SP:PUBLIC

The AI Buildout’s Cash-Flow Test

What hyperscaler earnings reveal about funding durability, monetisation and the next phase of infrastructure investment.

30 October 2026 · September versus June quarters · Microsoft, Alphabet, Meta and Amazon

For information and education. Not investment advice.

Highlights

Cash investment is outrunning operating cash at Alphabet, Meta and Amazon. Their September-quarter residuals after cash capex and finance-lease principal were −$10.18bn, −$2.12bn and −$33.89bn.
Microsoft retained a positive $17.63bn residual, but it excludes $12.95bn of other investing outflows. Deducting that entire line leaves $4.68bn, down from $8.30bn in June. This is a broader sensitivity, not a measure of AI capex.
These results pressure the self-funding assumption without proving a collective funding crisis. Cloud revenue, ad improvements and backlog show different parts of monetisation; none supplies a comparable measure of incremental AI returns.

Summary

The cash test has two different answers. The four companies report substantial operating cash generation and several forms of funding capacity. But after net cash PP&E and finance-lease principal, three of the four current-quarter residuals are negative. Microsoft remains positive in this sample, though the narrower cash-capex measure leaves material component-related investing activity outside it. Its broader sensitivity is also lower than in June.

That is a test of consolidated funding, not a verdict on AI returns. CFO and capex are company-wide. Management disclosures point to demand and monetization through cloud growth, AI-related products and ad-system improvements, but the measures are blended or management-defined. They do not isolate incremental AI cash profit against the capital invested. Workforce changes and contractual commitments add context, not a savings bridge or a single bill due now.

Quarterly company-wide cash residual after net cash capex and finance-lease principal, June versus September 2026.

Source: Microsoft, Alphabet, Meta and Amazon primary releases and lease notes; Salience Path calculations. Retrieved 30 October 2026. Microsoft’s chart endpoint excludes investing “Other, net”; the sensitivity below deducts that entire line.

Observations

Quarterly cash-flow bridge

USD bn; three months ended. Net cash PP&E equals cash PP&E less disclosed disposal or incentive proceeds. “House residual” is CFO less net cash PP&E less finance-lease principal. “Residual less SBC” is a noncash economic lens, not liquidity or a reported company metric. Outflow columns show positive magnitudes deducted in the formula. Microsoft June = FY26 Q4; September = FY27 Q1. Sources: company releases and filings. [M1][M2][M3][G1][G2][G3][F1][F2][F3][A1][A2][A3]

Company / quarter end	CFO	Net cash PP&E	Lease principal	FCF on company definition	Cash residual	SBC	Residual less SBC
Alphabet / Jun 2026	39.07	44.92	0.32	-5.86	-6.17	7.96	-14.13
Alphabet / Sep 2026	42.29	51.36	1.11	-9.07	-10.18	8.63	-18.81
Amazon / Jun 2026	45.39	53.08	0.40	-7.69	-8.08	6.04	-14.12
Amazon / Sep 2026	42.42	75.88	0.43	-33.46	-33.89	5.41	-39.30
Meta / Jun 2026	31.86	30.12	0.96	0.78	0.78	7.66	-6.87
Meta / Sep 2026	32.79	33.94	0.96	-2.12	-2.12	7.36	-9.48
Microsoft / Jun 2026	55.44	35.80	0.92	19.64	18.72	3.12	15.60
Microsoft / Sep 2026	58.00	39.37	1.00	18.63	17.63	3.13	14.50

FCF on company definition is CFO less net cash PP&E for Microsoft, Alphabet and Amazon. Meta’s definition also deducts finance-lease principal, so its reported figure matches the house residual here. Quarterly Amazon FCF is our calculation using its disclosed definition; Amazon reports FCF over trailing twelve months, not this quarterly figure. Company-definition FCF and house residual therefore differ for the other three. The residual excludes M&A, equity investments, buybacks, dividends, debt proceeds and repayments, and commitments; operating-lease payments and interest are already in CFO. Alphabet’s disclosed precommencement lease prepayments were $0.20bn in June and $0.16bn in September, outside the table’s principal line; including them would make its residual $6.37bn and $10.33bn negative. Amazon separately reported other financing-obligation principal of $0.06bn and $0.11bn; it is not included above. [G3][A3]

Trailing twelve months: a different test

USD bn; TTM ending June and September 2026. Company-definition FCF is not a uniform lease-adjusted measure. Alphabet’s comparable TTM principal remains unreconciled because the older line includes prepayments; dependent endpoints remain gaps, not zero. Sources: the same company statements plus annual/YTD reconciliations. [M1][M2][M3][M4][M5][G1][G2][G3][G7][F1][F2][F7][A1][A2][A3]

Company / quarter end	CFO	Net cash PP&E	Lease principal	FCF on company definition	Cash residual	SBC	Residual less SBC
Alphabet / Jun 2026	185.68	132.40	gap	53.27	gap	28.15	gap
Alphabet / Sep 2026	179.55	159.80	gap	19.75	gap	30.41	gap
Amazon / Jun 2026	161.40	169.01	1.60	-7.60	-9.20	19.31	-28.52
Amazon / Sep 2026	168.30	210.66	1.68	-42.36	-44.04	19.87	-63.91
Meta / Jun 2026	130.30	89.33	3.10	37.87	37.87	25.14	12.74
Meta / Sep 2026	133.09	104.44	3.52	25.13	25.13	26.94	-1.82
Microsoft / Jun 2026	182.94	115.95	3.10	66.99	63.89	12.41	51.48
Microsoft / Sep 2026	195.88	135.92	3.46	59.96	56.50	12.56	43.94

Source: Microsoft FY26/FY27, Alphabet, Meta and Amazon releases and filings; Salience Path calculations. Retrieved 30 October 2026. Meta’s reported nine-month CFO differs by $1m from the sum of its reported quarters; the TTM series retains the reported nine-month value.

Alphabet and Meta still had positive company-definition TTM FCF at September-end despite negative quarters. After SBC, Meta’s lease-adjusted TTM economic lens was negative $1.82bn; that is not cash burn. Alphabet’s company-definition TTM FCF less SBC was negative $10.66bn before a comparable lease adjustment, so it is not plotted beside the normalized endpoints. No buyback deduction is added to SBC. [G1][G7][F1][F7]

Cash residual less SBC is a noncash shareholder-economic lens, not liquidity.

Cash-flow quality changes the reading

Microsoft’s investing “Other, net” outflow rose from $10.416bn in June to $12.949bn in September. Its filing links the year-on-year increase primarily to facilitating component purchases. Deducting the whole line from the house residual leaves $8.301bn and $4.684bn respectively, before SBC. This sensitivity avoids treating the cash-capex residual as the whole buildout funding requirement; it does not classify every dollar as AI spending. Component-related current receivables rose from $27.8bn to $35.7bn. [M1][M2][M3, liquidity and contract balances]

The statements also show how far cash conversion can move without establishing AI productivity:

Company	CFO June → September	Operating-balance changes June → September	What the bridge cannot prove
Microsoft	55.44 → 58.00	4.62 → 2.12	Collections and deferred revenue are seasonal; not AI profit.
Alphabet	39.07 → 42.29	-11.70 → -17.43	TPU inventory and legal/tax payments affect cash timing.
Meta	31.86 → 32.79	0.50 → 6.60	Legal and other liability movements are not recurring earnings.
Amazon	45.39 → 42.42	-7.60 → -10.35	Retail seasonality and supplier/customer timing enter group CFO.

