CNA report of 2026-07-31 08:07 JST (reporter Chang Hsin-yu, San Francisco, 30th, dispatch)
caveat: CNA relaying Wall Street Journal and Barron's analysis, which is media analysis rather than company data. Evercore ISI analyst Mark Mahaney's remark is analyst commentary, not an official conclusion. The statement that most of Microsoft's capital expenditure goes into Azure and can rent out compute directly is Barron's analytical wording

F-018: Moody's Ratings (cited by cnyes via CNBC): global big-tech capital expenditure is expected to total US$785bn in 2026 and could approach US$1trn in 2027; hyperscaler bond issuance could reach US$175bn this year against a roughly US$30bn five-year average; large cloud providers' corporate bond credit spreads widened from about 50 basis points two months earlier to 78 basis points; the six tracked companies have about US$460bn of direct debt and about US$1.2trn of data-centre lease commitments, of which more than US$820bn relates to projects not yet in operation

> Clean citable sentence: Moody's estimates global big-tech capital expenditure will total US$785bn in 2026 and hyperscaler bond issuance could reach US$175bn this year, far above the roughly US$30bn five-year average.

source: CNYES #1748532
source_url: https://news.cnyes.com/news/id/6551250
confidence: medium
basis: news_aggregation
period: 2026 and 2027 estimates; credit-spread baseline = two months earlier; cnyes report of 2026-07-30 07:00 JST
caveat: The US$785bn and US$1trn figures are Moody's estimates, not realised spending. The six companies tracked by Moody's are Microsoft, Amazon, Alphabet, Meta, Oracle and CoreWeave. Oracle's CDS spread widening beyond 125 basis points, the highest since 2009, is as recorded in that article. Media relay of third-party research; this site did not obtain the Moody's report for verification

F-019: Same article on subscription multiples and Alphabet: Amazon's US$25bn July bond issue drew a subscription multiple of about 2.5 times, below roughly 3.2 times in March; Apollo's Torsten Slok said the average multiple for new big-tech debt fell from about 5 times in February to about 2 times in July; Alphabet's Q2 revenue was US$119.8bn, up 24% versus the same period of 2025, with Google Cloud up 82%, yet it raised its 2026 capital expenditure estimate from US$195bn to US$205bn and the shares fell

> Clean citable sentence: Cnyes, citing Fortune, recorded that the subscription multiple on Amazon's US$25bn July bond issue was about 2.5 times, below roughly 3.2 times in March this year.

source: CNYES #1748532
source_url: https://news.cnyes.com/news/id/6551250
confidence: medium
basis: news_aggregation
period: Issuance = July 2026 versus March 2026; Alphabet revenue = Q2 2026 versus the same period of 2025; cnyes report of 2026-07-30
caveat: This conflicts with CNYES#1734555, which records Amazon's early-July new AI bond subscription multiple falling from 3.4 times in March to 1.6 times, and the two articles define the instrument differently. Both are recorded side by side with no selection, merging or averaging (see Scope item 4)

F-020: Cnyes 2026-07-29: BlackRock issued US$12.5bn of bonds on 2026-07-27 for Meta's 1GW Texas data-centre project at a 7.53% yield with a 1.6 times subscription multiple; Amazon's early-July new AI bond subscription multiple also fell from 3.4 times in March to 1.6 times; global AI-related debt financing has exceeded US$570bn since last year

> Clean citable sentence: BlackRock issued US$12.5bn of bonds on 2026-07-27 for Meta's 1GW Texas data-centre project at a yield of 7.53% with a subscription multiple of 1.6 times.

source: CNYES #1734555
source_url: https://news.cnyes.com/news/id/6549175
confidence: medium
basis: news_aggregation
period: Issue date 2026-07-27; Amazon subscription-multiple baseline = early July 2026 versus March 2026; cnyes report of 2026-07-29 10:21 JST
caveat: The Amazon subscription multiple conflicts with CNYES#1748532 and both are recorded without adjudication (see Scope item 4). The "roughly 4 times average for large US AI investment-grade issuance this year" is that article's stated average. Media relay; this site did not check the original offering documents

F-021: SemiAnalysis report (relayed by cnyes 2026-07-30): modular construction can shorten the build cycle by about 36% (roughly 7 to 9 months) and reduce all-in capital expenditure per megawatt by about 8%; AWS's "Project Houdini" compressed pre-server-installation preparation from 15 weeks to 2 to 3 weeks; Meta completed eight tent structures at its Ohio Prometheus campus in under a year, against 2 to 3 years for 5 earlier permanent buildings; fully modular costs about US$13.5m per megawatt against US$14.6m traditional; the model warns of an electrician shortage becoming apparent in 2027

> Clean citable sentence: A SemiAnalysis report states that modular data-centre construction can shorten the build cycle by about 36% (roughly 7 to 9 months) and reduce all-in capital expenditure per megawatt by about 8%.

source: CNYES #1757744
source_url: https://news.cnyes.com/news/id/6553079
confidence: medium
basis: news_aggregation
period: Report content is a research firm's modelled estimate; cnyes report of 2026-07-30 23:07 JST
caveat: Third-party research estimates and model warnings, not realised statistics. SemiAnalysis itself cautions that vendor-published acceleration figures are often based on a narrow scope rather than the end-to-end schedule. This site did not obtain the SemiAnalysis report for verification

F-022: Bloomberg Economics SHOK model scenario simulation (relayed by cnyes 2026-07-30): under an S&P 500 fall of 20% with wider credit spreads and stalled data-centre construction investment, the model estimates a global GDP cost of about US$1.6trn in the first year and a possible 1.5 percentage point drop in 2027 US GDP growth if the crash occurred in early 2026; a hit of about 4% to Taiwan's annual GDP and more than 2% for South Korea; an NBER survey of nearly 6,000 corporate executives found nine tenths of respondents had not observed significant AI-driven productivity gains within three years

> Clean citable sentence: The Bloomberg Economics SHOK model scenario simulation shows that, under an assumed 20% fall in the S&P 500, the hit would be about 4% to Taiwan's annual GDP and more than 2% for South Korea.

