(Part 1 of 5) Midyear State of Data Centers and the AI Bubble Report - Pt. 1 Data Centers and Data Center Construction

In Part 1 of 5 of the Midyear State of Data Centers and the AI Bubble Report - we deep dive Data Centers and Data Center Construction in the first of a two part series.

Don Johnson (@DonMiami3), Chief Economist

Good Wednesday evening MacroEdge Readers and Community,

Hopefully you’ve had a great start to your summer and are enjoying these long July days. I know many of our readers tune in specifically to this report series (and not many of the other reports that are put out) because of its relevance and role in the economy and markets today. Over the last several years, the AI and data center story has been one that has put even the biggest disbelievers in a state of surprise for a number of reasons. We’re now hitting what I believe to be terminal altitude for the data center boom cycle - and we’re going to cover a lot in this Part 1 report.

I’ve decided to break this report into five parts to give us ample time to actually cover and break down different areas within this entire behemoth - that way we’re not lumping everything in when it shouldn’t be. The five parts of the ‘State of Data Centers and the AI Bubble Report’ will be as follows:

• Part 1: (Part 1) Data Centers and Data Center Construction

• Part 2: (Part 2) Data Centers and Data Center Construction

• Part 3: A Look at the LLMs and Assessing the Moats

• Part 4: How Big is the Bubble, and Will There be Bailouts?

• Part 5: A Summary and What’s Ahead for the Second Half

Parts 2, 3, 4, and 5 will be released on Wednesday for the next four weeks of reports (five including this release).

The data center cycle has been one of the most prolific infrastructure buildouts in decades - but there are so many pertinent questions about it all, namely with the ‘what, why, and how’ as the overarching themes… It’s one of those things that I think we will look back at and not in fact compare to the railroads, oil infrastructure, and highway buildouts of decades past. While we are undoubtedly continuing to propel ourselves into a more & more digital reality and this unique world we are living in has enabled people like me to reach tens of thousands of people in an instant - much of that technology has been around for decades now. AI itself - in its current form - is really not ‘AI’ in any way, shape, or form. It’s essentially an enormous content vacuum that’s permeating absolutely everything we do - and much more a data collection and ingestion mechanism than anything else. While I know some of the proponents will respond to this with ‘but, but, but’ - but the next time you log in to your email, or have your face scanned to check in at the airport, or send a message on Twitter (X) - just remember that you are part of the ‘AI’ project. As an ingestion system in its current form - current LLMs and AI technology is astoundingly inefficient - on X I’ve likened this to ‘Soviet-style’ technology and infrastructure where we’re building something to build it. When it comes to the commoditization of the space, US AI executives are panicking about innovation going on elsewhere in the world that threatens their business models through open source methodologies. The how or why in this technology is a story in and of itself - and now we find ourselves going - just how much infrastructure are we willing to commit to that could potentially be obsolete just a few years from now? Are we really supposed to sit here and believe that data centers and this technology won’t evolve? If so - is this particular technology really a technology at all - or are there forces behind it that don’t meet the eye? We will find all of those things out as this entire cycle continues to evolve.

The point of Part 1 is to look at the data centers supporting this entire thing - and I remember in the early days just how fast these things were getting approved. In a fun fact about me - I was actually dating a banker in West Texas at the time and recall with great detail just how fast Stargate approvals, permitting, and details were rammed through without second thought. Local (municipal) and county-level politicians for the first several years of this buildout cycle got away with committing to these deals behind closed doors, and behind a defense written by lobbyists that impacts to the community would be permanent and profound in many cases - which is where the story starts to falter. On that front, what I expect to see as we roll through the second half of the year, and construction teams start to wrap up their work at these ‘hyperscaler’ sites - is an interest from the Trump administration in figuring out how they can prolong the buildout of these facilities - even if they aren’t needed. Yes, to my point above, that puts this infrastructure buildout more closely to something we might expect in China or the Soviet Union - but as someone that believes we are at or very near ‘peak innovation’ for the current tech innovation cycle - politicians are far, far, far more interested in the short term benefits than any long term consequences - especially given what they (and I, and likely you) know about the attention spans of voters at any level. When you look at the current technology landscape - and it’s something that I do frequently as we have an entire practice at our firm delivering such solutions - I agree with Ed Zitron and others that ‘peak innovation’ is something that we are witnessing now, but maybe on more of a rolling basis than a single event.

AI and infrastructure buildout popped out of nowhere a few years ago - and all of the companies sitting on hundreds of billions of dollars in cash decided that it was the sexy new thing that they should chase, and that there would be some sort of return in the end. What I know about commoditized infrastructure buildouts is that they look very sexy in the beginning, but the endings across whatever the commodity might be do not look so hot. Months back, those that read our smaller Macro Note series will know that I likened the infrastructure buildout with data centers to the shale boom of the early 2010s - but on an even grander scale. The stakes could never be higher for the politicians that have committed wholeheartedly to making this (almost) the entire economic engine of the United States for the time being, and I expect that they’re going to ride on the wagon until they make it to the Oregon Trail. Is the next era of ‘innovation’ we want to leave behind for the future generations, one populated by infinite data centers (taxpayer-backed), an existence devoid of any social interaction, and one that we engage in these Chinese-style buildouts for ‘number go up’ economics where we dig holes and then hire people to refill them? That’s a question that collectively, people will have to decide for the remainder of the decade and beyond.

My goal has never been to make that decision for anyone else, and I think with the discussion focus areas and data below, there are many useful findings and facts about the current state of this major economic driver, so I hope you find this series very useful for portfolio strategy, business, or even personal decisions or choices outside of the financial or technology worlds. One of the most compelling parts of what I do is seeing our organization compile all of this data, and then knowing that there are tens of thousands of brilliant readers, minds, traders, business owners, and so much more interpreting it in their own light.

If you did not have a chance to read the first part of this broader series from February and March, I highly recommend that as a starting point so you can get a sense of how the sector and space has progressed since then, from the time I issued a warning that the Trump administration had better find its newest growth engine, to the Iran War starting shortly thereafter and inflation and equities booming side by side… It’s been an interesting year thus far and this five part series is not about my broader macro market commentary - which for the remainder of the year is focused on a very binary outcome - either the gates to inflation (hell) being opened with everything melting up concurrently, or inflation/war getting to a point where things finally stagnate - and along with things like data centers - this sideways stall out ends up materializing into a more significant slowdown - whether through stagflation or through cooling of said economic activity.

Maybe - in the end - this will sum up the whole phase of this buildout:

In Part 1 of 5 of this ‘Midyear State of Data Centers and the AI Bubble Report’, focusing on data centers and data center construction - we’re going to cover the following sections:

• Background to the Current Capex Cycle, Tax Benefits, Backroom Deals, and More

• Current Data Center Technology

• Consumer Power and Water Impacts

• State of Data Centers - Data Center Construction Data

• State of Data Centers - Moratoriums, Bans, and Much More

• (Part 2) State of Data Centers - the Public Equity Markets

• (Part 2) Are Data Centers Shale on Steroids?

• (Part 2) Data Centers and the Rings of Empty Chinese Cities

• (Part 2) Finishing Comments

Obviously, each section is going to be unique, but there’s enough information in here for everyone -regardless of what you’re here for - to take something useful away. I hope you find this compilation of our data, my thoughts, commentary, (yes, opinion too), useful on this very warm July week. It’s going to be a very interesting second half of the year for this entire capex-funded space, and I am most of all interested in seeing how things progress from where we are now to the December/January update when we get another many months under our belt.

Background to the Current Capex Cycle, Tax Benefits, Backroom Deals, and More

The current capex cycle

The current AI capex cycle is historic. We’re currently sitting at/near the cycle peak - and I expect us now to be slightly off of the highs from earlier in the year. Just because the hyperscalers are dropping ridiculous forward capex guidance does not mean that those figures cannot be adjusted if the math finally hits some of these enterprise leaders over the head (I expect that it will). In today’s Alphabet earnings, they announced a higher capex figure than scheduled by ~$20bn or so; markets did not react positively to that line. As someone who frequently utilizes Gemini and other Google AI solutions - the product quality seems to be rapidly degrading - and even though Alphabet is able to *handle* what they’ve taken on - I am not sure that means they should be. The amount of data collection occurring now through Google is also astonishing.

Not yet a MacroEdge subscriber? Upgrade to Ozone to get all of our research, data, commentary, portfolio strategy, and much more below: