Signals

All three large cloud providers reported faster growth in the September quarter. AWS revenue rose 39% year on year to US$45.98bn, ahead of the US$45.29bn StreetAccount consensus and up from about 37% in the June quarter, CNBC reports, while AWS operating income rose 55% to US$17.77bn. Microsoft disclosed Azure revenue in dollars for the first time: US$33.17bn, up 48% after 42% growth in the June quarter and above the US$32.32bn consensus. That puts Azure at 72% of AWS's size, against 58% two years ago. CNBC says Google Cloud more than doubled. Tomasz Tunguz puts it at US$30.7bn, up 103%, while The Register gives US$24.75bn. The higher figure is the one consistent with a doubling, though we have not reconciled the two against Alphabet's own filing. The sector's shares fell regardless, though not uniformly. Microsoft rose 4% in extended trading after its report, but Alphabet lost 12.6% in Thursday's session and ServiceNow almost 27%, on our dashboard data, and Amazon slipped 1.4% on Friday as the Nasdaq Composite fell 2.9%. We do not assign a cause to any of these moves. Mark's view is that accelerating cloud revenue makes Alphabet's fall hard to justify, and he treats ServiceNow as part of the SaaS-disruption story rather than the infrastructure one. The conversation's broader framing was that uncertainty can drive sharp market reactions, a claim about market behaviour rather than an explanation of this week. The reported numbers support something narrower. On CNBC's growth rates, AWS's operating margin was about 39%, up from about 35% a year earlier, so the largest cloud widened its margin while it built. That is one company and one quarter, and depreciation on recent spending may have further to run. For positioning, the test is whether the acceleration persists and what it costs. Microsoft's finance chief, Amy Hood, guided Azure to 49% to 50% growth at constant currency for the December quarter, against a 45.1% consensus. Amazon's Andy Jassy said AWS can grow north of 40% in each of 2027 and 2028 and become a trillion-dollar revenue business in five to 10 years, an ambition rather than a forecast. Watch Azure's delivery against that guidance, AWS's margin as new capacity is depreciated, and the gap between cloud revenue growth and capex growth over the next two quarters. Watch too whether application-software names keep falling while the infrastructure beneath them reports faster growth.

The New York Times reports that Amazon, Microsoft, Google and Meta said over the past two weeks that they are on track to spend more than US$720bn this year on data centres and other capital expenses, roughly triple the level of two years ago, and all are expected to spend more next year. The four set a spending record last quarter, and most now spend more cash on data centres than they generate from running their businesses. We have only the NYT's opening passage, not the full article, but the total checks out against CNBC's reporting: Amazon at US$220bn, Alphabet at US$195-205bn, Meta at US$135-145bn and Microsoft at about US$170bn come to US$720bn at the low end of those ranges. The word that matters is "most". Alphabet, Meta and Amazon have all reported negative free cash flow, while Microsoft's stayed positive at US$18.63bn, down 27%, and finance chief Amy Hood said it will remain positive through fiscal 2027. Microsoft still announced US$25bn of new bonds on 29 October, due to close on 4 November. Mark's reading is that even Microsoft has to use credit to build data centres, though with a AAA rating and positive cash flow it appears to be borrowing by choice rather than need: Hood spoke of tapping outside funding when conditions are attractive, and said leasing is already in use. What the money buys is shifting too. About 75% of Microsoft's capex now goes to short-lived assets such as chips, up from half a year earlier. CNBC reads that as favouring infrastructure that turns into revenue quickly; on our reading it also means more of the spending must earn its return before it needs replacing. The NYT adds that the boom accounts for much of the resilience of the US economy, which ties growth as well as equity returns to a curve that has to keep rising. For investors, the useful measure is now how much of this spending is funded from outside the business, and in what form. Microsoft is the marker: if its free cash flow turned negative inside Hood's horizon, none of the four would be self-funding. Watch how external funding divides between bonds, leases and, at Meta, equity, since leases are harder to track than public issuance. CNBC notes investors have grown warier of heavy debt as central banks raise rates, which matters more the longer the build depends on outside money. Watch also the useful lives companies assign to chips, and US growth data: a spending plateau would now register as a macro event, not only a sector one.

Intel shares fell 29% on Friday 30 October, which CNBC calls the stock's worst day since the 1970s; an earlier version said 1974. The fall came the day after third-quarter results in which adjusted earnings per share of 52 cents beat the 40 cents expected, and revenue of US$16.51bn, up 21% on a year earlier, was in line with the US$16.5bn consensus. For the fourth quarter Intel guided to adjusted EPS of 25 cents on revenue of US$15bn to US$16bn, against LSEG consensus of 43 cents and US$17.1bn. CNBC says the guidance reflects PC softness and a plant shutdown in Israel. Finance chief David Zinsner said, "We see the PC slowdown continuing at least through the first half of next year", and that an explosion at an Israeli fab in October, which killed two workers and seriously injured a third, will cost about US$600m in lost revenue over two quarters. The segment numbers complicate the usual story of a legacy drag on an AI winner. Client computing and physical AI fell 5% to US$8.1bn and missed estimates, while data centre and AI rose 78% to US$7.3bn, so the two groups are now close in size. CNBC puts PC weakness down to soaring memory and component costs. Intel entered the report up about 150% for the year, though down roughly 23% in October, helped by demand for the CPUs that orchestrate agentic workloads. How much of Friday's fall reflects the forecast, the valuation or the wider market cannot be read from one session: the Nasdaq lost 2.9% that day, Nvidia 4% and Qualcomm 11%. D.A. Davidson, which cut the stock to neutral from buy in a note titled "Not fixed yet", wrote that "it is hard to justify the current multiple when weakness is so significant". Mark regarded the reaction as excessive. John's counter-framing was that the PC is largely over, that memory is going to higher-margin systems, and that Intel is not exposed to the phone downturn. For positioning, the test is whether data-centre growth can outrun a PC decline Intel expects to last until at least mid-2027. AMD reports this week, giving a second read on CPU demand. Chief executive Lip-Bu Tan said Intel will have "a couple of customers ready to announce the next two quarters" for its foundry, which CNBC says still has no external customers for its new chips. Watch also the Israeli plant's return, since the shutdown hit 10-nanometre shipments. Friday's fall cut about US$12bn from the US government's 10% stake, still a profitable holding, and about US$6bn from Nvidia's.