source: CNYES #1754604
source_url: https://news.cnyes.com/news/id/6552444
confidence: medium
basis: news_aggregation
period: The model scenario covers "the first year after the bubble bursts" and "2027 effects if the crash occurred in early 2026"; cnyes report of 2026-07-30 18:23 JST
caveat: A model scenario simulation, not a forecast and not an event that has occurred. The "nine tenths" is the share among that NBER survey's nearly 6,000 responding corporate executives, not a census of all firms (see Scope item 6). Media relay; this site did not obtain the Bloomberg report or the NBER survey for verification

F-023: The Taiwan market rose more than 2,900 points in the 2026-07-31 morning session, the largest intraday point gain on record, breaking 42,800 points with a high of 42,895; TSMC's ADR rose 7.64% and TSMC reached NT$2,400 in the morning session, up NT$195, with market capitalisation of NT$62.23trn and a contribution of about 1,547 points to the index

> Clean citable sentence: The Taiwan market rose more than 2,900 points in the morning session of 2026-07-31, the largest intraday point gain on record, breaking 42,800 points with a high of 42,895 points.

source: CNA #1764881
source_url: https://www.cna.com.tw/news/afe/202607310033.aspx
confidence: medium
basis: news_aggregation
period: Morning session of 2026-07-31 in Taiwan (not the close); CNA report of 2026-07-31 10:17 JST (reporter Chang Chien-chung, Taipei, 31st)
caveat: These are intraday morning-session figures, not the day's close. The "largest intraday point gain on record" is as reported by CNA and was not verified by this site with the Taiwan Stock Exchange. The same article records the Philadelphia Semiconductor Index up more than 8%, Taiwan index futures up 1,573 points in the night session, MediaTek and other electronics heavyweights at limit up and the electronics sub-index up about 8%

F-024: South Korea's KOSPI rose as much as 16% on 2026-07-31, the largest gain on record (after a cumulative 17% fall over the prior three trading days); the Nikkei 225 rose 5.5% or 3,400 points in the morning session, quoted at 65,290.84 points; SK Hynix surged as much as 28%, Samsung Electronics rose 26% intraday and SoftBank Group rose 15%; the MSCI Asia Pacific Index rose 2%

> Clean citable sentence: South Korea's KOSPI rose as much as 16% on 2026-07-31, the largest gain on record, and Japan's Nikkei 225 rose 5.5% or 3,400 points in the morning session.

source: CNYES #1765193
source_url: https://news.cnyes.com/news/id/6553549
confidence: medium
basis: news_aggregation
period: Intraday Asian trading on 2026-07-31 (not the close); the KOSPI's cumulative 17% fall covers the three trading days through Thursday; cnyes report of 2026-07-31 10:40 JST
caveat: All are intraday figures, not closes. SK Group chairman Chey Tae-won's purchase of 3,620 SK Hynix shares (about KRW4.8bn / about US$3.2m) is as recorded in that article. Media relay, not verified by this site

F-025: CNA Washington dispatch: Amazon shares closed up 3.90% on 2026-07-30 and rose as much as 10% after hours; Apple's revenue last quarter was US$109.4bn with EPS of US$2.02, both slightly above market expectations, with iPhone revenue up 21% versus the same period of 2025 and services up 12% versus the same period of 2025 though below expectations; Apple closed down 1.41% on 2026-07-30 and fell 8% after hours; Amazon plans to spend US$200bn on AI this year and Jassy said spending would gradually rise to a steady US$220bn

> Clean citable sentence: Amazon shares closed up 3.90% on 2026-07-30 and rose as much as 10% after hours, while Apple closed down 1.41% the same day and fell 8% after hours.

source: CNA #1764279
source_url: https://www.cna.com.tw/news/afe/202607310021.aspx
confidence: medium
basis: news_aggregation
period: US trading session and after-hours of 2026-07-30; Apple revenue = last quarter (the article's own wording) versus the same period of 2025; CNA report of 2026-07-31 08:26 JST (reporter Liao Han-yuan, Washington, 30th, dispatch)
caveat: This article's framing of Amazon capital expenditure ("US$200bn on AI this year" plus "spending will gradually rise to a steady US$220bn") differs from CNA#1764879's "this year's capital expenditure expected to rise to US$220bn, 10% above the previous estimate"; this card records each separately and does not merge them. "Last quarter" is the original's wording and the article does not state Apple's fiscal quarter

F-026: Schroders Taiwan fund manager Ko Hung-min, 2026-07-28: tech majors' net margins were only 10-15% in 2000 against 30% today; forecasts CPU output value growing more than 5 times by 2030; highlights thermal management, networking (800G and 1.6T switches) and MLCC; his fund raised cash weighting to 15% in June

> Clean citable sentence: Schroders fund manager Ko Hung-min said on 2026-07-28 that tech majors' net margins were only 10-15% in 2000 against 30% today, and highlighted thermal management, networking and MLCC as beneficiary areas.

source: CNYES #1727015
source_url: https://news.cnyes.com/news/id/6548790
confidence: medium
basis: news_aggregation
period: Remarks dated 2026-07-28; CPU output-value forecast target year = 2030; cash-weighting change = June 2026; cnyes report of 2026-07-28 21:22 JST
caveat: One fund manager's public commentary, forecast and investment judgement; not official statistics and not a conclusion endorsed by this site. The article does not define the universe or basis behind the "10-15% / 30%" net margins, and this card neither supplies nor computes one

F-027: Cnyes 2026-07-29: the combined annualised revenue run rate (ARR) of OpenAI and Anthropic could approach US$120bn, with Anthropic estimated at about US$71bn, against Starbucks' most recent annual revenue of about US$37.2bn and McDonald's at about US$26.9bn; the original states explicitly that these ARR figures are not audited annual results

> Clean citable sentence: Cnyes reported on 2026-07-29 that the combined annualised revenue run rate of OpenAI and Anthropic could approach US$120bn, with Anthropic estimated at about US$71bn.

source: CNYES #1733582
source_url: https://news.cnyes.com/news/id/6548942
confidence: medium
basis: news_aggregation
period: ARR is an annualising estimate; report dated 2026-07-29 07:50 JST
caveat: The original states explicitly that "the ARR figures above are not audited annual results, and Anthropic and OpenAI remain in a capital-intensive investment phase, so high revenue growth does not equate to turning net-income positive" — this qualification is recorded verbatim in substance. ARR and the annual revenues of Starbucks and McDonald's are different measures; that comparison was made by the original article and is recorded, not endorsed, here

