Platformonomics TGIF #64: September 13, 2024

Oracle’s Data Center Difficulties: FY25 Q1

Database vampire and cloud laggard Oracle reported earnings for Q1 this week. Wall Street likes the story they are telling about (finally) moving to the cloud and cashing in on the AI gold rush. But, as always, CAPEX helps us discriminate between clouds and clowns.

Quarterly CAPEX was a mere $2.303 billion, up 75% from the year prior (which was a super easy comp) but down 18% from the previous quarter.

Some observations on the numbers:

• With FY25 CAPEX guidance of $13.7 billion, they now need to average $3.8 billion per quarter to make that number. Oracle has struggled to spend to their CAPEX guidance, even a quarter out.
• Unlike real clouds whose CAPEX takeoff was largely parabolic, Oracle’s (meager) spend continues to bounce around. This suggests they still lack the systematic ability to convert cash into cloud capacity. This is an extremely complex, multi-year process that takes a long time to refine, while abrupt start-stop oscillations are extremely expensive in both dollars and deployment schedule.
• These numbers remain tiny compared to the real hyperclouds and Oracle falls ever further behind with every passing quarter:

More interesting is the narrative (the database vampire can always be relied upon for some pointed storytelling), as Oracle seeks to convince us they’re actually in this game in terms of capacity, despite their minuscule CAPEX spending.

Oracle touts “162 cloud data centers in operation and under construction”. They intentionally conflate both the present with the future, and today’s Lilliputian Oracle data centers with larger data centers that are “under construction”.

Most of those “data centers” are just a few racks. Oracle brags a full region is just ten racks. They talk about “building” data centers for AWS, Azure and Google when in fact they’re just delivering few racks to sit inside hypercloud data centers, as the once dominant Oracle database franchise transitions from a platform unto itself to a feature of other clouds.

The “under construction” data centers are another example of Oracle “learning in public”. Instead of talking up data center scale in terms of parking 747s, Larry has learned we talk about data centers in terms of their power consumption (though he’s still throwing around “acres”).

Oracle’s challenge is they are very late to this game and there are lengthy, multi-year queues for grid interconnection, even assuming you can find the power. They now seem to understand this as they’re no longer talking somewhat helplessly about power being the long pole as they have in prior quarterly calls.

They are touting two more contemporary scale data centers. One is 800 megawatts that I assume is the 747 hangar previously discussed. Larry also chose to make a self-described “bizarre” disclosure in the quarterly q and a that Oracle is “in the middle of designing a data center that’s north of the gigawatt” powered by small modular reactors (SMRs).

What is missing in this discussion is any indication of when these larger data centers might come on-line given 5+ year interconnection queues. We’ll do some research as to where these are, but given there are no built, operating or licensed SMRs, any co-located data centers are years away. So the nuclear discussion, as much as I support it (see below), seems like misdirection from the fact that Oracle still can’t build data centers at scale.

The Nuclear Power Vibe Shift Continues

I’ve been proclaiming the inevitable marriage of cloud computing and nuclear power. While I wholeheartedly welcome Larry Ellison to the party (even if it is mostly misdirection on his part), it isn’t just Larry that is realizing nuclear needs to be a big part of our energy portfolio going forward.

SaaSmageddon?

This sounds like exaggeration as Klarna tries to recreate the media attention they got from their previous AI customer service announcement. Klarna seems to have migrated from (more expensive) Salesforce and Workday to other apps (Freshworks and Deel?) as opposed to building their own equivalents in-house with AI tools.

But it is indicative of the pressures on SaaS companies. Customer push-back on reduced innovation (hello private equity!) and ever higher prices (not just private equity!) is real. AI-based tools, for companies with their data act together, make internal application development more capable and productive. All will chip away at the current SaaS model.

Living in a Glass House: Inflation Numbers

Also from the New York Times the same day:

When you do the math it seems more like 15% inflation…

Living in a Glass House: It’s a Union Shop

Will this impact digital gaming giant/chocolate-chip cookie recipe and crossword puzzle colossus The New York Times’ ambitions to be a tech company? Is your Wordle streak in jeopardy if you don’t want to cross a picket line? Is the SQL UNION operator technically a union representative?

Software Migration Alert: MariaDB

Two! Four! Six! Eight! Migrate!

Private Equity in Action: Thrasio

Oaktree says the usually quiet part about private equity out loud:

But only for a week; they then rejoined the collective fantasy that private equity is good at managing the businesses they buy: