Oracle's infrastructure line supplied about 91% of the quarter's revenue growth

Strip the cloud line out and the rest of Oracle did not grow. Total revenue was $19.35bn and cloud was $11.6bn ^[1][3], leaving $7.75bn from everything else ^[1]. Work back through the growth rates: 30% off $19.35bn puts last year's total near $14.88bn, and 62% off $11.6bn puts last year's cloud near $7.16bn, so the non-cloud base went from roughly $7.72bn to $7.75bn ^[2][3][4].

The infrastructure line supplied almost all of the increase. Grown 121% to $7.4bn, it implies about $3.35bn a year earlier and an increase of $4.05bn ^[3][4][5]; total revenue rose about $4.47bn ^[6]. That is roughly 91% of the growth from one line, which was itself 38% of revenue ^[6][7]. Seoul Economic Daily's briefing describes Oracle as increasingly seen as having turned from a software firm into an AI infrastructure provider ^[14], and the mix is consistent with that.

Capital spending was $28.5bn, up from $8.5bn, with debt at $125bn ^[9]. Operating cash flow hit a record $23bn, up 184% ^[11]. Subtract the capex and you get about minus $5.5bn, close to the reported negative free cash flow of $5.4bn against $362m a year earlier ^[10][8]. Capex ran at $1.47 for every dollar of revenue booked ^[9].

Chief Financial Officer Hillary Maxson stressed that most of the revenue from new contracts does not require large cash outlays ^[12], and Co-Chief Executive Clay Magouyrk said Oracle signed more than $30bn of new AI-related contracts without raising additional capital ^[7]. Both statements can be true while the quarter's spending still exceeded operating cash flow by $5.5bn ^[8]. Capacity grew 850MW in the three months ^[5].

The shape of the beat says the same thing. Infrastructure came in $310m above the $7.09bn forecast while total revenue beat by $210m, so the other lines together landed about $100m under consensus ^[1][4][10]. Remaining performance obligations of $664bn were $33.4bn above the $630.6bn expected ^[8][11], roughly the size of the new contracts Magouyrk described ^[7]. At this quarter's rate, annualised revenue is $77.4bn, which makes the backlog about 8.6 years of it ^[12].

For anyone budgeting cloud spend next year, that queue matters more than the growth rate. The Abilene, Texas site that Magouyrk said trained GPT-6 Astra ran on 131,000 GPUs Oracle supplied ^[6], and it was dedicated to one customer's model.

Shares are down 22% this year and rebounded as much as 7% after hours, leaving them around 16% below where the year opened ^[13][13]. In my view the equity is now priced on delivery of the backlog at a cost of capital set by a $125bn debt stack ^[8][9]. Maxson's counter-case is not weak: if the signed contracts genuinely convert without heavy cash outlays, this quarter's $28.5bn is front-loaded capacity and free cash flow turns positive on its own ^[12][9]. I would be wrong if the next two quarters show operating cash flow growing faster than capex while capacity additions hold near 850MW ^[5][11]. I would be more confident if infrastructure growth slows toward 60% while the spending stays committed ^[3].