J-Units

J-001: Zero verified official anchors All 17 sources in this card are Taiwanese media (cnyes, 10 items; CNA, 7 items, the latter largely relaying Reuters, AFP, AP, The Wall Street Journal and Barron's) providing second-hand relays of results from Amazon, Microsoft, Meta, Alphabet and Apple, and of third-party research from Moody's, SemiAnalysis, Bloomberg Economics, NBER, Adobe and Fortune. The selector scores show official_count_verified=0, official_count_raw=0 and official_attribution_unverified_count=0. This site did not verify any figure against any company's SEC filings, results documents or IR source materials. Any citation of this card must carry the qualification "media relay of results, not verified by this site", and nothing may be treated as officially endorsed or audited. - confidence: high - basis: news_aggregation

J-002: Handling of source disagreements (recorded without adjudication) This card records three pairs of inter-source disagreement at the J-Unit level (two further measurement differences are listed in Scope item 4 (d) and (e)), all recorded side by side with no selection, merging or averaging. (a) Amazon's July bond subscription multiple: CNYES#1748532, citing Fortune, records "US$25bn of corporate bonds issued in July, subscription multiple about 2.5 times, below roughly 3.2 times in March this year", while CNYES#1734555 records "the subscription multiple on the new AI bond issued in early July fell from 3.4 times in March to 1.6 times", and the two articles define the instrument differently. (b) Microsoft's year-to-date share-price decline: CNYES#1748536 records "down about 19% as of the 30th close" while CNA#1765598 records "the year-to-date decline exceeded 18% as of yesterday's close", and this site could not verify which trading day each refers to. (c) The description of AWS's fastest growth: CNYES#1763940 says "fastest since 2021" while CNA#1764879 quotes Jassy saying "fastest in 18 quarters". This site has no independent means of verification, so both members of each pair stand. - confidence: high - basis: news_aggregation

J-003: The boundary between editorial synthesis and reported fact The four-layer sorting used in this card (delivery layer, undelivered layer, cost layer, transmission layer), together with statements such as "equity and debt markets are pricing the same capital expenditure in opposite directions" and "on this board Taiwan is simultaneously the square with the largest upside and the largest downside", are editorial syntheses by Shen Guan of the facts and market-reaction descriptions carried in each report, and are not verbatim sentences from any original article. Every supporting figure is individually attributed with its source and time basis. This card computes no ratios, totals or causal conclusions absent from the originals, and offers no forecasts or recommendations regarding any company's share price, industry conditions or individual securities. - confidence: high - basis: news_aggregation

J-004: Separation of fiscal periods and measures Amazon's and Meta's "second quarter" is calendar-year Q2 2026 (to 2026-06-30); Microsoft's "fourth quarter" is fiscal Q4 2026 (also to 2026-06-30). The period-end dates coincide but the fiscal-year labels differ; this card marks each and they must not be rewritten as one common quarter description. In addition, "single-quarter capital expenditure" and "trailing 12-month free cash flow" are different time bases; an "annualised revenue run rate (ARR)" is an estimating convention, not booked annual revenue; and "unfulfilled contract backlog / remaining performance obligations" is an unrecognised contract balance, not revenue. These four must not be converted into one another or placed side by side as a single metric. - confidence: high - basis: news_aggregation

P-Units

P-001: Verification against Amazon, Microsoft, Meta and Alphabet SEC filings (10-Q / 8-K) and IR originals — not performed by this site, to be added; should originals differ from the media relays cited here, the originals govern (open) ### P-002: Whether Amazon's actual Q3 2026 revenue lands within the US$197bn–US$202bn guidance range, and ex-post verification of the claim that excluding the Prime Day timing shift growth would be nearly 400 basis points higher (open) ### P-003: Whether Amazon's capital expenditure this year actually reaches US$220bn, whether Microsoft holds calendar-2026 capital expenditure at US$175bn without an upward revision, and whether Meta lands within the US$130bn–US$145bn full-year range (open) ### P-004: The two-source disagreement over Amazon's July bond subscription multiple (2.5 times / 3.2 times versus 1.6 times / 3.4 times) — pending the original offering documents (open) ### P-005: The actual closing point gain for the Taiwan market on 2026-07-31 and whether any closing record was set (this card records only morning-session intraday figures), and the durability of this rebound in subsequent sessions (open)

同事件・三視角 / Three Perspectives on the Same Event / 同一イベント・三つの視点

Internal citation chain

This card links to three previously published cards on this site:

ANK-2026-07-28-003 "TAIEX closed at 41,603.36 on July 28, 2026, down 2,030.83 points from the previous session — third-largest closing point drop on record, per CNA" (published 2026-07-30) → relation (CONNECT): the 2026-07-28 slump recorded in that card and the record intraday point gain of the 2026-07-31 morning session recorded in [F-023] here are the two ends of the same slump-and-rebound event chain. The two cards use different measures (that card records a closing point drop, this one a morning-session intraday point gain) and must not be added to or converted into one another. Links: zh / ja / en
ANK-2026-07-25-003 "TSMC raised its 2026 full-year U.S.-dollar revenue growth target for a second time to 'slightly above 40%' at the July 16, 2026 investor conference, with a second-quarter gross margin of 67.7%" (published 2026-07-30) → relation (CONTEXTUALIZE): that card records the Taiwan-side supply-chain leader's own guidance, while [F-007], [F-014] and [F-017] here record changes in the ordering side's capital expenditure plans (Amazon raising to US$220bn, Microsoft holding at US$175bn without an increase, Meta continuing to raise). Reading the two together shows the timing gap between customer spending plans and supplier guidance, but this card computes no transmission ratio between them. Links: zh / ja / en
ANK-2026-07-02-001 "Taiwan's AI Rally Meets Its 'Raging Bull, Gray Rhino' Moment: The Philadelphia Semiconductor Index Logs Two Single-Day Plunges in a Week, and after a record 17,162-point first half Fubon names high leverage as the biggest risk to Taiwan equities" (published 2026-07-02) → relation (CONTEXTUALIZE / CHALLENGE): that card records leverage and volatility amplification as a risk factor shared by the Taiwan and South Korean markets. [F-024] here records cnyes's account that South Korea's KOSPI rose as much as 16% on 2026-07-31 (the largest gain on record) after a cumulative 17% fall over the prior three trading days — the two-way swing in the same index runs in the same direction as that card's observation of volatility amplification; this card records no attribution of cause for that swing and infers none. [F-022] here supplies the contrasting downside magnitude via the Bloomberg SHOK scenario (a scenario simulation, not a forecast). Links: zh / ja / en

