Hyperscaler Capex Q3 2026: Big Tech AI Spending Guide

The four hyperscalers are Amazon, Microsoft, Alphabet, and Meta. Q3 2026 did not show a demand cut. It showed a delivery split: Amazon absorbed a wave of equipment the Street had not modeled, while Microsoft and Alphabet slipped below estimates because third-party capacity and chips arrived late. Full-year plans barely moved. The 2027 step-up is still the base case, but only if the tally’s universe and the lease definition are kept straight.

Q3 was a timing print, not a capex peak

Amazon’s cash purchases of property and equipment hit $77.3 billion in the September quarter, about $18–20 billion above pre-earnings estimates, and the company still left 2026 cash capex at about $220 billion. That combination only works if Q4 steps down hard: nine-month purchases were $175.7 billion, so the unchanged guide implies roughly $44 billion in Q4, versus $77.3 billion just printed. CEO Andy Jassy separately said 2027 spend will be “meaningfully more” than 2026, so the quarterly spike is a delivery bulge, not the new run-rate. [1] [2] [3]

Microsoft and Alphabet went the other way. Microsoft’s capital expenditures including finance leases were $45.2 billion, below StreetAccount’s $50.67 billion and a Financial Times-cited $51.7 billion, because third-party capacity expected in the quarter was delayed into the first half of 2027. Cash paid for property and equipment was $39.4 billion; finance leases were $7.4 billion. Alphabet’s purchases of property and equipment were $51.36 billion, about $3.7 billion under the $55.09 billion analysts expected, and free cash flow was negative $9.1 billion. Both companies refused to raise 2026 guides. Microsoft cut calendar-2026 capex to about $170 billion from about $175 billion on those delays; Alphabet held $195–205 billion and said 2027 will “increase significantly,” with a number only on the Q4 call. [4] [5] [6] [7] [8]

Meta sat in between. Company-defined capex (purchases of property and equipment plus finance-lease principal) was $34.91 billion — $33.94 billion of cash purchases plus $964 million of lease principal — and free cash flow turned negative at $2.12 billion. The 2026 range was narrowed to $135–145 billion from $130–145 billion, a higher floor, not a cut. The midpoint stayed in line with estimates around $138–140 billion. [9] [10] [11]

Quarterly table: cash purchases vs. company-defined capex

Cash purchases of property and equipment are the only line that is comparable across all four. Microsoft’s and Meta’s headline “capex” add lease items, and those items are not the same thing.

$ billions	Definition	Q3 2025	Q4 2025	Q1 2026	Q2 2026	Q3 2026	Q3 vs Q2	Q3 vs Q3’25	Trailing 4Q (Q4’25–Q3’26), cash
Amazon	Cash PPE purchases	35.1	39.5	44.2	54.2	77.3	+23.1	+42.2	215.2
Alphabet	Cash PPE purchases	24.0	27.9	35.7	44.9	51.4	+6.4	+27.4	159.8
Microsoft	Cash PPE	19.4	29.9	30.9	35.8	39.4	+3.6	+20.0	136.0
Microsoft	Capex incl. finance leases	—	—	31.9	~41.0	45.2	+4.2	—	—
Meta	Cash PPE	18.8	21.4	19.0	30.1	33.9	+3.8	+15.1	104.4
Meta	Capex incl. finance-lease principal	19.4	22.1	19.8	31.1	34.9	+3.8	+15.5	—

Sources for the cells: Amazon cash path and the $77.3 billion print; Alphabet’s five-quarter PPE series ($23.95 / $27.85 / $35.67 / $44.92 / $51.36 billion, rounded above); Microsoft cash from cash-flow history and the $39.37 / $45.2 billion split; Meta cash from cash-flow history and company-defined series of $19.37 / $22.14 / $19.84 / $31.08 / $34.91 billion. [1] [3] [12] [13] [5] [14] [15] [16] [17]

Trailing four quarters of cash PPE sum to about $616 billion. That is not a run-rate for 2027. Amazon’s Q3 is an outlier inside that sum, and Microsoft’s trailing cash figure excludes several billion dollars a quarter of finance leases that its own guidance includes.

What the definitions actually do. Amazon guides “cash capital expenditures,” which tracks purchases of property and equipment. Alphabet’s quarterly figure and its $195–205 billion guide are the same cash PPE line. Meta’s guide adds principal payments on finance leases — a financing cash outflow, not the initial lease addition. Those payments were only $2.77 billion of the $85.82 billion spent in the first nine months, so the gap is small, on the order of 3 percent. Microsoft’s headline capex adds finance-lease additions at inception ($5.6 billion in fiscal Q4 2026, $7.4 billion in fiscal Q1 2027). In July it also lengthened datacenter useful lives from 15 to 25 years, which reclassifies future datacenter leases from finance leases (inside capex) to operating leases (outside capex). That accounting change, not less building, is why the calendar-2026 figure fell from about $190 billion in April to about $175 billion in July. The October trim from $175 billion to $170 billion is the separate delivery delay. Adding the four guides as if they were one GAAP line overstates a pure cash total by roughly Microsoft’s finance-lease slice and slightly overstates Meta. [18] [19] [20] [21]

Consensus: two underspends, one large overspend, one in-line year

Company	Q3 actual (basis analysts use)	Pre-earnings consensus	Variance	Full-year guide vs. prior
Amazon	$77.3B cash PPE	$58.7B FactSet; about $20B above LSEG; one tally at $57.47B	Overspend of roughly $18–20B	Held at ~$220B cash
Alphabet	$51.36B cash PPE	$55.09B (CNBC); $55.1–55.2B in other tallies	Underspend of about $3.7B	Held at $195–205B
Microsoft	$45.2B incl. finance leases	$50.67B StreetAccount; $51.7B cited via FT	Underspend of about $5.5–6.5B	Cut to ~$170B calendar from ~$175B
Meta	$34.91B incl. lease principal	No widely cited quarterly capex consensus in retrieved coverage	Not scored as a quarterly beat or miss	Narrowed to $135–145B from $130–145B; midpoint in line with ~$138–140B estimates

[1] [22] [2] [6] [4] [7] [10] [23]

Markets treated the overspend and the underspends differently, and the cash-flow line explains why. Amazon’s trailing-12-month free cash flow fell to an outflow of $42.4 billion. Alphabet’s quarterly free cash flow was negative $9.1 billion, worse than a near-break-even expectation. Microsoft still produced about $18.6 billion of free cash flow because operating cash flow of $58 billion covered $39.4 billion of cash capex, and the stock rose. Meta’s quarterly free cash flow was negative $2.12 billion. The pattern: a capex “miss” that is a delay is tolerable when free cash flow stays positive; a capex “beat” that consumes operating cash is not. [1] [6] [5] [9]

2026 total: about $730 billion on mixed definitions

Latest company guidance, post Q3 prints:

Company	Latest 2026 guide	Basis	Implied Q4 if the guide is hit
Amazon	~$220B	Cash capex	~$44B after $175.7B in nine months
Alphabet	$195–205B (mid $200B)	Cash PPE	~$63–73B after ~$132B in nine months
Microsoft	~$170B calendar	Capex incl. finance leases	~$52B after ~$118B in Q1–Q3 on that basis; company also guided fiscal Q2 (calendar Q4) at over $50B
Meta	$135–145B (mid $140B)	Incl. finance-lease principal	$49.2–59.2B after $85.8B in nine months
Four-company sum	$720–740B; midpoint $730B	Mixed	Q4 must rise at Alphabet and Meta and fall at Amazon

[24] [1] [6] [4] [21] [3]

Outside tallies for 2026 do not match that $730 billion, because they do not cover the same companies or the same line:

• Bloomberg, reported October 30, put AI spending by America’s tech giants at about $750 billion, funded by ad, consumer, and retail cash flows. A Janus Henderson summary put Amazon, Alphabet, Meta, and Microsoft calendar-2026 capex tracking near $750 billion. Both sit a bit above the post-trim midpoint, consistent with figures compiled before Microsoft’s $5 billion cut or with a slightly broader spend definition. [25] [26]
• S&P Global Market Intelligence consensus, cited in an October 28 Amazon preview, had calendar-2026 hyperscaler capex at $762 billion, up from $362 billion. That set is wider than the big four. [27]
• Goldman Sachs Research, September 23–25, had the five largest U.S. hyperscalers — Amazon, Alphabet, Microsoft, Meta, and Oracle — on track for about $800 billion in 2026. [28] [29]
• Morgan Stanley’s mid-October global model, covering hyperscalers and neoclouds and adding lease investment, put 2026 total investment near $978 billion. [30]
• UBS, via Reuters on October 30, estimated hyperscaler capex up 76 percent this year to $673 billion. That is the low outlier versus company guides and should not be averaged in without noting the gap. [31]