Source: company June/September cash-flow statements; Salience Path calculations, retrieved 30 October 2026. “Operating-balance changes” sums all statement lines below changes in operating assets and liabilities, including taxes/other balances; it is not a pure recurring working-capital measure. Full line items reconcile exactly to each quarter’s CFO. [M1][M2][G1][G2][F1][F2][A1][A2]

Alphabet’s September cash included a $5.2bn payment for an older EC fine and inventory build for TPU-system sales. Meta recorded a $9.89bn legal charge; expense recognition does not establish the quarter’s cash payment. Amazon’s net-income comparison includes investment-gain effects, while Prime Day shifted into Q2 in most large countries. None should be labelled an AI return. Depreciation rises as capacity enters service; useful lives and capitalisation can change earnings timing without creating current cash. The available comparison does not establish a four-company bridge for accounting-estimate changes or incremental AI profit. [G5, CFO remarks][F1][F3][A1][M4]

Liquidity and financing: capacity, not payoff

Microsoft reported $64.51bn cash and short-term investments versus $76.84bn in June; supplier-restricted investments included within that balance were $4.0bn and $3.8bn. Alphabet’s September $265.10bn cash/securities included $90.4bn marketable equities, so it is not all cash-like funding. Its long-term debt was $125.10bn, against $98.2bn in June. Meta’s cash/securities fell to $82.45bn from $90.26bn. Amazon’s cash/securities fell to $114.23bn from $122.99bn while long-term debt rose to $173.13bn from $128.89bn; it raised $48.10bn in long-term debt during Q3. These observations show funding choices and buffers, not unconstrained liquidity or AI payback. Long-term debt here excludes current debt and leases; no uniform net-debt ranking is implied. [M1][M2][M3][G1][G2][G5][F1][F2][A1][A2][A3]

Workforce and commitments

Workforce figures are dated snapshots, not a productivity or savings measure. Microsoft reported 223,000 full-time employees at 30 June; no September headcount is established here. Its July workforce update describes an earlier reduction and voluntary-retirement programme; it does not supply a reconciled September savings bridge. On 26 October it announced a separate voluntary exit offer for eligible employees in participating teams in Canada, India, the UK and US at level 70 and below. The offer’s take-up, departures, severance cash and recurring savings were not quantified. The programmes cannot be added together, and the October offer does not establish September savings. Hood’s next-quarter expense guidance includes an estimate for the offer, not quantified realised savings. [M6] [M4][M8][M9]

Alphabet’s headcount was 198,933 at 30 June and 204,757 at 30 September. Meta reported 75,472 and 66,775 at those dates; the June figure included about 8,000 employees affected by a May reduction, most of whom were expected at June-end to leave the reported count by the end of Q3. The net change is not a verified programme-acceptance count. Meta’s $1.18bn severance expense was reported for both Q2 and the nine months, not as a separately reconciled cash-savings bridge. Amazon reported 1,595,000 full- and part-time employees at 30 June and 1,638,000 at 30 September. Its total includes retail and logistics and excludes contractors and temporary personnel; it is not an AWS workforce measure. [G3][G4][F1][F2][F3][F4][A1][A2]

Commitment disclosures use different categories and horizons. They do not form a single comparable debt-like total, and maximum contractual amounts are not immediate cash demand.

Microsoft: uncommenced leases rose from $329.1bn at June-end to $364.6bn at September-end; these are future contractual amounts, not September cash capex. [M3, Note 12][M4, Note 13]
Alphabet: Q3 purchase commitments were $924.40bn, with $213.40bn due within twelve months; uncommenced leases were $130.50bn. [G3, Future Cash Commitments] Q2 material purchase commitments and other obligations were $811.0bn, with $200.7bn short term. [G4] Category wording differs between quarters and between the filing and management call, so these amounts are not combined. [G5]
Meta: uncommenced leases were $355.73bn at Q3-end versus $278.99bn at Q2-end. Other noncancelable commitments were $454.52bn versus $349.31bn; of the Q3 category, $45.85bn was due in the remainder of 2026 and $121.41bn in 2027. These are distinct categories, not a summed obligation. [F3][F4]
Amazon: Q3 uncommenced leases were $164.92bn versus $137.21bn at Q2-end; the filing separately showed $159.64bn of PP&E commitments and $94.76bn of other unconditional purchase commitments. The Q2 filing presented a combined unconditional-purchase category of $130.07bn, so it is not a like-for-like comparison. [A3][A4]

