The consensus is that the money the four North American cloud providers spend on equipment in 2027 will still rise. The document that overturns it will arrive at the end of January: in Microsoft's (MSFT) quarterly report cash flow statement, the "purchases of property and equipment" line—if it turns negative quarter-on-quarter, the thesis breaks. In between, is there any other number that can settle the matter in advance?

The consensus is that the money the four North American cloud providers spend on equipment in 2027 will still rise. The document that overturns it will arrive at the end of January: in Microsoft's (MSFT) quarterly report cash flow statement, the "purchases of property and equipment" line—if it turns negative quarter-on-quarter, the thesis breaks. In between, is there any other number that can settle the matter in advance?

Let me give the answer first: there's no need to wait until the end of January—Microsoft's September 30 quarterly report already contains one number that settles the direction: "other receivables incurred to facilitate the purchase of server components," which was 35.7 billion USD on September 30, up 7.9 billion in a single quarter, versus only 14.4 billion a year earlier. This money is the portion where the equipment hasn't arrived but the cash has already gone out. It is still climbing, so the "purchases of property and equipment" line at the end of January will find it very hard to turn negative quarter-on-quarter.

This receivable sits in the note section "Contract Balances and Other Receivables," under other current assets, and is disclosed every quarter. The series: June 2025 8.2 billion, September 14.4 billion, December 15.1 billion, March 2026 17.8 billion, June 27.8 billion, September 35.7 billion. Over the past four quarters it rose 21.3 billion, versus only 6.2 billion in the same period a year earlier.

The cash flow statement reconciles. "Investing activities—other, net" was an outflow of 12.949 billion this quarter versus 6.209 billion a year earlier, 6.7 billion more outflow, and the main body of the quarterly report explains this increase as facilitating component purchases. Cash out 6.7 billion, receivables up 7.9 billion—the two sides mesh: the 6.7 billion was actually paid and under accounting rules does not yet count as equipment spending.

Why it can speak ahead of time: prepayments go out first, goods arrive later, and only on arrival are they counted as equipment spending. It is a one-way indicator—if it rises, next quarter isn't guaranteed to rise (if delivery slips past quarter-end it rolls into the next quarter), but if it falls, next quarter will definitely fall with it. Three of the past four quarters matched: September 2025 up 6.2 billion, next quarter up 10.48 billion quarter-on-quarter; December up 0.7 billion, next quarter up 1.0 billion; March 2026 up 2.7 billion, next quarter up 4.93 billion. June rose 10.0 billion, and the next quarter rose only 3.57 billion—that cell doesn't match. So it can only serve as a direction sign, not a measuring stick.

The second number is harder, in the lease section of the same note: as of September 30, the total payments over the entire lease term of Microsoft's leases "not yet commenced" were 364.6 billion USD, with commencement dates written between fiscal 2027 and 2034; on June 30 it was 329.1 billion, and a year earlier 106.2 billion. These leases do not enter "purchases of property and equipment": equipment under finance leases goes into the finance lease line, and operating leases don't even put the asset into equipment spending. So if it really turns negative at the end of January, it may simply be money shifting from "buying it ourselves" to "leasing it." This quarter is exactly that direction: finance leases added 5.192 billion of right-of-use assets versus 9.147 billion a year earlier; operating leases added 6.523 billion versus 1.417 billion a year earlier.

The depreciation and amortization line is also accelerating: 11.744 billion this quarter versus 8.147 billion a year earlier, up 44%. It is only charged once equipment is put into use, showing that the previous batch of equipment is entering the data centers.

The "four North American companies" consensus itself also cannot be verified with a single Microsoft line. This quarter the four companies' reported equipment spending: Microsoft 39.367 billion, Google 51.35 billion, Amazon 77.26 billion, Meta 33.94 billion, summing (self-calculated) to 201.9 billion; last quarter 165.0 billion, up 22.3% quarter-on-quarter; a year earlier 97.3 billion, up 107.6% year-on-year. Microsoft alone rose only 10% quarter-on-quarter, while Amazon rose 42.5%. The four report very close together: Meta six days before Microsoft, Google and Amazon one to two days later. The claim "wait for Microsoft's end-of-January report" doesn't hold up—within one week all four are in, and if Microsoft turns negative while the other two don't, the consensus won't break. What reports later are all suppliers—the Nvidia, Broadcom, Micron layer—reporting how much they themselves sold, not how much the cloud companies themselves committed to spend.

The market has already priced in "equipment money is fine." US stocks moved together across equipment, compute, and optics: Applied Materials +4.2%, Lam Research +8.2%, KLA +6.7%, ASML +5.0%; AMD +11.0%; Coherent +13.6%, Fabrinet +11.7%. A-share turnover is also concentrated on the same chain: Eoptolink +4.57% on 5.36 billion turnover, Foxconn Industrial Internet +6.61% on 2.78 billion turnover, Victory Giant Technology +7.72% on 2.36 billion turnover.

The strongest opposing argument: that 35.7 billion of receivables and 364.6 billion of not-yet-commenced leases are both "commitments," not "capacity already spent." The 364.6 billion is total payments over the entire lease term, spread over 1 to 20 years, so the amount falling into a single year in 2027 may be only a few tens of billions; "not yet commenced" also means terms can still be negotiated and timing can still be delayed. More critically, the 35.7 billion cannot be split into price and quantity—with this round of storage and memory price increases, the same batch of servers costs more, and the quarterly report gives only the amount, not the number of units. If what is rising is mainly unit price, then this number going up precisely shows that physical volume hasn't risen much, which is instead weaker evidence for "the money spent on equipment in 2027 will still rise."

What would count as me being wrong: in the December 31 quarterly report, this receivable falls below 30.0 billion USD; or the "not yet commenced" leases are no longer above 364.6 billion and the quarterly outflow in "investing activities—other, net" narrows to within 7.0 billion—two of these three hit, and it means the equipment pipeline was already contracting in the December quarter, and a negative turn at the end of January would not be an accounting issue. Another more direct one: if Microsoft's end-of-January guidance for next quarter's equipment spending is below this quarter's 39.367 billion, then what I said above about "it can't come down" is wrong on the spot.

Source:
1 Microsoft 10-Q msft-20260930.htm 2026-10-28 0001193125-26-436133
2 Microsoft 8-K msft-ex99_1.htm 2026-10-28 0001193125-26-436122
3 Microsoft 10-K msft-20260630.htm 2026-07-29 0001193125-26-323660
4 Microsoft 10-Q msft-20260331.htm 2026-04-29 0001193125-26-191507
5 Microsoft 10-Q msft-20251231.htm 2026-01-28 0001193125-26-027207
6 Microsoft 10-Q msft-20250930.htm 2025-10-29 0001193125-25-256321
7 Google 10-Q goog-20260930.htm 2026-10-29 0001652044-26-000084
8 Amazon 10-Q amzn-20260930.htm 2026-10-30 0001018724-26-000040
9 Meta 10-Q meta-20260930.htm 2026-10-22 0001628280-26-066773