Question Zen 6 Speculation Thread

[bearmoo]

Intel already did all that back when they were cash rich (Swan years, 2018-2021).

Must be where their Capex went

 

[adroc_thurston]

Must be where their Capex went

Swan period was that and buybacks lmao.
But that's opex.
Pat funneled it all into new hires and fab capex instead.

 

[ETI4711]

That's because they built their own shitty module standard. LPDDR6 will come in a JEDEC-specified CAMM2 modules.

CAMM2 and LPCAMM2 ther was much talk. But who does use it?

JEDEC has already annouced this:

JEDEC’s SOCAMM2: Low Power Compact LPDDR5X Modules Poised to Power Next-Gen AI Servers | JEDEC

www.jedec.org

SOCAMM2 is on the road map of SK Hynix

 

[Doug S]

Right, on paper it looks like OpenAI and Meta effectively get their AMD GPUs for free, but AMD's stock price would go up 3x to accomplish that ($200 to $600). I don't think shareholders would be disappointed in that outcome.

The stock price may go up as long as the bubble holds but the fundamental valuation of the company isn't helped by a transaction where Meta pays you $100 billion and you give them $100 billion in product AND $100 billion in stock. That's dilutive, you're giving away a piece of the company on the bet that they will keep buying more after they've passed the threshold.

OK probably not truly dilutive at the time of the transaction, as there are some financial thresholds for AMD, presumably the stock will have to have gone up by enough that it isn't dilutive then, but all bets are off after that date. If AMD's stock falls below that treshold later (which it 100% for sure would if the AI bubble bursts) then it is a bad deal. Lisa Su and her team don't care, they will make a lot of money off this from the stock bump it creates (due to their bonuses and stock options) even if that turns to be shorter term in nature.

The idea as I said is that Meta will continue to buy from AMD after that threshold has hit, and that will generate the value that makes this worth it in real terms rather than "stock price of the moment" terms. But what guarantee does AMD have that Meta continue with purchases once they hit $100 billion? Since you made that deal with them once, they might pause their purchases after hitting the $100 billion milestone and have their hand out for another stock deal. What do you do then? Or even if they fully intend to continue purchases what happens if six months after hitting the threshold the AI hypecycle has peaked, and everyone realizes they've massively overbuilt and purchases dry up?

Deals like this are just another "bubble is getting frothier" indicator to me.

 

[Thunder 57]

The stock price may go up as long as the bubble holds but the fundamental valuation of the company isn't helped by a transaction where Meta pays you $100 billion and you give them $100 billion in product AND $100 billion in stock. That's dilutive, you're giving away a piece of the company on the bet that they will keep buying more after they've passed the threshold.

OK probably not truly dilutive at the time of the transaction, as there are some financial thresholds for AMD, presumably the stock will have to have gone up by enough that it isn't dilutive then, but all bets are off after that date. If AMD's stock falls below that treshold later (which it 100% for sure would if the AI bubble bursts) then it is a bad deal. Lisa Su and her team don't care, they will make a lot of money off this from the stock bump it creates (due to their bonuses and stock options) even if that turns to be shorter term in nature.

The idea as I said is that Meta will continue to buy from AMD after that threshold has hit, and that will generate the value that makes this worth it in real terms rather than "stock price of the moment" terms. But what guarantee does AMD have that Meta continue with purchases once they hit $100 billion? Since you made that deal with them once, they might pause their purchases after hitting the $100 billion milestone and have their hand out for another stock deal. What do you do then? Or even if they fully intend to continue purchases what happens if six months after hitting the threshold the AI hypecycle has peaked, and everyone realizes they've massively overbuilt and purchases dry up?

Deals like this are just another "bubble is getting frothier" indicator to me.

Certainly seems to dilute the stock IMHO but I am far from an expert. AMD giving away so much stock (20%) in recent deals seems like a real liability. It isn't guaranteed of course but still seems unwise and I would expect better from Lisa. But, as you said, she will make well with this deal.

 

[Saylick]

The stock price may go up as long as the bubble holds but the fundamental valuation of the company isn't helped by a transaction where Meta pays you $100 billion and you give them $100 billion in product AND $100 billion in stock. That's dilutive, you're giving away a piece of the company on the bet that they will keep buying more after they've passed the threshold.

OK probably not truly dilutive at the time of the transaction, as there are some financial thresholds for AMD, presumably the stock will have to have gone up by enough that it isn't dilutive then, but all bets are off after that date. If AMD's stock falls below that treshold later (which it 100% for sure would if the AI bubble bursts) then it is a bad deal. Lisa Su and her team don't care, they will make a lot of money off this from the stock bump it creates (due to their bonuses and stock options) even if that turns to be shorter term in nature.

The idea as I said is that Meta will continue to buy from AMD after that threshold has hit, and that will generate the value that makes this worth it in real terms rather than "stock price of the moment" terms. But what guarantee does AMD have that Meta continue with purchases once they hit $100 billion? Since you made that deal with them once, they might pause their purchases after hitting the $100 billion milestone and have their hand out for another stock deal. What do you do then? Or even if they fully intend to continue purchases what happens if six months after hitting the threshold the AI hypecycle has peaked, and everyone realizes they've massively overbuilt and purchases dry up?

Deals like this are just another "bubble is getting frothier" indicator to me.

All valid and fair points. Yes, there's nothing that guarantees that AMD stock stays elevated even after the last tranche is paid out and it would be catastrophic if that drop was initiated by a pop in the AI bubble. Like you say, all of these deals, irrespective of whoever is making them, are dependent on this AI fueled mania to continue long enough for the deals to reach fruition.

