The AI Trade is getting crushed...AXTI(-67%), AAOI(-56%), NBIS(-41%), SNDK(-38%) etc.

[DegenerateGambler]

The AI Trade got obliterated the last month, with leading names down a ton.  I wouldn't really call this a buying opportunity but it seems like the crash of the internet bubble in 2000 in the early stages.  A lot of these names still trade at >20 times sales and still has ways to go down.  

The question is is this just a pullback or is the AI trade effectively "over"?

[Gregmal]

When you look at what was driving the rally, who was driving it, and then the why and how, it all kinda lines up with a top. Qualitatively and gut feel wise too, I think Ive got it right. Q2 very much felt like Q1 2021 when the Covid beneficiaries and hyped SPAC stuff just began crumbling after it exhausted the last of time dumb money. And just the same as then, the “you can’t short this market” refrains were consensus whenever any sort of bearish sentiment arose.

[brobro777]

I don't know about other names but just looking at da action, Micron seems to be headed straight to $600, no bounce

 

I got no position in MU so whatever man but that's what it feels like for the memory names - straight down, no bounce

[Sinbius]

1 hour ago, DegenerateGambler said:

The AI Trade got obliterated the last month, with leading names down a ton.  I wouldn't really call this a buying opportunity but it seems like the crash of the internet bubble in 2000 in the early stages.  A lot of these names still trade at >20 times sales and still has ways to go down.  

The question is is this just a pullback or is the AI trade effectively "over"?

Well...at least you are coherent with your nickname...

[DegenerateGambler]

1 hour ago, Gregmal said:

When you look at what was driving the rally, who was driving it, and then the why and how, it all kinda lines up with a top. Qualitatively and gut feel wise too, I think Ive got it right. Q2 very much felt like Q1 2021 when the Covid beneficiaries and hyped SPAC stuff just began crumbling after it exhausted the last of time dumb money. And just the same as then, the “you can’t short this market” refrains were consensus whenever any sort of bearish sentiment arose.

yeah i don't see the Mag7 hyperscalers to keep increasing capex 30% per year for the next 5 years, they would need to start seeing return on their capital.  So the market is forward looking and started to sell these AI names that were trading at hundreds of times sales knowing capex won't go up forever.

[DegenerateGambler]

1 hour ago, brobro777 said:

I don't know about other names but just looking at da action, Micron seems to be headed straight to $600, no bounce

 

I got no position in MU so whatever man but that's what it feels like for the memory names - straight down, no bounce

memory is still probably cyclical so a low PE might signal the cycle top.  I would agree MU could tank a lot more here.

[Gregmal]

I mean I think the most awe inspiring aspect of this is that you can look back, 3-5 years and just chart "AI spend". Clearly, 2026 is a massive, massive FOMO binge, and 2027 likely will be too. Maybe even 28 as well. But in terms of a total capital expenditure cycle, what is insane and jaw dropping is seeing investors gamblers and such, citing what is clearly within an earshot of whatever one would call "peak", especially for crappy cyclical commodity stuff like chips/memory, and then taking THOSE numbers, and extrapolating valuations based on historic multiples. In what universe has it ever been a good idea to invest like that? "Ah, only 15x 2026 numbers"...like WTF? You expect 15 years of this sort of mania?

Edited 21 hours ago by Gregmal

[Peregrine]

8 minutes ago, Gregmal said:

I mean I think the most awe inspiring aspect of this is that you can look back, 3-5 years and just chart "AI spend". Clearly, 2026 is a massive, massive FOMO binge, and 2027 likely will be too. Maybe even 28 as well. But in terms of a total capital expenditure cycle, what is insane and jaw dropping is seeing investors gamblers and such, citing what is clearly within an earshot of whatever one would call "peak", especially for crappy cyclical commodity stuff like chips/memory, and then taking THOSE numbers, and extrapolating valuations based on historic multiples. In what universe has it ever been a good idea to invest like that? "Ah, only 15x 2026 numbers"...like WTF? You expect 15 years of this sort of mania?

 

Wall street estimates that SK, Samsung and Micron will earn more than $700 billion pre-tax next year. Apply a 70% pre-tax margin and that implies $1 trillion in total revenues - that's more than the ~$700 billion in total CAPEX from hyperscalers this year. People have completely lost all sense of math and common sense to rationalize this bubble.

[Gregmal]

33 minutes ago, Peregrine said:

 

Wall street estimates that SK, Samsung and Micron will earn more than $700 billion pre-tax next year. Apply a 70% pre-tax margin and that implies $1 trillion in total revenues - that's more than the ~$700 billion in total CAPEX from hyperscalers this year. People have completely lost all sense of math and common sense to rationalize this bubble.

Not to mentioned the degradation of the core FANG/MAG whatever stocks as they shift from net cash, high margin, asset lite companies to more debt burdened, asset heavy, capex beholden ones. 

[frommi]

The house of cards is getting its basement pulled away. Kimi now beats OpenAI and Anthropic models -> funding of them slows down/stops->no money for compute->revenue of hyperscalers slows down->Hyperscalers reduce Capex->less money for semis->Bubble pops. Mag7 down->Semis down->40% of the index in freefall.

[Spekulatius]

17 hours ago, Peregrine said:

 

Wall street estimates that SK, Samsung and Micron will earn more than $700 billion pre-tax next year. Apply a 70% pre-tax margin and that implies $1 trillion in total revenues - that's more than the ~$700 billion in total CAPEX from hyperscalers this year. People have completely lost all sense of math and common sense to rationalize this bubble.

Well, obviously Capex will have to increase to hit these numbers. If you do the math and think this through then a trillion $ in Capex needs at least $500B in revenue to sustain it - assuming $150-$200B in depreciation , a meager $100B in profit  (10% ROIC), and a 50% gross margin. It’s not going to come from coding and making slide decks or  marketing material and graphics.

Edited 3 hours ago by Spekulatius

[brobro777]

 

Hey a sale! Sandisk SSD discounted from $1459 to $599!

 

 

[Peregrine]

2 hours ago, Spekulatius said:

Well, obviously Capex will have to increase to hit these numbers. If you do the math and think this through then a trillion $ in Capex needs at least $500B in revenue to sustain it - assuming $150-$200B in depreciation , a meager $100B in profit  (10% ROIC), and a 50% gross margin. It’s not going to come from coding and making slide decks or  marketing material and graphics.

 

The same consensus estimates have hyperscaler CAPEX hitting $1 trillion next year. So all of it's gonna be memory spend? These guys are twisting themselves in knots trying to make up numbers that just don't make sense.

[gfp]

What dollar amount of memory will Apple buy?  Is that included as a "hyperscaler" ?  What about Dell, etc?