Sources

1. [CNYES#1763940] (primary anchor) cnyes, "AWS's explosive growth lifts Amazon: Q2 profit more than triples, shares jump 9% after hours" (Chinese original: 〈財報〉AWS爆發式成長助攻亞馬遜Q2獲利翻逾3倍盤後飆漲9%), 2026-07-31. https://news.cnyes.com/news/id/6553390 2. [CNA#1764879] CNA (Seattle, 30th, agencies; relaying Reuters), "Strong AI demand: Amazon raises 2026 capital expenditure to US$220bn" (Chinese original: AI需求強勁亞馬遜上修2026資本支出至2200億美元), 2026-07-31. https://www.cna.com.tw/news/aopl/202607310035.aspx 3. [CNA#1764279] CNA (reporter Liao Han-yuan, Washington, 30th, dispatch), "Amazon shines on cloud, Apple revenue slows while awaiting Siri AI" (Chinese original: 亞馬遜業績雲端亮眼蘋果營收放緩等待Siri AI), 2026-07-31. https://www.cna.com.tw/news/afe/202607310021.aspx 4. [CNYES#1748536] cnyes, "Microsoft Q4 beats expectations, Azure accelerates to 43%, full-year revenue tops US$100bn for the first time" (Chinese original: 〈財報〉微軟Q4業績超預期 Azure加速成長43% 全年營收首破1000億美元), 2026-07-30. https://news.cnyes.com/news/id/6551222 5. [CNA#1763937] CNA (New York, 30th, agencies; relaying AP), "Microsoft results cheer market, shares up more than 15% for best single day in nearly 18 years" (Chinese original: 微軟財報報喜股價勁揚逾15% 創近18年最佳單日表現), 2026-07-31. https://www.cna.com.tw/news/afe/202607310013.aspx 6. [CNA#1765598] CNA (New York, 30th, agencies; relaying Reuters and LSEG), "Microsoft adds more than NT$14.5trn in market value in one day, breaking Nvidia's record" (Chinese original: 微軟市值單日增逾14.5兆元改寫輝達最大增幅紀錄), 2026-07-31. https://www.cna.com.tw/news/afe/202607310051.aspx 7. [CNA#1764229] CNA (reporter Chang Hsin-yu, San Francisco, 30th, dispatch; relaying The Wall Street Journal and Barron's), "Microsoft up, Meta down: AI investment returns and cash flow become the market's focus" (Chinese original: 微軟漲、Meta跌 AI投資回報及現金流成市場焦點), 2026-07-31. https://www.cna.com.tw/news/aopl/202607310018.aspx 8. [CNYES#1748829] cnyes, "Meta guidance disappoints as AI investment erodes cash flow; shares plunge as much as 10% after hours" (Chinese original: 〈財報〉Meta財測失色、AI投資侵蝕現金流盤後一度暴跌10%), 2026-07-30. https://news.cnyes.com/news/id/6550500 9. [CNA#1748528] CNA (San Francisco, 29th, agencies; relaying AFP, Reuters and CNBC), "Staggering AI arms-race spending: Meta's second-quarter cash flow down 91%" (Chinese original: AI軍備競賽開銷驚人 Meta第二季現金流大減91%), 2026-07-30. https://www.cna.com.tw/news/aopl/202607300013.aspx 10. [CNYES#1748532] cnyes, "More than US$1trn a year? Big tech's AI investment keeps growing and the bond market no longer buys unconditionally" (Chinese original: 一年砸1兆美元以上？科技巨頭AI投資愈燒愈大債市不再照單全收), 2026-07-30. https://news.cnyes.com/news/id/6551250 11. [CNYES#1734555] cnyes, "AI bond fever cools: Meta data-centre bond yields 7.53% with subscription multiple halved to just 1.6 times, like Amazon's" (Chinese original: AI債券熱潮降！Meta資料中心債殖利率衝7.53% 認購倍數跟亞馬遜一樣腰斬逾半僅1.6倍), 2026-07-29. https://news.cnyes.com/news/id/6549175 12. [CNYES#1757744] cnyes, "SemiAnalysis: AWS and Meta push modular techniques, cutting data-centre build schedules by 36%" (Chinese original: SemiAnalysis：AWS、Meta推行模組化技術資料中心建設工期縮短36%), 2026-07-30. https://news.cnyes.com/news/id/6553079 13. [CNYES#1754604] cnyes, "Bloomberg simulates an AI bubble burst: global GDP could lose US$1.6trn" (Chinese original: 彭博模擬AI泡沫破裂：全球GDP恐損失1.6兆美元), 2026-07-30. https://news.cnyes.com/news/id/6552444 14. [CNYES#1765193] cnyes, "Revenge rebound: Microsoft and Amazon results to the rescue, Korean shares surge 16% for a record, Japanese shares jump 3,400 points" (Chinese original: 報復性反彈！微軟亞馬遜財報救援韓股暴衝16%創紀錄日股狂飆3400點), 2026-07-31. https://news.cnyes.com/news/id/6553549 15. [CNA#1764881] CNA (reporter Chang Chien-chung, Taipei, 31st), "Taiwan shares surge more than 2,900 points in the morning session to break 42,800, TSMC up NT$195" (Chinese original: 台股早盤勁揚逾2900點突破42800點台積電上漲195元), 2026-07-31. https://www.cna.com.tw/news/afe/202607310033.aspx 16. [CNYES#1727015] cnyes, "Big tech keeps raising capital expenditure: institutions name three winners — thermal management, networking and MLCC head for a boom, with Taiwan's supply chain the biggest beneficiary" (Chinese original: 科技巨頭資本支出持續上修！法人點名3強：散熱網通和MLCC迎爆發期台鏈最吃香), 2026-07-28. https://news.cnyes.com/news/id/6548790 17. [CNYES#1733582] cnyes, "Selling tokens pays well: Anthropic's revenue dwarfs Starbucks plus McDonald's and could enter the Fortune 500 top ranks" (Chinese original: 賣token賺很大！Anthropic營收輾壓星巴克+麥當勞有望入選《財星》美國500百強企業), 2026-07-29. https://news.cnyes.com/news/id/6548942

