Here’s a piece making the rounds today, if you haven’t seen it yet.
Jim
3 Likes
I don’t know if I believe everything I just read but regardless, this may be the most important thing I have read in months.
For those that lack the time or interest to read the full article, spend 30 seconds on this graph:
2 Likes
We’re starting to see the “AI bubble” narrative gain traction and enter the Overton window. A few companies seem like they’re taking steps to prepare for at least the potential of one, helping increasing the anxiety around it and maybe even helping to precipitate it.
Jim
This is a map of the data centers completed, planned and being built in the panhandle of Texas. Over 2/3 rds of these are planned or under construction. These projects take years to completion.This is only a partial map of TX. It’s really amazing how many are being built in the whole state. I am not too worried about the AI bubble as I feel in the face of facts that it is not true - just my opinion. That big black dot on the middle top is the Fermi project - largest in the world. I’m guessing the total investment just in this part of the map is in the trillions - my guess. I’m wondering if a small part of the hundreds of millions being spent at each location will find its way into the local communities. The workers fly in and out weekly, get hotels, get rent cars and eat out. They don’t pay for any of these expenses as they get a per diem for everything. The financial scale is amazing. Someone from Abilene TX told me that the smallest older 1000 sq foot homes are selling for $300,000 up. They are desperate for housing for these workers who are cash rich. It’s got to be a boon for the locals communities especially the rural communities that are literally out in the sticks. You can check out your community and state at the link down below…doc
US Data Center Map: 374.7 GW Buildout, 318.6 GW Pipeline
edit: the dark small dots that are burnt red are the completed data centers. From what I gleened they are much smaller than the new ones being built. This should be a huge demand for everything used in constructing and then filling them up with compute power.
2 Likes
Seems like a tortured analogy to express the simple sentiment that those building AI data centers are taking a huge risk based on an assumption of future AI demand.
Who are these companies… can you share more details. I am trying to understand it better.
Thanks for the share. Everyone is aware of OpenAI/ Anthropic commitments. While their revenue, cash flow profile is unknown. My thesis is when these 2 companies come public, file their 10-Q’s, it will become clear. Yes, currently Anthropic is experiencing 10x revenue growth for the last 3 years, at the same time they are already over $100 B, is 10x possible next year? Even 1x is awesome $200B, which may be just sufficient to pay the bills.
Right now, select few, read it as VC’s and some hedge funds, get to see their balance sheet. Most of us are at dark.
Separately, OpenAI, Anthropic is now systemically important companies, can they be forced make their financials public irrespective of their public listing? Especially given how much credit rides on them.
My instinct is to de-risk my portfolio and go cash big time. But, the amount of CG tax I will end up paying stops me from that. I strongly feel, when this bubble pops we are going to see not only big drawdown, but a decade long bear market to digest all the capacity build.
I am trying to find areas where your can rotate to… looks like everything will be impacted.
Here’s a Bloomberg story highlighting a couple: https://www.bloomberg.com/news/newsletters/2026-08-21/ai-suppliers-are-trying-to-prepare-for-the-data-center-bubble-to-burst
Jim
Anthropic’s financials will become public in weeks, if not days. We’re going to see these quite soon. We have high-level financials from OpenAI that are absolutely horrific. So much of AI runs through these two companies and it seems like they might never be able to turn a profit.
We see key fundamentals of the token market that look horrendous. Token prices are cratering. Users are switching to the lowest-cost provider with acceptable quality. Can any AI lab offer a defensible niche product or is the current crop of AI inherently commoditized?
In the dotcom/telecom bust we saw telecom overcapacity for a decade and more. Similar with the housing bust, with malaise for at least a decade. I think you’re right to expect a decade to absorb the excess capacity.
Jim
4 Likes