During the week, I published a piece on recent AI infrastructure developments that can be found here. I also published my latest holdings, portfolio management approach & updated performance vs. the S&P 500.

Capacity Plans:

While it was a quiet news week for most public companies, the same was not true for Meta. Following a $10B Canadian data center announcement, headlines from Bloomberg broke about the company planning to double compute capacity in 2027. It’s worth noting that this does NOT mean CapEx will double Y/Y in 2027. That is highly unlikely, with a lot of this added capacity coming from spend incurred before 2027. Some large accounts on social media are concluding that CapEx will double based on this news. That’s not the right takeaway. Furthermore, while this is absolutely a sign that Meta is pushing ahead on capacity plans, we don't know if this is a reiteration of previous internal intentions or a change. So while this does still seem bullish for AI infrastructure, that makes it hard to gauge exactly how bullish. That 2027 CapEx commentary on the Q3 call (if we get it and I think we will) should be interesting. Growth rates will almost surely slow compared to 2026, but the question is... by how much?

Public Cloud Entry Context:

Next, it sounds like recent news that Meta is joining the public cloud market has everything to do with how compelling the offers it's receiving are. They’re too good to resist. So while they are renting out some capacity, it sounds like that’s coming at the expense of using the compute elsewhere rather than just not needing it anymore. That makes complete sense, considering they’ve spoken constantly about how many use cases they have for this spend and how fungible it is.

Improving Model Momentum:

And for more Meta news, they’re now rolling out paid developer APIs to access their new multi-modal Muse Spark 1.1 model as they add another leg to their AI monetization. While this new model isn't the best out there, Meta is very pleased with its performance and even went as far as saying it's better than they were hoping for.

This general model momentum is even allowing them to swap existing 3rd party model usage with their own image creation model released this past week. The encouraging news comes just months after discussing disappointment in pace of improvement. Zuck sees a clear ability to use improving models and new APIs like the one discussed above to offer cheaper tokens and intelligence at still strong margin. It sounds like they plan to deeply undercut OpenAI and Anthropic.

Netflix is exploring different ways to provide its content via channels and different bundles. Per WSJ, this is a response to softer engagement trends and declining TV viewership market share (now at 12+ month lows). I know this has been a darling for quite some time, but I do think these struggles could easily continue. Google, Apple & Amazon all continue to push further into entertainment. And? They don’t need to make any money on these libraries. They can operate at razor thin margin & use this as a vehicle to cross-sell more things/ads to consumers. Netflix does not have that luxury. They have to compete with several of the richest companies in the world and do so more profitably. This (I think) belongs in the too hard pile. I know that’s not a popular opinion, and I hope in a year I look extremely wrong and bulls can laugh at me.