News AMD 4Q23 Earnings

[branch_suggestion]

And now specifically on AMD, Q1 is clearly rock bottom, every market declining partially offset by MI300 ramping.
Every quarter afterwards should be sustained growth from the Q1 baseline, H2 has AMD pulling an enormous number of levers with the only resistance being H200, which is restricted by the very limited HBM3e bin supply that won't fully ramp until 2025.
Zen 5 in general is a colossal win that should force a big upgrade cycle, even in the face of DCGPU demand.
RDNA4 should help inject new life into gaming rev as it faces no direct price competition until 2025. PS5 Pro should launch late 2024 so is really a 2025 story.
Embedded should recover gradually so that is nice for the bottom line.
MI300 should have significant growth rate per quarter, potentially above 50% that should lead to a lot of momentum across the whole company.
I can see AMD breaking the 10B/quarter mark by Q42024 if things go well.

But, there is that b word. To be honest, a bubble is inevitable, but the really scary scenario is a global recession.
Good thing is, a complete cratering in demand won't hurt AMD nearly as much as NV or especially Intel.
AMD has similar revenue diversity to Intel, they have far less collateral from a demand crash than either and can reboot to healthy capacity with the least risk.
NV has enough cash to weather the storm, but that capacity they have acquired might take years to get back to good utilisation, and all that inventory/prepaid capacity they would need to write down.
Intel would be absolutely doomed though, they have no leadership products and their fabs would be dead weight.
All in all, once the current fad has died down, AMD will be well positioned for the next fad, NV won't have first mover advantage this time.
NV can choose to maintain share and tank margins with price cuts and MDF, or maintain margins at all costs. They won't get to have both now, those days are finished.

 

[adroc_thurston]

Microsoft musta sold a lot less XBoxes.

Sony.
Demand is leveling off since PS5 spent some time under a supply duress and was overselling after.

 

[yuri69]

I've read many stating the 96c of Genoa should make AMD rich in 2023-2024. Seems not.

Putting any faith in RDNA4 after the RDNA3 fiasco followed by rumors cancelling some SKUs from the upcoming lineup...

MI300 is very complex. There might be manufacturing/logistical bottlenecks, right?

 

[branch_suggestion]

I've read many stating the 96c of Genoa should make AMD rich in 2023-2024. Seems not.

There will be plenty of supply when they pull the trigger, and it will be sold as it is made if it is as big a leap as hinted.

Putting any faith in RDNA4 after the RDNA3 fiasco followed by rumors cancelling some SKUs from the upcoming lineup...

I'm hoping for a lot of volume this time, specifically in laptop, it is hyperfocused on that market now and has first movers advantage. AMD cannot screw up twice.

MI300 is very complex. There might be manufacturing/logistical bottlenecks, right?

Defect rate seems to be a non-issue, all about ramping the supply chain and letting all the people with money know that it just works.
Speaking of, AMD has made plenty of sales/supply chain side changes, they really want to make the most of this.

 

[moinmoin]

the expected numbers are surprisingly low for some reason

Did you look at the QA notes/transcriptions so far? The stated numbers are specifically committed bookings. So from the AI event in December to the call now the committed bookings increased from 2b to 3.5b revenue wise. That only sounds low for people who misread it. Supply doesn't seem to be the limit, ramp and demand seem.

AMD had virtually zero share until after MI100, that was just there to establish the CDNA chain for the future, MI200 was the first class leading part, but only in HPC but supplying those contracts certainly helped prepare for higher demand with a more general purpose accelerator, which is what MI300 is.

One thing to keep in mind is that the whole MI line only exists as it was a requirement to fulfil the contracts for Frontier (MI250) and El Capitan (MI300A) exascale supercomputers. If it weren't for DARPA's several Forward sponsoring in the past decade those wouldn't have happened at all. If AMD didn't design MI300 to be flexible MI300X wouldn't exist.

 

[jpiniero]

Sony.
Demand is leveling off since PS5 spent some time under a supply duress and was overselling after.

Sony hasn't reported yet but MS apparently said that XBox sales in the quarter were below their expectations.

 

[moinmoin]

Sony hasn't reported yet but MS apparently said that XBox sales in the quarter were below their expectations.

Sony is coming from a higher high, Xbox is a smaller part of the game console market. But yeah, both don't do too well currently.

 

[Thunder 57]

I've read many stating the 96c of Genoa should make AMD rich in 2023-2024. Seems not.

Putting any faith in RDNA4 after the RDNA3 fiasco followed by rumors cancelling some SKUs from the upcoming lineup...

MI300 is very complex. There might be manufacturing/logistical bottlenecks, right?

AMD has been killing it in servers for awhile. RDNA4 is not targeting the high end. Seems your only goal is to bash AMD.

 

[jpiniero]

Sony is coming from a higher high, Xbox is a smaller part of the game console market. But yeah, both don't do too well currently.

Sony did outsell the previous year in the last 2 quarters...

 

[Vattila]

Here is my graphical overview of AMD's business segment results over the last couple of years:

(Full data at my OneDrive.)

The collapse in the Client segment since 2022-Q3 is remarkable. In the 5 prior quarters, the segment generated a total of $2.9B in operating income. Since then it has lost $0.2B in total. However, there is some glimmer of hope that the segment will improve this year:

"We are aggressively driving our Ryzen AI CPU roadmap to extend our AI leadership, including our next-gen Strix processors that are expected to deliver more than three times the AI performance of our Ryzen 7040 series processors. Strix combines our next-gen, Zen 5 core with enhanced RDNA graphics and an updated Ryzen AI engine to significantly increase the performance, energy efficiency and AI capabilities of PCs. Customer momentum for Strix is strong with the first notebooks on track to launch later this year. Looking at 2024, we are planning for the PC TAM to grow modestly year on year, weighted toward the second half as AI PCs ramp. We continue to see strong growth opportunities for our client business as we ramp our current products, extend our AI PC leadership and launch our next wave of Zen 5 CPUs. [...] For 2024, we expect the demand environment to remain mixed, with strong growth in our Data Center and Client segments, offset by declines in our Embedded and Gaming segments."

2024-Q3 Earnings Call transcript

 

[Khato]

Regarding claims of there being no issues with supplying as much MI300 as customers want... Last I checked NVIDIA H100 supply is constrained by TSMC CoWoS capacity. How exactly is MI300 not going to also be constrained by CoWoS capacity? Not like NVIDIA's going to cancel their reserved capacity and gift it to AMD just because customers want to buy MI300 instead of H100.

 

[adroc_thurston]

How exactly is MI300 not going to also be constrained by CoWoS capacity?

AMD actually works with OSATs like ASE.

just because customers want to buy MI300 instead of H100.

That they do lmao.

 

[Khato]

AMD actually works with OSATs like ASE.

If you want to claim that availability of additional advanced packaging services outside of TSMC is going to result in an end to that being the bottleneck, fair enough. The implication that AMD works with OSATs while NVIDIA won't though? Thanks for a good laugh.

 

[adroc_thurston]

The implication that AMD works with OSATs while NVIDIA won't though?

Yeah they've been doing that since 2015. Xilinx, even earlier.

 

[moinmoin]

Yeah they've been doing that since 2015. Xilinx, even earlier.

I think AMD's proficiency with packaging technology went under the radar for many. It's wasn't by chance that AMD went big with MCM and chiplets that early even while going through financially trying times. And Xilinx is one of the few companies being even more adventurous.

 

[AtenRa]

They gotta play safe assuming NV retaliates heavily in cloud by chopping prices off H100/H200 to regain cloud share.

NVIDIA is wafer capacity constrained , they will not lower prices, in fact they may as well increase prices since they will sell every chip they will produce.

 

[adroc_thurston]

NVIDIA is wafer capacity constrained

No.

they will not lower prices,

That's what lost them marketshare at Meta and MS.

in fact they may as well increase prices since they will sell every chip they will produce.

It seems you don't understand how the market works.

 

[jpiniero]

That's what lost them marketshare at Meta and MS.

But if they are really that packaging limited, cutting prices won't really help NV.

And I am still of the opinion that AI is just a fad that will go away once Wall Street stops losing it's mind over anything AI.

 

[adroc_thurston]

But if they are really that packaging limited, cutting prices won't really help NV.

Yes it will, losing key hyperscale customers is bad vibes.

And I am still of the opinion that AI is just a fad that will go away once Wall Street stops losing it's mind over anything AI.

It's a bubble but it's not a fad.
More dotcom bubble than NFT .png sellers.

 

[tajoh111]

They're gonna smash that.

She has to be, it's a new market and NV still has levers to pull.

yea, that's why MI300 is eating GPGPU share to begin with.
AMD has a roadmap, no one else but NV does.

The question is, what roadmaps will slip to keep AMD's most important roadmaps on track or accelerated? This is the most pragmatic choice with the smallest gamble.

From AMD's Q1 projections and 2023's overall revenue and profit, AMD doesn't quite have the resources to keep every road map on track. Mi300 was supposed to be a mid 2023 product, not a 2024 product. So does AMD make cuts or make a gamble and takes some heavy losses to put more spending into R and D, hiring more engineers and prebooking wafers at TSMC to keep their priority.

With Nvidia having the cashflow and money to dump into R and D for Data center(which they are clearly doing with the accelerated roadmap), Lisa if anything should be optimistic with datacenter projections because AMD needs to go on a hiring spree because current revenue cannot justify it. If AI is a bubble every month counts because if they gravy train is over in 2025 or 2026, AMD doesn't have any time to waste.

 

[adroc_thurston]

The question is, what roadmaps will slip to keep AMD's most important roadmaps on track or accelerated?

yea.
The question is funny given that every NV product for 2022/2023 timeframe slipped.
Some are outright dead (let's play where's Waldo with infiniband).

From AMD's Q1 projections and 2023's overall revenue and profit, AMD doesn't quite have the resources to keep every road map on track.

Yes they do their R&D spend is at $1.5B now.

Mi300 was supposed to be a mid 2023 product, not a 2024 product

It was a Q1'23 product, chefe.

and prebooking wafers at TSMC to keep their priority.

You don't do that, TSM has spare capacity aplenty.

With Nvidia having the cashflow and money to dump into R and D for Data center(which they are clearly doing with the accelerated roadmap)

They 'accelerated' roadmap at the tiny little cost of insane work-hours at their hweng galley.
It's all comp pressure stuff.

 

[moinmoin]

Mi300 was supposed to be a mid 2023 product, not a 2024 product.

MI300 reportedly has a delivery lead time of 7 to 8 months. First MI300A were produced since that's required for El Capitan. MI300X production and demand is still ramping.

Also it bears repeating that the 3.5b number is not a projection but committed bookings, mainly due to said delivery lead time.

 

[jpiniero]

AMD posted the 10-K... Sorny was 18% of AMD's revenue for the whole year (4.1B). They were 16% last year (3.8B)

Note that's the whole year.

 

[branch_suggestion]

AMD posted the 10-K... Sorny was 18% of AMD's revenue for the whole year (4.1B). They were 16% last year (3.8B)

Note that's the whole year.

Gonna guess MS will usurp them this year.

 

[positivedoppler]

Yes it will, losing key hyperscale customers is bad vibes.

It's a bubble but it's not a fad.
More dotcom bubble than NFT .png sellers.

I don't think AI will become a footnote in history like buy.com, egg.com, onsale.com, etc... There is no ceiling on the demand for intelligence. At no point in time can I see a future in which tech companies stop advancing AI because machines are smart enough.