Semiconductor Growth Forecasts Are Fantasy … Correction Is Inevitable

[mgp-1]

Indeed! Of course they have a point, that the current pricing environment will not last forever, but their reasoning is rather peculiar.

As an example, take this nonsensical sentence: “Even more disturbingly, growth is being driven by ASPs, not unit demand. This, we believe, makes the current growth rates untenable and a downstream correction inevitable.”

I simply cannot get my head around what they are trying to say. Do they think the price rises without a demand increase in an industry, where you usually do not take supply out, which would indeed be really strange?

More likely, we are just witnessing a simple supply-demand gap, and the question becomes, how long does it take for supply and demand to rebalance. But the authors do not discuss this issue at all.

Furthermore, in their demand arguments they focus on AI and forget the other demand drivers at play. Traditional demand drivers like consumer electronics, phones and pc are subdued and will likely continue to be so for a while due to the semi-led price increases. But new price-inelastic demand catgories are emerging, which are easily overlooked when compared to the gargantuan AI-driven demand (likewise inelastic). Take FPV and autonomous military drones for example. The yearly amount of memory being burnt in the Russia-Ukraine war is staggering, but the semi price increases do not change the fact that drones offers the most cost-efficient way of killing the opponent, so the warring parties will keep buying.

ASPs rise in a shortage situation, and collapse when supply catches up with demand. The long-term ASP growth is ZERO! Moore's second law stated "IC ASPs eventually reach $1" but no-one liked that one!

 

[mgp-1]

Hi everyone ... thanks for the interesting discussion. We go into a LOT more detail and analysis in the full report; this extract was just a teaser. DM me if you want more details

 

[count]

ASPs rise in a shortage situation, and collapse when supply catches up with demand. The long-term ASP growth is ZERO! Moore's second law stated "IC ASPs eventually reach $1" but no-one liked that one!

Good thing Moore's law is dead

 

[Ole Hoejlund]

Something which is maybe 1% or less of the market is not a demand driver in any possible meaning of the word... ;-)

Military/defence has not been a demand driver for a *very* long time, if ever -- it's a follower which takes advantage of the technologies developed (and paid for) by the real demand drivers, which have shifted from PC to mobile and now to HPC/AI.

But even so it doesn't have enough volume/revenue to justify developing custom chips in the very expensive bleeding-edge nodes, things like drones use technology from a few nodes back which is perfectly capable of doing the job and far quicker/easier/cheaper to develop.

Non-drone military applications like traditional planes/missiles/munitions/radar take so long to develop and are needed in such small volumes they're not even a pimple on the face of military demand... ;-)

You are spot on looking at the market right now and historically. But I’m not sure you will be right going forward.

Disclaimer: the following numbers are AI-generated and I do not have the time right now to do the necessary, but tedious fact-checking. Though, the words and opinions are mine.

The wars in Ukraine, Nagorno-Karabahk, Syria and Iran has made it is obvious that large scale drone saturation will be an integral part of warfare going forward.

Based on the current run-rate production of just shy of 10 million drones per year by the warring parties and a per unit DRAM-content of 0.5-4.0 GB, you’re looking on a burn of up to 0.1 pct. of the 2025 bit supply.

That’s not interesting in itself. What is interesting is first, that all frontline states and quite a few on top of that are actively looking at building scaled military drone capabilities across surveillance, deep/mid/short range strike, counter drone etc. including supply depth. We will probably witness quite a build up in places like China, Japan, Taiwan, Indonesia, Pakistan, India, most of the middle east, central/eastern Europe etc. And it’s not only a significant build-up. When you reach scale, there will be attrition due to training, obsolescence etc.

Second, developments in electronic warfare are forcing several drone capabilities towards a higher degree of autonomy, which raises the memory content per unit quite a bit to handle video buffer, map, AI models etc.

My gut feeling is that we see a demand driver of app. 1-2+ pct. of 2025 bit supply in the coming years. In my eyes, this is not inconsequential in a situation where we are witnessing a severe supply gap.

And this is just one use case. We have additional new demand drivers growing in parallel and that comes on top of the gargantuan AI/LLM demand. Not large in and by themselves, but they are typically price inelastic and they arrive at the same time in a very tight market.

Of course, this is pure speculation with several uncertainties.

 

[Ole Hoejlund]

ASPs rise in a shortage situation, and collapse when supply catches up with demand. The long-term ASP growth is ZERO! Moore's second law stated "IC ASPs eventually reach $1" but no-one liked that one!

Totally agree on that.

I just don’t think that is the interesting question. The question is this: is the current demand uptick and structural supply gap of a magnitude that will change the market dynamics for a longer period? In this period memory and compute may become a strategic, price inelastic enabler in short supply and leave the commodity market dynamics of yesterday. It will eventually return to a more price sensitive commodity-like behavior again, but it is certainly not the case now and how long will this continue?

As I read you, you consider the current situation a short-lived deviation. My take is that you’re so focused on Yesterday, that you overlook the current dynamics and signs that something different may be at play.

Another demand driver: shift to EV and significant growth in memory content per unit of both EVs and BEVs will probably drive related memory demand from 5-6 pct. of global bit supply in 2025 to up to 15 pct. of 2025 bit supply in 2030. Again, this demand is rather price inelastic compared to the historical demand and the supply side including WFE is not geared to handle growth rates of that order on top of AI.

Disclaimer: non-validated AI-generated numbers which may not hold up to scrutiny.

 

[bilau]

ASPs rise in a shortage situation, and collapse when supply catches up with demand. The long-term ASP growth is ZERO! Moore's second law stated "IC ASPs eventually reach $1" but no-one liked that one!

Micron guided FQ3 GM to 81%!

At those GM levels, it makes zero sense for the suppliers to let supply catch up with demand as that would not grow, but shrink the total profit.

In fact, Mike Murphy reiterated, or telegraphed to his cohorts if you are cynical like me, that they will modulate the pace of adding new capacity.

If it were me, I would modulate the bit growth to be just below the demand grow to maintain the gap.

So, any supply/demand balance must come via demand destruction?

 

[hist78]

Except that every military application of semiconductors put together -- drones in Ukraine included -- is orders of magnitude smaller than the demand for AI datacenter applications...

Apple alone shipped between 240.6 million and 247.4 million iPhones globally in 2025.

By comparison, in 2025 US defense contractors produced 191 F‑35 fighters, about 100 Tomahawk cruise missiles, 2,400 Javelin anti‑tank missiles, two submarines, and launched and delivered more than 100 military satellites into orbit.


There is a 2021 semiwiki.com article talked about this situation.

"The commercial market demands on the semiconductor supply chain dwarfs market demand from the DoD. With not many, if any, incentives for suppliers to support DoD production needs, availability and access to manufacturing capacity is low. "

~ Dr. Yadunath Zambre, chief microelectronics technology officer at the Air Force Research Laboratory (AFRL).

Keynote from Air Force Research Laboratory at CadenceLIVE Americas 2021 - Semiwiki

Cadence hosted its annual CadenceLIVE Americas conference June 8th-June 9th. Four keynotes and eighty-three different talks on various topics were presented. The talks were delivered by Cadence, its customers and partners. The C-suite keynotes were delivered by Lip-Bu Tan (CEO) and Dr. Anirudh...

semiwiki.com

 

[Ole Hoejlund]

Micron guided FQ3 GM to 81%!

At those GM levels, it makes zero sense for the suppliers to let supply catch up with demand as that would not grow, but shrink the total profit.

In fact, Mike Murphy reiterated, or telegraphed to his cohorts if you are cynical like me, that they will modulate the pace of adding new capacity.

If it were me, I would modulate the bit growth to be just below the demand grow to maintain the gap.

So, any supply/demand balance must come via demand destruction?

Yes, and I think there has been a lot of signaling that support that take on the DRAM market. But CXMT are currently in the middle of an aggressive build out of wafer name plate capacity, and SK Hynix announced a path to tripling in 2034, so the discipline may be eroding. Sometimes greed is good and benefits the masses.

 

[Barnsley]

Micron guided FQ3 GM to 81%!

At those GM levels, it makes zero sense for the suppliers to let supply catch up with demand as that would not grow, but shrink the total profit.

In fact, Mike Murphy reiterated, or telegraphed to his cohorts if you are cynical like me, that they will modulate the pace of adding new capacity.

If it were me, I would modulate the bit growth to be just below the demand grow to maintain the gap.

So, any supply/demand balance must come via demand destruction?

Fake supply issues.

You have to believe these guys are loving these numbers.

They have ZERO incentive to address "supply issues"