 

[Khato]

It isn't guaranteed of course but still seems unwise and I would expect better from Lisa. But, as you said, she will make well with this deal.

I certainly hope that Lisa doesn't follow in the footsteps of certain other CEOs who were once similarly respected for their accomplishments before going off the deep end... But just how much she's promoting AI internally with the same manner of talk used to convince investors is not a good sign.

On the other hand, these deals are likely the only option AMD was able to negotiate in order to get their foot in the door. Otherwise regardless of how good their hardware is the great wall of NVIDIA software would remain intact.

 

[Saylick]

One thing I realized is that each tranche of stock warrants require both delivery and stock price milestones to be met in order to pay out for OpenAI/Meta. There is a scenario where AMD delivers 6MW of GPUs to OpenAI and/or Meta, thus satisfying the first criteria, but if the stock price fails to rise in concert with the purchases, thus not satisfying the second criteria, there's no dilution. In such a scenario, AMD gets paid for the GPUs without giving out a piece of the company. Of course, such a scenario where AMD records the revenue but the stock price fails to rise is unlikely, but just pointing out that the market can be irrational and it doesn't always price things according to the underlying fundamentals.

 

[adroc_thurston]

One thing I realized is that each tranche of stock warrants require both delivery and stock price milestones to be met in order to pay out for OpenAI/Meta. There is a scenario where AMD delivers 6MW of GPUs to OpenAI and/or Meta, thus satisfying the first criteria, but if the stock price fails to rise in concert with the purchases, thus not satisfying the second criteria, there's no dilution. In such a scenario, AMD gets paid for the GPUs without giving out a piece of the company. Of course, such a scenario where AMD records the revenue but the stock price fails to rise is unlikely, but just pointing out that the market can be irrational and it doesn't always price things according to the underlying fundamentals.

It's just a way for AMD to keep customers very loyal despite mountains of FCF NV is sitting on.
OpenAI is unlikely to deploy beyond 1MW even, they're hardly solvent at this point.

 

[RnR_au]

OpenAI is unlikely to deploy beyond 1MW even, they're hardly solvent at this point.

Anthropic, Google and local AI are eating their lunch. Its also why OpenAI have talked about ads and an adults only model.

 

[Joe NYC]

It's just a way for AMD to keep customers very loyal despite mountains of FCF NV is sitting on.
OpenAI is unlikely to deploy beyond 1MW even, they're hardly solvent at this point.

Here is an interesting possible outcome:

Up through 2026, CPU server, client etc. paid for OpEx, such as R+D etc for the data center GPU division. That put a cap on investment that could go into client etc.

Starting Q4 2026, datacenter GPU can more than pay for their own OpEx. So perhaps, the CPU and client come out of starvation mode.

 

[adroc_thurston]

So perhaps, the CPU and client come out of starvation mode.

starvation mode?

 

[Joe NYC]

starvation mode?

Let's say "underinvestment mode"

 

[adroc_thurston]

Let's say "underinvestment mode"

Underinvestment mode?

 

[MadRat]

MoP is nightmarishly margin dilutive and is never ever at AMD.
Forget about iti.

Why use moly interconnects for adding traces in the CPU's package?

 

[adroc_thurston]

Why use moly interconnects for adding traces in the CPU's package?

?

 

[MadRat]

?

What did you take his abbreviations to mean?

 

[soresu]

i love venice so much bros...

Indy agrees... 😂🤣

 

[coercitiv]

Indy agrees... 😂🤣

Don't you guys have Venice at home?

 

[soresu]

Don't you guys have Venice at home?

View attachment 138802

Ye I've brought that up before - I think that was the gen just before my first AMD chip, the Athlon X2.

 

[yuri69]

Certainly seems to dilute the stock IMHO but I am far from an expert. AMD giving away so much stock (20%) in recent deals seems like a real liability. It isn't guaranteed of course but still seems unwise and I would expect better from Lisa. But, as you said, she will make well with this deal.

In my (deeply negative) PoV it seems like AMD still desperately needs somebody to buy their AI accelerators at meaningful scale. One would thought that a viable AI product would sell itself at times of an AI craze...

 

[gdansk]

In my (deeply negative) PoV it seems like AMD still desperately needs somebody to buy their AI accelerators at meaningful scale. One would thought that a viable AI product would sell itself at times of an AI craze...

Well, their earnings show they do sell. But the market expects them to hockey stick like Nvidia did. They're not Nvidia. Dumb expectations, dumb deals. It's the AI craze.

 

[ETI4711]

In my (deeply negative) PoV it seems like AMD still desperately needs somebody to buy their AI accelerators at meaningful scale. One would thought that a viable AI product would sell itself at times of an AI craze...

No product is selling itself.

 

[coercitiv]

In my (deeply negative) PoV it seems like AMD still desperately needs somebody to buy their AI accelerators at meaningful scale. One would thought that a viable AI product would sell itself at times of an AI craze...

IMHO the big datacenter spenders probably tried to use AMD and others as bargaining chips to get sweeter deals from Nvidia (better prices, volumes etc). This might have backfired because Nvidia is very... vindictive, so now the likes of Meta and OpenAi have to make it work with AMD while their Nvidia chip allocation is "suboptimal".

Spoiler

 

[tajoh111]

IMHO the big datacenter spenders probably tried to use AMD and others as bargaining chips to get sweeter deals from Nvidia (better prices, volumes etc). This might have backfired because Nvidia is very... vindictive, so now the likes of Meta and OpenAi have to make it work with AMD while their Nvidia chip allocation is "suboptimal".

Spoiler

View attachment 138834

Nvidia is fortunate to be in the position where they have buyers lined up.

Nvidia earnings and outlook make it look like AMD is stuck in the mud.