The cleanest 2026 number for the four named companies is the guidance sum: $720–740 billion, midpoint $730 billion, not additive on a single cash basis. A pure cash-PPE total would be a little lower, almost entirely because Microsoft’s $170 billion includes finance leases and because the July reclassification already removed future operating leases from that guide.

2027: company language is “up a lot”; sell-side is $1.1–1.3 trillion depending on who is counted

None of the four has issued a 2027 dollar guide.

• Amazon: spend “meaningfully more” in 2027 than in 2026, with a line of sight to returns, and AWS growth projected above 40 percent in both 2027 and 2028. [32] [3]
• Alphabet: capex to “increase significantly” in 2027; details on the Q4 call. A $300 billion scare-number circulating in coverage is an investor comment, not company guidance. [6] [33]
• Meta: still building the 2027 budget, but “continue investing aggressively.” [34]
• Microsoft: fiscal 2027 capex expected to grow year over year; spend accelerates in the second half of the fiscal year as delayed third-party capacity and higher component costs land. Hood also said the company expects to stay free-cash-flow positive in fiscal 2027. Pre-print analyst expectations had fiscal 2027 capex above $190 billion. [35] [36] [37]

Sell-side aggregates, newest first where the date is known:

Source	Date	2026	2027	Who is in the number	Definition note
CNBC group comment	Oct 29, 2026	—	“Top $1 trillion”	“Hyperscalers,” unspecified	Qualitative
UBS via Reuters	Oct 30, 2026	$673B (+76%)	+25% (implies ~$840B)	“Hyperscalers”	Low vs. company guides; +6% in 2028
S&P consensus in Amazon preview	Oct 28, 2026	$762B	$1.3T	Hyperscalers, set not fully specified in the preview	Consensus estimates
Morgan Stanley	~Oct 15, 2026	~$978B	$1.58T	Global hyperscalers and neoclouds	$1.388T cash capex + $195B lease investment; leases lift the total ~14%
Goldman (Hammond / Snider)	Sep 23–25, 2026	~$800B	$1.2T	Five U.S. hyperscalers incl. Oracle	Above then-consensus of $1.1T; $1.4T in 2028
S&P Global six-company model	Early Sep 2026	$870B	$1.3T	AMZN, GOOGL, MSFT, META, ORCL, SpaceX	Company-level 2027 figures in that model: Alphabet $357B, Amazon $319B, Microsoft $189B, Meta $164B, Oracle $95B, SpaceX $197B

[8] [31] [27] [38] [30] [28] [29] [39]

The spread is mostly scope, not disagreement about Amazon or Google. Strip Oracle, SpaceX, and neoclouds out of the $1.2–1.3 trillion figures and the big four alone are not a published post-earnings consensus in the sources retrieved. The early-September S&P company splits (Alphabet $357 billion, Amazon $319 billion, Microsoft $189 billion, Meta $164 billion) sum to about $1.03 trillion for the four, but those estimates predate Q3 and Microsoft’s guide cut, and Microsoft’s $189 billion in that table is not on the same calendar or lease basis as Amazon’s cash guide. Treat them as a pre-print snapshot, not an updated total.

Morgan Stanley’s lease adjustment is the cleanest illustration of the definition problem at 2027 scale: cash capex of about $1.39 trillion plus $195 billion of lease investment equals $1.58 trillion, and the bank said hyperscalers already hold more than $1.1 trillion of lease commitments that have not started. Those commitments never hit the cash PPE line in the year they are signed. A model that uses only cash capex will look “light” against a model that books the lease. [30] [40]