Monetization evidence: type and timing

Microsoft — blended operating proxies, current quarter: the current call reported “Azure” growth of 48% year on year, while the prior call reported “Azure and other cloud services” at 43%. The labels and segment structure changed, so this is not a clean like-for-like growth bridge; Foundry revenue more than tripled, without a disclosed dollar denominator. Commercial remaining performance obligations were $677bn, about 30% expected in the next twelve months. These measures do not isolate incremental AI revenue, cash collection or return. Management reported third-party capacity delays, with some deliveries shifted to the second half of FY27, and expects FY27 free cash flow to remain positive; its capex figure including leases is not the same as cash PP&E paid. [M1][M6][M7]
Alphabet — management-defined and blended measures, current quarter/YTD: Q3 Cloud revenue was about $30.70bn versus $24.80bn in Q2. Management linked much of the growth to AI infrastructure and solutions, including TPU systems, and described tens of billions of dollars of AI revenue year to date. TPU hardware sales also put cash into inventory before delivery, unlike a pure hosted-compute model. No standalone audited AI profit, invested-capital base or cash yield was disclosed. [G1][G2][G5][G6]
Meta — attributed current-product benefit; prospective new-product monetization: management attributed benefits to AI-driven recommendations in its core advertising business. It described Muse as compute-intensive near term and future efficiency/profitability as a goal. The CFO said the “hundreds of billions” equity-financing reference was a theoretical founder-control calculation, not present financing needs or plans. Existing Family of Apps revenue per user is not Muse revenue; no aggregate measured AI return was disclosed. [F5][F6]
Amazon — blended cloud proxy, current quarter: AWS reported $46.00bn revenue and $17.80bn operating income. AWS includes non-AI activity, so those totals are not AI returns. Bedrock spending and token growth in the release show use, but not a separately disclosed dollar profit or cash yield. The Q3 call wording available for this analysis is a provisional Investing.com secondary transcript; its AI payback assertions are not used as verified findings. The available evidence does not provide a primary-verified AI-specific payback timeline. [A1][A3]

Key Judgments

Named assumption: the investment interval remains financeable at the consolidated-company level before AI-specific returns are fully visible. This is a funding-capacity assumption, not an AI-ROI conclusion.

Read	What the evidence says	What would change the read
Supports	All four generated positive CFO in the latest quarter. [M1][G1][F1][A1] Reported balance-sheet liquidity and observed financing sources provide context for funding capacity; Amazon disclosed $48.10bn of long-term debt proceeds in Q3. [M1][G1][F1][A1][A3] This does not show that capex earns an adequate return.	A sustained cash-funding shortfall, a material loss of financing access, or a disclosed inability to meet obligations would weaken this read. Conversely, repeated positive residuals would strengthen the self-funding case.
Pressures	Three current-quarter house residuals were negative; Amazon’s was the largest at -$33.89bn. Microsoft’s residual was positive at $17.63bn but lower than its prior-quarter $18.72bn. This narrower investment-cash measure exceeds CFO at three companies. Microsoft’s additional investing sensitivity reduces its apparent buffer to $4.68bn; it is not an unrestricted self-funding verdict.	Repeated quarters of lower net cash PP&E spend relative to CFO, rising positive residuals, or a verified cash bridge from deployed capacity to incremental returns would ease the pressure.
Leaves unchanged	The evidence does not establish actual AI ROI. Cloud growth, backlog, management-defined AI revenue, and ad-system benefits are not a comparable measure of incremental AI profit or cash yield.	A company-level series linking AI-specific revenue to incremental margins, invested capital and collected cash, with timing and attribution disclosed, would permit a return assessment.