Submitted by Peter Tchir of Academy Securities

We have been attacking this issue orthogonally for the past few weeks.

• Last weekend’s Cheap China Compute brought up several issues facing the AI Spend.
• On Thursday we published Braggawatts (which should probably be BragCompute or something), but the concept is that a lot of the announced deals are missing some, or all of the following:
  • Enough electricity, especially at peak usage times, to fulfill their commitments.
  • Access to water and other resources to function.
  • Getting the various chips on time, connected and installed (hearing China is threatening to restrict exports of fiber-optic cables (another, in a long list of reasons why the U.S. (and others) need to pursue ProSec™).
  • Municipal, State, or even Federal regulatory approval (to the extent they are necessary).

To the extent this is true (and we also see construction cost overruns and delays) this is probably good for credit spreads, but negative for equity valuations.

Indirectly, we have been addressing two issues, for even longer. While these issues have been in the background, they are rising to the forefront more quickly than anticipated:

• Market structure (ETFs, leveraged ETFs, 0DTE options, etc.).
• The need for the AI industry to rapidly adopt far better community outreach! Our somewhat silly, AI-generated, picture of workers (dressed for casual Friday), carrying torches, storming a data center, seems less silly by the day. The AI Revolution is growing faster than we thought and is already influencing state and local politics coming into the midterms.
  • While it might be easy to ignore New York State’s recent “moratorium” (though be careful doing that, as upstate New York is far less “liberal” than New York City), it is more difficult to ignore the change in Texas.
  • Governor Abbott now seems to be discussing a “prohibition in rural neighborhoods.” That goes beyond previous discussions introducing rules around electricity, water, noise, etc. This is a far cry from when the Governor was attempting to make Texas a dominant hub for AI and datacenters!

It is quite possible we won’t see a slowdown in AI spending (that still seems to be what markets are pricing in), but the case that this narrative experiences a serious “hiccup” is growing.

The Market is Always Right

Since I spend half my time trying to fight markets, I’m not sure I agree with that, but it seemed like a good way to highlight 3 important things that happened late this week. Yes, the Philadelphia Semiconductor index bounced back this week (up 1.2%), but the Nasdaq 100 slumped 1.6%.

• INTC earnings seemed great. I don’t attempt to forecast earnings, but when the earnings hit the tape, virtually everyone I trust on social media and traditional media seemed to view them as very positive. Yet INTC dropped about 8% on Friday.
• On Wednesday, Anthropic and AMD announced a deal. Maybe I’m confused, but it seems to me that a month or two ago, that sort of announcement would have been very positive for AMD stock. Yet AMD stock fell 5.4% from Wednesday’s close.
• On Thursday, ORCL announced a $7 billion deal with the Pentagon. Seems impressive (and very much in line with our ProSec thesis). Yet, ORCL hit a 52-week low on Friday, falling on both Thursday and Friday.

At “best” this is telling us that the market is setting a very high bar for further upside.

At “worst” it is telling us that positioning is overly long, and it elevates our market structure concerns.

What Goes Up Must Come Down?

The inflows into the semiconductor space have been quite incredible.

Source: Bloomberg Finance L.P. (SOXX US Equity — iShares Semiconductor ETF)

It is difficult to look at this chart and not see:

• A decent correlation between inflows and performance (momentum and the narrative have worked hand in hand to bolster the market).
• A chart that looks “parabolic” in nature, which is always concerning (at least to me).

SOXX assets under management grew from $20 billion at the end of March to over $47 billion (a combination of price and inflows). Some serious wealth effect.

SOXL hasn’t had the same pace of inflows (it has had outflows since the rally began in April). But this 3X leveraged ETF has assets of $20 billion, representing $60 billion that needs to be rebalanced daily (the bigger the move up, the more it has to buy; conversely, the bigger the move down, the more it has to sell). That daily rebalancing is separate from inflows or outflows.

This combination of ETFs (and other ETFs focused on semis, including a large number of single stock leveraged ETFs) adds to my concern.

Distilling

Not the fun kind of distilling (which you may need after reading this report), but the “distilling” Chinese AI is using to speed their model “training” and make their “training” far cheaper is a real concern. You are seeing the U.S. government examining what can be done about this.

We will get into more detail on this later this week, as I’m having several conversations with Academy’s GIG members on this subject.

Increasingly I’m worried we are seeing a “rinse and repeat” for China:

• Flood the market with cheap “something” (in this case compute).
  • Maybe the “thing” isn’t as good, but it is so darn cheap, it is tempting.
  • Maybe it is cheap due to a variety of factors (unfair government support, loose (if any) enforcement of Intellectual Property protection, etc.).
• Use that pricing power to slow global competition.
• Add in some legitimate advantages China has (no concept of NIMBY, a decade or more of rapid expansion of energy production and their grid, their own legit Intellectual Property, and the production, at scale, of a variety of lower level, but useful chips).