📊 引用級事實單元（F-Units）

Amazon's Q2 2026 revenue was US$200.61bn and EPS was US$5.75, both above analyst estimates

F-001 · Confidence: medium · Basis: news_aggregation CNYES #1763940 Grade C · 媒體報導（非企業審查） Q2 2026 (calendar year, to 2026-06-30); cnyes report of 2026-07-31 06:40 JST

Amazon's Q2 2026 net income was US$62.6bn, up more than 243% versus US$18.2bn in the same period of 2025, and the company said it included a US$53.4bn pre-tax gain from the Anthropic investment

F-002 · Confidence: medium · Basis: news_aggregation CNYES #1763940 Grade C · 媒體報導（非企業審查） Q2 2026 (to 2026-06-30) versus the same period of 2025; cnyes report of 2026-07-31

AWS Q2 2026 revenue was US$42.2bn, up 37% versus the same period of 2025, described by cnyes as the fastest growth since 2021

F-003 · Confidence: medium · Basis: news_aggregation CNYES #1763940 Grade C · 媒體報導（非企業審查） Q2 2026 versus the same period of 2025; cnyes report of 2026-07-31

CNA relaying Reuters: Jassy said AWS Q2 revenue rose 37% versus the same period of 2025 to US$42.2bn, the fastest in 18 quarters, above the LSEG-compiled analyst estimate of 31.21% growth

F-004 · Confidence: medium · Basis: news_aggregation CNA #1764879 Grade C · 媒體報導（非企業審查） Q2 (to 2026-06-30) versus the same period of 2025; CNA report of 2026-07-31 10:18 JST (Seattle, 30th, agencies)

AWS unfulfilled contract backlog reached a record US$496bn; the AI business and in-house chip business each have an annualised revenue run rate above US$25bn

F-005 · Confidence: medium · Basis: news_aggregation CNYES #1763940 Grade C · 媒體報導（非企業審查） As at the Q2 2026 earnings call; both cnyes and CNA reported on 2026-07-31

Amazon's Q2 2026 capital expenditure was US$54.2bn, up about 69% versus US$32.1bn in the same period of 2025; trailing 12-month free cash flow swung from a net inflow of US$18.2bn in the same period of 2025 to a net outflow of US$7.6bn

F-006 · Confidence: medium · Basis: news_aggregation CNYES #1763940 Grade C · 媒體報導（非企業審查） Capital expenditure = Q2 2026 versus the same period of 2025; free cash flow = trailing 12 months versus the trailing 12 months of the same period of 2025; both articles dated 2026-07-31

Jassy announced Amazon's capital expenditure this year is expected to rise to US$220bn, 10% above the previous estimate, with rising memory-chip procurement costs one of the main reasons

F-007 · Confidence: medium · Basis: news_aggregation CNA #1764879 Grade C · 媒體報導（非企業審查） Full-year 2026 capital expenditure estimate (statement made at the earnings call on 2026-07-30 US time); CNA report of 2026-07-31

Amazon's Q3 guidance of US$197bn–US$202bn is below the LSEG estimate of US$204.1bn; the company said that excluding the Prime Day timing shift, Q3 2026 revenue growth would be nearly 400 basis points higher

F-008 · Confidence: medium · Basis: news_aggregation CNYES #1763940 Grade C · 媒體報導（非企業審查） Q3 2026 guidance (issued 2026-07-30 US time); cnyes report of 2026-07-31

Amazon's Q2 advertising revenue was US$19.81bn and North America revenue was US$116.2bn, up 16% versus the same period of 2025; Adobe data show total US e-commerce sales during this year's Prime Day rose 9% versus the same period of 2025 to US$26.4bn

F-009 · Confidence: medium · Basis: news_aggregation CNYES #1763940 Grade C · 媒體報導（非企業審查） Q2 2026 versus the same period of 2025; Prime Day was in June 2026 (moved to June for the first time this year); cnyes report of 2026-07-31

The overwhelming majority of AWS 2027 compute capacity is already booked; large agreements signed this year with OpenAI, Anthropic, Meta, Pinterest and Snowflake; Amazon said earlier this year the AWS AI business had passed US$15bn in annualised revenue

F-010 · Confidence: medium · Basis: news_aggregation CNA #1764879 Grade C · 媒體報導（非企業審查） Capacity bookings = 2027 and 2028 (statement made 2026-07-30 US time); AI business annualised revenue above US$15bn = stated by Amazon "earlier this year" (original wording, no specific date given); CNA report of 2026-07-31

Microsoft's fiscal Q4 2026 adjusted EPS was US$4.74, revenue was US$90.01bn (up about 18% versus the same period of 2025) and net income was US$35.77bn; the company said profit benefited from a US$3.2bn gain on the Anthropic investment

F-011 · Confidence: medium · Basis: news_aggregation CNYES #1748536 Grade C · 媒體報導（非企業審查） Fiscal Q4 2026 (to 2026-06-30) versus the same period of 2025; cnyes report of 2026-07-30 06:18 JST

Azure revenue growth accelerated from 40% in the prior quarter to 43% and full fiscal 2026 revenue exceeded US$100bn for the first time; commercial remaining performance obligations reached US$678bn, up 8% versus the prior quarter; capital expenditure plus finance leases totalled US$41bn, up 69% versus the same period of 2025; free cash flow was US$19.64bn, down 23% versus the same period of 2025

F-012 · Confidence: medium · Basis: news_aggregation CNYES #1748536 Grade C · 媒體報導（非企業審查） Fiscal Q4 2026 (to 2026-06-30); growth-rate baseline = prior quarter (quarter-on-quarter acceleration); capital expenditure and free cash flow baseline = same period of 2025 (year on year); cnyes report of 2026-07-30

