Dipping after hours

So… apparently the market is unimpressed by AMD’s guidance.

  • EPS: 70 cents adjusted, versus expectations of 68 cents per share
  • Revenue: $5.8 billion, versus expectations of $5.7 billion

For the fourth quarter, AMD said it expects about $6.1 billion in sales, while analysts were looking for revenue of $6.37 billion.

Net income in the third quarter rose to $299 million, or 18 cents per share, from $66 million, or 4 cents per share a year ago. Revenue increased 4% from $5.6 billion a year earlier.

Revenue in AMD’s Client group, which includes sales from PC processors, rose 42% on an annual basis to $1.5 billion, driven by PC chips.

Data center, which includes AMD’s server processors and AI chips, reported $1.6 billion in sales, flat from a year earlier. AMD said its sales of server CPUs grew. AMD also said that it expects strong growth in its data center business in the fourth quarter.

AMD said its forthcoming AI chips, the MI300A and MI300X, are “on track” for volume production in the current quarter.

“Our data center business is on a significant growth trajectory based on the strength of our EPYC CPU portfolio and the ramp of Instinct MI300 accelerator shipments,” AMD CEO Lisa Su said in a statement, highlighting the company’s AI business.

AMD’s embedded segment revenue declined 5% to $1.2 billion, which the company blamed on a weak communications market. That includes parts for networking as well as the company’s FPGA unit that it acquired when it bought Xilinx.

Sales in AMD’s gaming segment declined 8% to $1.5 billion, because of fewer “semi-custom” chip sales. That’s what the company calls its business that makes processors for consoles like Sony’s PlayStation 5.


I would call this a really solid performance by AMD, but AMD is priced for more than “really solid”. The tepid Q4 guidance didn’t help. I would ignore the afterhours price as it seems the actual morning open can be significantly different.


So client sales rose 42%, but data center sales were flat (yet server CPU sales grew somehow)? Isn’t that client performance surprising? And the server side looks contradictory to me (flat sales, but CPUs grew). Help me to understand this.


They sell a lot of things to the data center now – the Xilinx products, Pensando networking offerings, the existing GPU offerings which probably have some presence… It would help to know what’s working and what’s not. (I hear some things about how NVidia and Intel’s networking stacks up against Pensando, we’ll see if that move plays out well in the end.)

1 Like

94 last night, 108 today? Maybe the Street had more time to think about how the AI story may play out, but something sure changed in the last 24 hours… :stuck_out_tongue:

I managed to make a tiny buy last night (30 shares at 94.81)-- Wish I had had more cash available! I had also recently bought a somewhat larger chunk at 94.05, and more around 101 (wishing I had held off on that until the mid-90s).


So client sales rose 42%, but data center sales were flat (yet server CPU sales grew somehow)?

The new datacenter GPUs (Instinct MI300) were to start shipping this quarter, and if they weren’t shipping now AMD would have said something. But it is no surprise that orders grew last quarter, but shipments were steady. Note that if a customer ordered a system from one of the server manufacturers with EPYC CPUs and Instinct MI300 GPUs, that revenue, even if the CPUs were shipped to the system vendor, would not get booked until the server manufacturer (shipped and) billed the system to the end user. Technically, the revenue is recognized when the CPU or GPU is put into the hardware. However, for large systems, the hardware will be put together on-site.

Do I expect lots of MI300 sales in the next few quarters? Sure. And every MI300 sale will also be an EPYC sale. Yes, MI300s can now share (main) memory with Intel CPUs. However, MI300s can be cache-coherent with EPYC CPUs. (Technically, the sharing is of L2 and L3 caches, but L1D data is duplicated in the L2 cache.)

Hmm. Sharing data aggressively is tricky. If the CPU writes to memory, the data will go into a write pipe then into L1D cache. The L2 cache entry will be invalidated and then written. If the same cache line is written again while the data is still in the write pipe, it will be updated there. You need to have a synchronization point correctly placed to avoid creating garbage. This is on a cache line (64 bytes) basis. Yes, you can create spin locks between CPU and GPU, but that is way off topic for here. The part that is relevant here is that end users will not do this. Database vendors and other (AI?) software vendors will. When benchmarks use these libraries (and after they are created ;-), the results may show significant improvements.

So you should feel fine about 113 today and take a quick profit?

1 Like

I sold a little… I will probably sell a little more over the next couple of weeks… maybe sell 100 shares total. Iif we break 120 I’d sell more.

I have realized finally that if money is just sitting in cash in my brokerage account it’s getting nearly 5%. Honestly that’s not a bad return for doing absolutely nothing.

1 Like

Hindsight being 20/20 I wish I would have bought some calls when it was $94 so I think Caromero1965 did great since its now over $113 today…doc

1 Like

Not to be a ticker but now we’re at almost $120 :slight_smile: Either a top or a breakthrough, what say you all?