What this means if you are selling into the build, or trying to fund one

The binding constraint in this print was delivery, not orders. Microsoft cut $5 billion from calendar 2026 because GPUs and third-party halls slipped into the first half of 2027, and it still guided the December quarter above $50 billion and a faster second half. Alphabet underspent in Q3 and still needs $63–73 billion in Q4 to hit an unchanged guide. Amazon overspent by ~$20 billion and did not raise the year, which means Q4 cash capex is guided, by arithmetic, well below Q3. Suppliers should read Q4 purchase orders, not the Q3 headline, and should not treat Microsoft’s lower reported capex as less demand: part of the 2026 “cut” is leases that moved from finance to operating and therefore left the capex line without leaving the build. [35] [18]

Cash coverage is now the competitive filter inside the group. In the September quarter, purchases of property and equipment ran at about 1.8 times operating cash flow at Amazon ($77.3 billion vs. $42.4 billion), 121 percent at Alphabet ($51.4 billion vs. $42.3 billion), and about 104 percent at Meta ($33.9 billion vs. $32.8 billion). Microsoft was at 68 percent ($39.4 billion vs. $58 billion) and remained free-cash-flow positive. That is why a capex undershoot helped Microsoft’s stock and a capex overshoot hurt Amazon’s. Anyone underwriting 2027 capacity — a neocloud, a lessor, a chip supplier — should assume the incremental dollar is increasingly debt- and equity-funded, not operating-cash-funded, at three of the four. [24] [41]

For a 2027 planning number, use two stacks and do not blend them. Big-four company language supports another leg up from a ~$730 billion 2026 midpoint, but no company has put a dollar on it. The sell-side cluster of $1.1–1.3 trillion is a five- or six-name figure (Oracle, and in some models SpaceX). The $1.58 trillion Morgan Stanley figure is global hyperscalers plus neoclouds, with leases added back. Quote the universe in the same sentence as the number, or the total is not usable.

---

Recent Findings Supplement (October 2026)

The four major hyperscalers—Amazon (AWS), Microsoft (Azure), Alphabet (Google Cloud), and Meta—reported Q3 2026 (or equivalent fiscal periods ending ~Sept 30, 2026) results in late October 2026, revealing sharply higher capex driven by AI infrastructure, with mixed beats/misses versus estimates, negative or sharply lower free cash flow at three of the four, and updated 2026 guidance that keeps the big-four total near $720–745 billion.[1][2][3][4]

Most report cash purchases of property and equipment (PPE) from the cash flow statement; Meta explicitly includes principal payments on finance leases in its capex definition and guidance, while Microsoft distinguishes cash additions to PPE (~$39.4B in the quarter) from a higher total capex figure (~$45.2B including leases in some reports). Amazon emphasizes “cash capital expenditures.” These distinctions matter because lease accounting changes (e.g., Microsoft reclassifying some to operating expenses) and financing can shift reported numbers without altering underlying spending.[5][6]

Amazon’s Q3 capex of $77.3 billion (PPE purchases) beat analyst estimates of ~$57–59 billion by a wide margin, but drove TTM free cash flow to a $42.4 billion outflow; the company held its 2026 cash capex guidance at ~$220 billion while signaling “meaningfully more” spending in 2027.[2][7]

• Q3 2026: $77.259 billion PPE purchases (vs. $35.095 billion in Q3 2025 and $54.208 billion in Q2 2026).[7][8]
• Beats: Significantly exceeded FactSet/Street estimates (~$58.7 billion or $57.47 billion).[2]
• 2026 guidance: Reiterated ~$220 billion cash capex (previously raised from ~$200 billion).[1]
• 2027: CEO Andy Jassy stated the company expects to “spend meaningfully more.”[2]
• Context: AWS revenue +39% YoY to ~$46 billion; heavy AI-driven investment outpaced operating cash flow (~1.8x ratio in the quarter).[7]

Microsoft’s calendar Q3 2026 (fiscal Q1 FY2027) cash PPE additions reached $39.4 billion (total capex ~$45.2 billion including leases), below its own ~$50 billion projection due to third-party capacity delays, while free cash flow of $18.6 billion beat estimates; calendar 2026 guidance was trimmed to ~$170 billion (from prior ~$175 billion or higher) partly due to lease reclassifications.[9][5]