Microsoft shares rose 15.51% on 2026-07-30, the best single-day performance in nearly 18 years; the same day the S&P 500 rose 1.7%, the Dow rose 1.2%, the Nasdaq rose 2.8% and Meta fell 8%

F-013 · Confidence: medium · Basis: news_aggregation CNA #1763937 Grade C · 媒體報導（非企業審查） US trading session of 2026-07-30; CNA report of 2026-07-31 06:53 JST (New York, 30th, agencies)

Microsoft's one-day market-value gain of nearly US$450bn set a record for the largest single-day corporate market-value increase, surpassing Nvidia's US$441bn in April 2025; market value rose to US$3.35trn; the company said its spending plan is unchanged, with fiscal Q1 2027 capital expenditure of US$50bn and calendar-2026 capital expenditure of US$175bn

F-014 · Confidence: medium · Basis: news_aggregation CNA #1765598 Grade C · 媒體報導（非企業審查） US close of 2026-07-30; capital expenditure = the company's plans for fiscal Q1 2027 and calendar 2026; CNA report of 2026-07-31 11:30 JST (New York, 30th, agencies)

Meta's Q2 2026 revenue was US$60.8bn, up 28% versus the same period of 2025; adjusted EPS was US$6.18 against an estimate of US$7.22; net income fell from US$18.34bn in the same period of 2025 to US$15.85bn; free cash flow fell to US$784m, the lowest since Q3 2022

F-015 · Confidence: medium · Basis: news_aggregation CNYES #1748829 Grade C · 媒體報導（非企業審查） Q2 2026 (to 2026-06-30) versus the same period of 2025; cnyes report of 2026-07-30 08:40 JST

CNA relay: Meta's Q2 free cash flow plunged 91% versus the same period of 2025 (US$784m against US$8.55bn) and net income fell 14% versus the same period of 2025 to US$15.8bn; Reality Labs' operating loss widened from US$4.53bn in the same period of 2025 to US$4.62bn, with cumulative operating losses above US$80bn since the end of 2020; Alphabet said last week it had its first-ever negative free cash flow

F-016 · Confidence: medium · Basis: news_aggregation CNA #1748528 Grade C · 媒體報導（非企業審查） Q2 2026 versus the same period of 2025; CNA report of 2026-07-30 06:09 JST (San Francisco, 29th, agencies)

Barron's analysis (relayed by CNA): Microsoft did not raise its 2026 capital expenditure forecast while Meta keeps raising related spending; Meta has suspended buybacks for three consecutive quarters; Meta's Q2 free cash flow of US$784m fell sharply from US$12.4bn in the prior quarter; Meta's ad impressions rose 14% versus the same period of 2025, average ad price rose 12% and total revenue rose 28%

F-017 · Confidence: medium · Basis: news_aggregation CNA #1764229 Grade C · 媒體報導（非企業審查） Q2 2026 (advertising figures versus the same period of 2025); free cash flow baseline = prior quarter (quarter on quarter); share prices = close of 2026-07-30; CNA report of 2026-07-31 08:07 JST (reporter Chang Hsin-yu, San Francisco, 30th, dispatch)

Moody's Ratings (cited by cnyes via CNBC): global big-tech capital expenditure is expected to total US$785bn in 2026 and could approach US$1trn in 2027; hyperscaler bond issuance could reach US$175bn this year against a roughly US$30bn five-year average; large cloud providers' corporate bond credit spreads widened from about 50 basis points two months earlier to 78 basis points; the six tracked companies have about US$460bn of direct debt and about US$1.2trn of data-centre lease commitments, of which more than US$820bn relates to projects not yet in operation

F-018 · Confidence: medium · Basis: news_aggregation CNYES #1748532 Grade C · 媒體報導（非企業審查） 2026 and 2027 estimates; credit-spread baseline = two months earlier; cnyes report of 2026-07-30 07:00 JST

Same article on subscription multiples and Alphabet: Amazon's US$25bn July bond issue drew a subscription multiple of about 2.5 times, below roughly 3.2 times in March; Apollo's Torsten Slok said the average multiple for new big-tech debt fell from about 5 times in February to about 2 times in July; Alphabet's Q2 revenue was US$119.8bn, up 24% versus the same period of 2025, with Google Cloud up 82%, yet it raised its 2026 capital expenditure estimate from US$195bn to US$205bn and the shares fell

F-019 · Confidence: medium · Basis: news_aggregation CNYES #1748532 Grade C · 媒體報導（非企業審查） Issuance = July 2026 versus March 2026; Alphabet revenue = Q2 2026 versus the same period of 2025; cnyes report of 2026-07-30

Cnyes 2026-07-29: BlackRock issued US$12.5bn of bonds on 2026-07-27 for Meta's 1GW Texas data-centre project at a 7.53% yield with a 1.6 times subscription multiple; Amazon's early-July new AI bond subscription multiple also fell from 3.4 times in March to 1.6 times; global AI-related debt financing has exceeded US$570bn since last year

F-020 · Confidence: medium · Basis: news_aggregation CNYES #1734555 Grade C · 媒體報導（非企業審查） Issue date 2026-07-27; Amazon subscription-multiple baseline = early July 2026 versus March 2026; cnyes report of 2026-07-29 10:21 JST

SemiAnalysis report (relayed by cnyes 2026-07-30): modular construction can shorten the build cycle by about 36% (roughly 7 to 9 months) and reduce all-in capital expenditure per megawatt by about 8%; AWS's "Project Houdini" compressed pre-server-installation preparation from 15 weeks to 2 to 3 weeks; Meta completed eight tent structures at its Ohio Prometheus campus in under a year, against 2 to 3 years for 5 earlier permanent buildings; fully modular costs about US$13.5m per megawatt against US$14.6m traditional; the model warns of an electrician shortage becoming apparent in 2027

F-021 · Confidence: medium · Basis: news_aggregation CNYES #1757744 Grade C · 媒體報導（非企業審查） Report content is a research firm's modelled estimate; cnyes report of 2026-07-30 23:07 JST

