After hanging back for a number of weeks the patterns of the technology sector are starting to make sense again, at least somewhat. I still believe the "AI trade" is over-valued and the circular nature of Nvidia's (NVDA) investments into its downstream customers further underpins that belief.

The valuations of AI-related companies jumped up this week, regardless of the yellow warning flags, which means that there was some good selling to be had. I do think there is strong "AI" business out there to be had, but the total market share of that "AI" business doesn't line up to the capital expenditures or valuations. For example, Duolingo had fantastic earnings for Q2 2025 because their domain fits very well with the strengths of Large Language Models (LLMs) and they came up with a very clever product spin in "Dualingo Max".

As an aside, "Dualingo Max" makes me laugh because it's basically... an AI chat bot. The underlying concept is the original and overplayed trope of "AI" in consumer products, but Duolingo successfully built something that is actually compelling for its users, rather than a chat bot box in the lower right corner of their web site.

Not all businesses are Duolingo and not all businesses benefit from the inherent strengths of these LLMs. I think there will still be strong Q3 earnings related that ties to AI, which will further feed the frenzy, but I don't believe the total profit growth from "AI" will match the capital expenditures being made by Google, Meta, OpenAI, etc.

Two different comments I heard on Bloomberg this week, paraphrased:

Data center vacancy rates around 2-3%

and

These investments in GPUs, even if they don't power the applications we think of, will still be useful like "dark fiber" was in the 2000s

Some mental gymnastics are required to continue to be an optimist about the big AI tech company's investments. The impression I come away with is that "we need more data centers!" and "even if they aren't useful, we'll use them eventually!"

Dark fiber was useful because it can literally be inert in the ground, backhoes accidentally damaging it notwithstanding, until it becomes valuable again. Data centers are more closely analogous to the backhoes than the dark fiber. Data centers are expensive to operate and are fundamentally depreciating assets, if you are not able to extract value out of the investment within a certain time-period, the value of the data center drops considerably.

Dark fiber was useful and powered the "web 2.0" boom, YouTube, video streaming, and so much more. But the companies who laid much of that fiber evaporated when the dot-com bubble burst.

Personally I would not treat that as a positive comparison for AI investments.

Log

• The Costco (COST) earnings were good, and the company is looking good to me. I picked up some shares since it seems to me that Costco is undervalued in today's open.
• Shopify (SHOP) also dipped, picked up some there too.

Portfolio