Platformonomics TGIF #71: November 1, 2024

Platformonomics TGIF is a weekly roll-up of links, comments on those links, and perhaps a little too much tugging on my favorite threads.

A Post About The Washington Post

Why was the media so strongly motivated to ignore the first and most important point in Jeff Bezos’ explanation of the Washington Post’s Presidential endorsement decision?

News

Big CAPEX Scorecard – Q3 2024

Summary:

• Big CAPEX spent over $65 billion in Q3. 90%+ of that goes to cloud/AI infrastructure except Amazon (AWS is roughly half their spend — we’ll know precisely in their annual report).
• Big CAPEX has spent over $171 billion year-to-date.
• Amazon: $22.79 billion +79% (includes fulfillment spending). New all-time quarterly high.
• Google: $13.06 billion +62%. Nearly equals full CY17 spend ($13.18 billion).
• Meta: $9.2 billion +36%. Below guidance, which they attributed to lumpy payments.
• Microsoft: $20 billion +79%. New all-time quarterly high. Nearly equals full CY20 spend ($20.6 billion).
• Amazon and Microsoft are both emphasizing their data centers are multi-decade assets.
• The short duration pivot table operators on Wall Street hate this spending.

Amazon CAPEX – Q3 2024

Google CAPEX – Q3 2024

Guidance:

Meta CAPEX – Q3 2024

Guidance:

Turning now to the capex outlook. We anticipate our full year 2024 capital expenditures will be in the range of $38-40 billion, updated from our prior range of $37-40 billion. We continue to expect significant capital expenditures growth in 2025. Given this, along with the back-end weighted nature of our 2024 capex, we expect a significant acceleration in infrastructure expense growth next year as we recognize higher growth in depreciation and operating expenses of our expanded infrastructure fleet.

So your second question was about Q4 CapEx. The expected step-up in Q4 CapEx from Q3, part of that comes from increases in server spend and to a lesser extent, data center CapEx.
But with servers there are these timing dynamics at play that you referred to because we had these server deliveries that landed late in Q3. And so the cash doesn’t go out the door basically until Q4, and that’s when you’ll see the CapEx show up.
And given the nature of capital expenditures, generally, there is some — actually quite a bit of lumpiness quarter-to-quarter. So it’s a little bit hard to sort of extrapolate from any particular quarter. Overall, I’d say we’re growing our infrastructure investments significantly this year, and we expect significant growth again in 2025.

We’re training the Llama 4 models on a cluster that is bigger than 100k H100s or bigger than anything that I’ve seen reported for what others are doing.

It is worth noting that companies who spend a lot more on CAPEX than Meta don’t disclose the size of their training clusters. But lets get that long-overdue cage match between Elon and Zuck scheduled.

Microsoft CAPEX – Q3 2024

Guidance:

Roughly half of our cloud and AI related spend continues to be for long-lived assets that will support monetization over the next 15 years and beyond. The remaining cloud and AI spend is primarily for servers, both CPUs and GPUs, to serve customers based on demand signals.

Demand continues to be higher than our available capacity.

We expect capital expenditures to increase on a sequential basis given our cloud and AI demand signals.

One of the things that may not be as evident is that we are not actually selling raw GPUs for other people to train. In fact, that’s a business we turn away, because we have so much demand on inference that we are not taking what I would – in fact, there’s a huge adverse selection problem today where people, it’s just a bunch of tech companies still using VC money to buy a bunch of GPUs.

Also interesting (and maybe helps explain why Benioff is freaking out):

We now have over 16,000 paid Fabric customers, including over 70% of the Fortune 500.

Microsoft’s Copilot Stack sits on top of Fabric.

Cloud Growth Rates – Q3 2024

Platformonomics Repatriation Index™ – Q3 2024

Remember cloud repatriation? Cloud customers were supposedly yanking lots of workloads from the cloud for the pleasure of racking and managing servers themselves.

Despite that repatriation headwind, AWS/Azure/GCP all saw growth accelerate this quarter. And despite benefiting from the double-barrelled tailwinds of both repatriation and AI, our repatriation poster children Digital Realty Trust and Equinix turned in blistering growth of 2% and 7%, respectively.

The Platformonomics Repatriation Index™ hit another all-time low, meaning repatriation keeps getting harder and harder to find outside of VC blog archives.

O Scaling Laws, Where Art Thou?

Amazon Even More Bullish on Finding a Compression Algorithm for Experience

They’re still touting their feature count:

In the last 18 months, AWS has released nearly twice as many machine learning and gen AI features as the other leading cloud providers combined.

Yet one feature still isn’t on that list: a competitive frontier model.