• Q3 2026: Cash additions to PPE $39.367 billion (vs. $19.394 billion YoY); total capex cited as $45.2 billion in some reports.[10]
• Miss vs. own guidance/projection: Below ~$50 billion expectation (delays in third-party deliveries).[9]
• 2026 guidance: ~$170 billion calendar year (adjusted lower via accounting changes, not reduced plans); Q2 FY2027 expected >$50 billion.[11]
• Context: Azure +48% YoY; strong FCF beat (vs. ~$8.5 billion Visible Alpha est.) despite spending.[9]

Alphabet’s Q3 PPE purchases of $51.4 billion missed analyst estimates (~$55 billion), widened negative free cash flow to $9.1 billion, and prompted the company to hold 2026 guidance at $195–205 billion while flagging a “significant” increase in 2027.[3][12]

• Q3 2026: $51.355 billion PPE purchases (vs. $23.953 billion in Q3 2025 and $44.924 billion in Q2 2026).[12]
• Miss: Below ~$55.1–55.2 billion estimates.[3]
• 2026 guidance: Unchanged at $195–205 billion (raised earlier in the year from $180–190 billion).[13]
• 2027: CFO Anat Ashkenazi stated capex will “increase significantly”; some market speculation around $300+ billion.[3]
• Context: Google Cloud +~103% YoY to $30.7 billion; capex ~121% of operating cash flow in the quarter.[14]

Meta’s Q3 capex (including finance lease principals) reached $34.91 billion (~$33.94 billion PPE + $0.96 billion leases), pushing quarterly free cash flow negative at $2.12 billion; the company narrowed 2026 guidance upward to $135–145 billion.[4][15]

• Q3 2026: $34.91 billion total capex incl. finance lease principals (PPE $33.942 billion); vs. ~$19.37 billion total in Q3 2025 and ~$31 billion in Q2.[4]
• 2026 guidance: $135–145 billion (narrowed/raised floor from prior $130–145 billion range).[16]
• 2027: No specific guidance yet (still developing plans); expects continued aggressive investment.[17]
• Context: Revenue +26% YoY; capex outran operating cash flow (~106% ratio).[4]

Big-four 2026 capex guidance aggregates to roughly $720–745 billion (Amazon ~$220B cash; Alphabet $195–205B; Microsoft ~$170–175B incl. leases post-adjustment; Meta $135–145B incl. leases), with sell-side estimates for broader hyperscaler/AI spend (often including Oracle or others) ranging from ~$650–800+ billion in 2026 and $1.1–1.58 trillion in 2027.[18][19]

• Company-implied total (midpoints): ~$732.5 billion.[18]
• Goldman Sachs (Sept 2026): ~$800 billion in 2026 and $1.2 trillion in 2027 for five largest US hyperscalers (above consensus ~$1.1T for 2027).[20]
• Morgan Stanley (Oct 2026): ~$980 billion in 2026 and ~$1.58 trillion in 2027 for AI compute investment (cash capex baseline ~$1.388T + leases); cumulative shortfall ~$600 billion through 2028 for top four.[19]
• Broader context: Multiple reports cite >$700 billion collective Big Tech/hyperscaler spend in 2026; 2027 seen as peak cash-burn year before potential moderation.[9]

Prior-quarter context (selected, from recent filings and analyses) shows rapid sequential and YoY ramps: Amazon Q2 2026 ~$54.2B (Q3 2025 $35.1B); Alphabet Q2 2026 ~$44.9B (Q3 2025 ~$24.0B); Meta Q2 ~$30–31B range; Microsoft sequential cash PPE also rising sharply (e.g., prior quarter ~$35.8B in some reports). Full prior-four-quarter series would require pulling individual 10-Q/8-K cash flow statements, but the pattern is consistent acceleration into 2026.[18][8]

For competitors or new entrants, the scale and pace imply massive barriers (data center/power/supply constraints), but also opportunities in financing, efficiency tools, alternative chips/energy solutions, or leasing structures; however, negative FCF at three firms and heavy reliance on debt/equity raises (~$325B noted for three in 2026) highlight execution and return risks if AI monetization lags.[21]

These Q3 reports (published Oct 21–29, 2026) represent the most recent verified updates; 2027 details will likely emerge in Q4 earnings. Aggregate figures can vary by exact scope (cash vs. leases, four vs. five+ firms) and should be cross-checked against primary filings.