As much emphasis as Academy has placed on ProSec™, I’m fearful that we underestimated the potential for Cheap Chinese Compute to disrupt not just our AI/Data Center Industry, but also at some level, our National Security.

This “Is It Worth It?” Narrative Shift

The media is an incredible source of information. I incredibly value my engagement with media. Not just the brief moments in front of the camera or the microphone, but all the discussions we have. Some on background. Some on views that never get published. That has been incredibly helpful, but one other thing has been of incredible use to me as a strategist:

• Seeing the shift in the media narrative before it plays out.

If you don’t think the media influences markets, you can skip this section.

What I see (and experience) are “cycles” that develop over time.

A few weeks/months ago, the media wasn’t that interested in negative stories on the AI Spend (except maybe to highlight troubles in the bond market).

Then, no strategist or analyst wanted to be involved in negative stories on the space, because it was dead wrong, at least based on stock prices.

But I believe I can “sense” a shift in what is being asked, what is being published, and more importantly, what is going to become the narrative!

This affects everyone from CEOs to strategists. What CEO was going to say they might slow down spending (either on the build-out, or the use side)? Could any CEO really say, “we’ve been seeing our token costs increase, and despite trying to use this stuff (that apparently everyone else is having success with), we are struggling to get a lot of value out of it (we discussed simple Return on Investment as a potential issue in Thursday’s report).

Any CEO who was willing to do that might as well have branded Luddite on their forehead and waited for their stock to crash as “everyone knows you need to be using AI.”

Yet, we’ve mentioned this, but one conference organizer’s comments are worth repeating.

• In 2025 AI sessions were wildly popular and received high scores.
• In 2026, attendees wanted case studies and examples (which I tend to think means that they have been experimenting, with limited success).

Source: Bloomberg Finance L.P. (SDLLMTN Index — Silicon Data LLM Token Expenditure Index)

I will admit that I don’t have a good grasp of how accurate this chart is, but it sounds cool. The token expenditure index has fallen further since the first time we published this. It is possible token use is increasing, only if they are buying cheaper tokens.

The reality is that this chart may support my “trying but frustrated” view, that many people seem to be experiencing (at least anecdotally and in private conversations).

How many people out there are thinking:

• Finally!
• Whew, I’m really happy it isn’t just me!
• I told you so!

None of this means that AI and Data Centers are not useful. They are useful and will continue to see their usage and adoption grow. But…

• Are current growth expectations too high? That seems possible.
• Are valuations susceptible to changes in growth sentiment, coupled with market structure? Seems possible as well.

The voice of those questioning the current utility and cost of that utility, and therefore growth is likely to rise in the coming weeks, which would be a headwind for valuations.

Bottom Line

It is completely valid to have the following thoughts at the same time (at least I hope it is valid, because these are my thoughts):

• AI and Data Center usage will continue to grow.
• The onslaught of Cheap Chinese Compute is not good for profit margins of the providers, or the picks and shovels.
• Plans to build out compute may be far less feasible than previously thought, once again changing the profit margin outlook going forward.
• While useful, the cost to use AI has increased, and it is unclear that there is widespread belief that the cost vs benefit is truly there given today’s technology and prices. Investments in technology based on “Fear of Missing Out” may slow, if companies don’t fear they are missing out.
• The media narrative may shift rapidly too, giving their bears a bigger stage to express their concerns.
• The importance of leveraged ETFs in the AI Spend/Data Center space (add in Nasdaq 100, etc.) certainly helped propel stocks higher (which people seem to ignore), and it will amplify any sell-off (and already has).

From an investment standpoint:

• Buy bonds in the space! Spreads are wide. Given Credit Default Swap activity, one can assume there are decent short positions that have been built (always a nice catalyst if direction reverses). Money managers across the globe are “making room” to absorb the “certain onslaught of new issuance.” All it takes is for someone on the build side to “flinch.” Not even going on the full Debt Diet, but cautioning on how much they will spend how quickly. Maybe it won’t happen, but while we may not be facing a “perfect storm” for the space, there are a lot of risks that don’t seem fully priced in yet.
• Buy “completed projects.” Companies with projects that are completed or nearly completed will have a competitive advantage if we see any slowdown.
• Be cautious on equities in the space. All are great companies. Almost all fit into our ProSec™ narrative. But valuations may be questioned, and as we’ve seen in these markets, prices move fast when they start to move.

Iran is my biggest concern for Treasuries. Not just the energy price inflation, but also the need for countries to spend more on defense. Even in the Middle East, countries that once gobbled up Treasuries are facing their own economic slowdown, while seeing their need to spend increase. I should probably throw in the towel for rate cuts before hikes. But, and this remains a big but, if we do see spending on the AI / Data Centers slow at all, that will help on the inflation front and will put a question mark on jobs, as so much of what has been driving the economy on the positive side is related to the Capex spending in this area! Maybe the market will just put all these questions on hold, until the end of the summer, but it doesn’t have that sort of feeling!

Hope you are all able to get some vacation with friends and family this summer, while only having to keep one eye glued to markets!