AI's Capital Cycle: A Bull's Worst Nightmare

The last time I found a setup like this, our options returned over 600%.

It happened in South Korea.

I spotted an unsustainable combination of leverage, concentration, and deteriorating market mechanics before the indexes broke.

Now I’m seeing another setup worth paying attention to.

This time, it’s hiding inside the AI trade…

Just one week after I gave you my dispersion read, essentially every major volatility measure has collapsed to cyclical lows.

There’s little - if any - volatility left to sell.

With September approaching, I believe there’s a high probability of a VIX breakout.

Today, your job is simple:

I spent the weekend looking for an answer.

Not by guessing which stocks have run too far.

But by studying the capital cycle underneath the entire AI trade.

Here’s what I found:

That sort of explains why these memory stocks trade at such low forward P/E multiples after enormous rallies.

Maybe the market isn’t making a mistake.

Maybe it’s just looking past a peak earnings cycle.

AI’s Capital Cycle: A Bull’s Worst Nightmare.

Over the past 15 years, I put together all of the necessary data to come up with the following read.

Using regressions, ETF weightings, and other industry data, I was able to answer the question of:

It should be obvious that the extreme ultraviolet (EUV) and foundry monopolies like ASML and TSMC are the least affected given their market positioning.

On the other end,

DRAM/NAND memory players and neoclouds are the most affected during the cycle, considering their products are themselves a cyclical commodity and competition tends to come in waves at the peak profit cycle.

Which is what we’re seeing now…

Just in the past month, we’ve seen SK Hynix, Samsung, CXMT, Tower Semis, and dozens of other players jump on the memory wagon.

If you truly think this is a memory “super cycle” and that cyclicality will be no more,

Stop reading right now.

But if you realize that every commodity is by nature cyclical, that supply and demand are not fixed forever, then this will bring you to a massive trade opportunity, so let’s keep going.

Let’s take a look at the memory capital cycle through three very important lenses:

As you can see, the last proper cycle happened between 2014 and 2018 when it peaked.

Today, it seems we are roughly at the same extremes on the upside which inherently increases the odds of a turn lower under the right conditions.

Capex Intensity

Capex relative to D&A is now 1.76x compared to 1.8x in 2018.

What’s interesting about this measure is that, despite all the compute shortage that’s being touted by all these players…

They’re choosing to roll out relatively normal levels of capital expenditures.

Game theory would quickly lead you down a very logical conclusion:

This becomes especially important as most depreciation schedules (and charges) have been extended beyond reason.

Simply put,

Memory companies are assuming their chips and machinery will last for ~6 years on average, despite newer - more powerful - models making those current chips obsolete.

This is why their usual lifetime value is usually 2-3 years.

When you account for those reporting choices, the cycle has already peaked on a capex basis.

Margins

This is the outlier, as both gross and operating margins have reached new all-time highs for memory.

Because of the backlogs and lead time delays, I don’t expect these to come lower unless supply (or depreciation adjustments) hit them hard.

For now, you could say that margins haven’t peaked and therefore could continue to see higher levels to extend the cycle further.

Knowing that supply is the one to stop margin expansion in its tracks, we’ll cover that next.

Inventory Turnover

Inventory keeps churning along at less than 100 days on average for the memory players.

However,

When I dug into how these companies are reporting their levels of inventory, an extremely important detail came up:

Over 75% of all inventory is made up by work in progress (WIP) items.

Meaning,

When over $500 billion worth of chip supply comes online all at once, I suspect both margins and inventory turnover will mark the peak of the cycle.

At which time we should see the capex intensity slow down below ~1.2x on average for the sector.

I will likely post a follow up to the hyperscaler and data center economics after this.

For now,

Know that hyperscaler and data center inventories are quickly building up in terms of compute capacity.

The big spenders are carrying the highest capex intensity out of the group at 3.9x with data centers at 1.62x.

What’s becoming dangerous here is that this investment intensity combined with building capacity that remains unused…

Creates the perfect storm to send memory stocks to new 52-week lows by the time all of that WIP inventory comes online.

Think about those dots connecting for a second.

This is why I believe markets are beginning to treat DRAM names differently now.

Micron, SK Hynix, and SanDisk all trade at massive forward P/E discounts for the very reasons I have outlined above.

On the other end,

Foundries like AMD and Intel carry the most premium, a direct linkage to the ROIC and margin exposure multiples through the capital cycle.

Based on these findings, I am now working on a put calendar spread trade for a select memory name on this list.

Just like we played South Korea.

Join Offside Premium to get the trade structure and alert.

Signs of exhaustion are beginning to show up on the SMH ETF now.

A rising wedge with declining volume is happening right at the level of inflection I gave you a week ago.

If the September volatility kicks in, and it effectively drags these memory names lower, I suspect the rest of this ETF will join in the downside as well.

Because I have now landed on a high-conviction directional trade for AI, a proper hedge is required.

You’ve been exposed to some of my long/short equity ideas before, but now the stakes are much higher.

Option calendar spreads can be complex, unless you have the proper guidance on your side.

This is where the 1-1 support feature of Premium can help you.

Because whether SMH comes off this level is uncertain…

That’s why you want to get paid to wait and finance your entire position appropriately.

Meanwhile, here’s the latest from Jordi Visser (the retail authority on AI right now) talking about how AI has shifted time itself.

I remember when his views were more rooted in reality and numbers earlier this year. It looks like he’s now gone into the quantum realm, a great indicator of where sentiment could be for the entire trade:

Until next time,

OFFSIDE RESEARCH

Against the Tape, Ahead of the Curve.