Bloomberg Economics SHOK model scenario simulation (relayed by cnyes 2026-07-30): under an S&P 500 fall of 20% with wider credit spreads and stalled data-centre construction investment, the model estimates a global GDP cost of about US$1.6trn in the first year and a possible 1.5 percentage point drop in 2027 US GDP growth if the crash occurred in early 2026; a hit of about 4% to Taiwan's annual GDP and more than 2% for South Korea; an NBER survey of nearly 6,000 corporate executives found nine tenths of respondents had not observed significant AI-driven productivity gains within three years

F-022 · Confidence: medium · Basis: news_aggregation CNYES #1754604 Grade C · 媒體報導（非企業審查） The model scenario covers "the first year after the bubble bursts" and "2027 effects if the crash occurred in early 2026"; cnyes report of 2026-07-30 18:23 JST

The Taiwan market rose more than 2,900 points in the 2026-07-31 morning session, the largest intraday point gain on record, breaking 42,800 points with a high of 42,895; TSMC's ADR rose 7.64% and TSMC reached NT$2,400 in the morning session, up NT$195, with market capitalisation of NT$62.23trn and a contribution of about 1,547 points to the index

F-023 · Confidence: medium · Basis: news_aggregation CNA #1764881 Grade C · 媒體報導（非企業審查） Morning session of 2026-07-31 in Taiwan (not the close); CNA report of 2026-07-31 10:17 JST (reporter Chang Chien-chung, Taipei, 31st)

South Korea's KOSPI rose as much as 16% on 2026-07-31, the largest gain on record (after a cumulative 17% fall over the prior three trading days); the Nikkei 225 rose 5.5% or 3,400 points in the morning session, quoted at 65,290.84 points; SK Hynix surged as much as 28%, Samsung Electronics rose 26% intraday and SoftBank Group rose 15%; the MSCI Asia Pacific Index rose 2%

F-024 · Confidence: medium · Basis: news_aggregation CNYES #1765193 Grade C · 媒體報導（非企業審查） Intraday Asian trading on 2026-07-31 (not the close); the KOSPI's cumulative 17% fall covers the three trading days through Thursday; cnyes report of 2026-07-31 10:40 JST

CNA Washington dispatch: Amazon shares closed up 3.90% on 2026-07-30 and rose as much as 10% after hours; Apple's revenue last quarter was US$109.4bn with EPS of US$2.02, both slightly above market expectations, with iPhone revenue up 21% versus the same period of 2025 and services up 12% versus the same period of 2025 though below expectations; Apple closed down 1.41% on 2026-07-30 and fell 8% after hours; Amazon plans to spend US$200bn on AI this year and Jassy said spending would gradually rise to a steady US$220bn

F-025 · Confidence: medium · Basis: news_aggregation CNA #1764279 Grade C · 媒體報導（非企業審查） US trading session and after-hours of 2026-07-30; Apple revenue = last quarter (the article's own wording) versus the same period of 2025; CNA report of 2026-07-31 08:26 JST (reporter Liao Han-yuan, Washington, 30th, dispatch)

Schroders Taiwan fund manager Ko Hung-min, 2026-07-28: tech majors' net margins were only 10-15% in 2000 against 30% today; forecasts CPU output value growing more than 5 times by 2030; highlights thermal management, networking (800G and 1.6T switches) and MLCC; his fund raised cash weighting to 15% in June

F-026 · Confidence: medium · Basis: news_aggregation CNYES #1727015 Grade C · 媒體報導（非企業審查） Remarks dated 2026-07-28; CPU output-value forecast target year = 2030; cash-weighting change = June 2026; cnyes report of 2026-07-28 21:22 JST

Cnyes 2026-07-29: the combined annualised revenue run rate (ARR) of OpenAI and Anthropic could approach US$120bn, with Anthropic estimated at about US$71bn, against Starbucks' most recent annual revenue of about US$37.2bn and McDonald's at about US$26.9bn; the original states explicitly that these ARR figures are not audited annual results

F-027 · Confidence: medium · Basis: news_aggregation CNYES #1733582 Grade C · 媒體報導（非企業審查） ARR is an annualising estimate; report dated 2026-07-29 07:50 JST

❓ FAQ

Does Amazon's "profit more than tripled" in Q2 2026 mean the core business got stronger?

How fast did AWS actually grow this quarter, and why are there two versions of "fastest"?

Microsoft surged and Meta slumped in the same week — what is the market looking at?

Per CNA's 2026-07-31 relay of Barron's analysis, it comes down to cash flow and capital expenditure: most of Microsoft's capital expenditure goes into Azure, which can rent out compute directly and generate revenue, and Microsoft did not raise its 2026 capital expenditure forecast. Meta keeps raising related spending, has now suspended share buybacks for three consecutive quarters, and saw cash flow after capital expenditure fall sharply in the latest quarter. Microsoft shares closed up more than 15% on 2026-07-30, while Meta closed down 7.95%. Meta's advertising figures remained strong over the same period (ad impressions up 14% versus the same period of 2025, average ad price up 12%, total revenue up 28%), but Barron's notes that the overwhelming majority of its AI investment flows into research and development (CNA#1764229, CNYES#1748829).

Where is the money for this AI capital expenditure coming from, and how is the bond market responding?

Per cnyes on 2026-07-30 citing Moody's Ratings, global big-tech capital expenditure is expected to total US$785bn in 2026 and could approach US$1trn in 2027, while hyperscaler bond issuance could reach US$175bn this year, far above the roughly US$30bn five-year average. Over the same period, credit spreads on large cloud providers' corporate bonds widened from about 50 basis points two months earlier to 78 basis points, and Oracle's CDS spread widened beyond 125 basis points, the highest since 2009. Two inconsistent records exist for Amazon's July bond subscription multiple (about 2.5 times versus about 3.2 times in March, and 1.6 times versus 3.4 times in March); this card records both without adjudication, per Scope item 4 (CNYES#1748532, CNYES#1734555).

What does this mean for Taiwan?

Have the figures in this card been officially verified?

No. All 17 sources are Taiwanese media (cnyes and CNA) providing second-hand relays of company results and third-party research, and this site did not check SEC filings or IR originals from Amazon, Microsoft, Meta or Alphabet (scope: see Scope item 1). Any citation of figures here must carry the qualification "media relay of results, not verified by this site" (see Scope item 1). ---

🧠 編輯判斷（J-Units）

Zero verified official anchors

Confidence: high

Handling of source disagreements (recorded without adjudication)

Confidence: high

The boundary between editorial synthesis and reported fact

Confidence: high

Separation of fiscal periods and measures

Confidence: high

🔮 待驗證假設（P-Units）

Verification against Amazon, Microsoft, Meta and Alphabet SEC filings (10-Q / 8-K) and IR originals — not performed by this site, to be added; should originals differ from the media relays cited here, the originals govern (open)

Status: open

Whether Amazon's actual Q3 2026 revenue lands within the US$197bn–US$202bn guidance range, and ex-post verification of the claim that excluding the Prime Day timing shift growth would be nearly 400 basis points higher (open)

Status: open

Whether Amazon's capital expenditure this year actually reaches US$220bn, whether Microsoft holds calendar-2026 capital expenditure at US$175bn without an upward revision, and whether Meta lands within the US$130bn–US$145bn full-year range (open)

Status: open

The two-source disagreement over Amazon's July bond subscription multiple (2.5 times / 3.2 times versus 1.6 times / 3.4 times) — pending the original offering documents (open)

Status: open

The actual closing point gain for the Taiwan market on 2026-07-31 and whether any closing record was set (this card records only morning-session intraday figures), and the durability of this rebound in subsequent sessions (open)

Status: open

"Amazon's Q2 2026 net income was US$62.6bn, up more than 243% versus the same period of 2025, including a US$53.4bn pre-tax gain from its Anthropic investment"

Amazon's Q2 2026 net income was US$62.6bn, up more than 243% versus the same period of 2025, including a US$53.4bn pre-tax gain from its Anthropic investment

TL;DR

One anti-misreading pin

Frame deconstruction (Shen Guan: where these four sets of results are most often misread)

Body

1. Delivery layer, part one: Amazon Q2 2026 (media relay of results)

2. Delivery layer, part two: Microsoft fiscal Q4 2026 and a record one-day market-value gain

3. Undelivered layer: Meta's cash flow and the missing "AI monetisation data"

4. Cost layer: the bond market is repricing the same issuers

5. Counter-evidence on cost structure: schedule compression and the labour bottleneck

6. Scenario layer: the modelled magnitude for Taiwan (a model scenario, not a forecast)

7. Transmission layer: Taiwan's record intraday gain on 7/31 and the simultaneous Asian rebound

8. The Taiwan supply-chain view and AI software-layer revenue scale (each recorded separately; not mutual corroboration)

Scope (source limitations, stated once)

Risk factors

FAQ

Q: Does Amazon's "profit more than tripled" in Q2 2026 mean the core business got stronger?

Q: How fast did AWS actually grow this quarter, and why are there two versions of "fastest"?

Q: Microsoft surged and Meta slumped in the same week — what is the market looking at?

Q: Where is the money for this AI capital expenditure coming from, and how is the bond market responding?

Q: What does this mean for Taiwan?

Q: Have the figures in this card been officially verified?

F-Units

F-001: Amazon's Q2 2026 revenue was US$200.61bn and EPS was US$5.75, both above analyst estimates

F-002: Amazon's Q2 2026 net income was US$62.6bn, up more than 243% versus US$18.2bn in the same period of 2025, and the company said it included a US$53.4bn pre-tax gain from the Anthropic investment

F-003: AWS Q2 2026 revenue was US$42.2bn, up 37% versus the same period of 2025, described by cnyes as the fastest growth since 2021

F-004: CNA relaying Reuters: Jassy said AWS Q2 revenue rose 37% versus the same period of 2025 to US$42.2bn, the fastest in 18 quarters, above the LSEG-compiled analyst estimate of 31.21% growth

F-005: AWS unfulfilled contract backlog reached a record US$496bn; the AI business and in-house chip business each have an annualised revenue run rate above US$25bn

F-006: Amazon's Q2 2026 capital expenditure was US$54.2bn, up about 69% versus US$32.1bn in the same period of 2025; trailing 12-month free cash flow swung from a net inflow of US$18.2bn in the same period of 2025 to a net outflow of US$7.6bn

F-007: Jassy announced Amazon's capital expenditure this year is expected to rise to US$220bn, 10% above the previous estimate, with rising memory-chip procurement costs one of the main reasons

F-008: Amazon's Q3 guidance of US$197bn–US$202bn is below the LSEG estimate of US$204.1bn; the company said that excluding the Prime Day timing shift, Q3 2026 revenue growth would be nearly 400 basis points higher

F-009: Amazon's Q2 advertising revenue was US$19.81bn and North America revenue was US$116.2bn, up 16% versus the same period of 2025; Adobe data show total US e-commerce sales during this year's Prime Day rose 9% versus the same period of 2025 to US$26.4bn

F-010: The overwhelming majority of AWS 2027 compute capacity is already booked; large agreements signed this year with OpenAI, Anthropic, Meta, Pinterest and Snowflake; Amazon said earlier this year the AWS AI business had passed US$15bn in annualised revenue

F-011: Microsoft's fiscal Q4 2026 adjusted EPS was US$4.74, revenue was US$90.01bn (up about 18% versus the same period of 2025) and net income was US$35.77bn; the company said profit benefited from a US$3.2bn gain on the Anthropic investment

F-013: Microsoft shares rose 15.51% on 2026-07-30, the best single-day performance in nearly 18 years; the same day the S&P 500 rose 1.7%, the Dow rose 1.2%, the Nasdaq rose 2.8% and Meta fell 8%

F-015: Meta's Q2 2026 revenue was US$60.8bn, up 28% versus the same period of 2025; adjusted EPS was US$6.18 against an estimate of US$7.22; net income fell from US$18.34bn in the same period of 2025 to US$15.85bn; free cash flow fell to US$784m, the lowest since Q3 2022

J-Units

P-Units

同事件・三視角 / Three Perspectives on the Same Event / 同一イベント・三つの視点

Internal citation chain

Sources

📊 引用級事實單元（F-Units）

❓ FAQ

🧠 編輯判斷（J-Units）

🔮 待驗證假設（P-Units）

Verification Record

Related Analyses