DC approvals are being granted on the ~basis that the desired locales grid capacity, is so constricted, that approval is both delayed and of limited MW

Relative power efficiency is no longer a preference, but an absolute must. Supply of cloud instances is so constrained that rationing is in place. Cloud instance providers must chose the hardware which yields the most per MW.

I don’t think for a second that Sapphire Rapids and Emerald Rapids can compete well in terms of performance or price with AMD’s Genoa or Bergamo, and then Turin is right there to take on Granite Rapids. AMD will sell as many CPUs as it can make; Intel will get what is left until it can deliver better bang for the buck and better thermals.

https://www.theverge.com/23294064/intel-deny-meteor-lake-delay-2023-2024

and TSMC:

https://www.tomshardware.com/news/tsmc-comments-on-intel-n3-orders-report

This is in addition to the 10nm 60 core (up from 56 core) Xeon chips launching in 1Q2023.

So please justify the “2024-2025” thing?

Intel is still not above huge mistakes. I mentioned earlier that picking TSMC 4 over Samsung 6 was a mistake for their GPUs, and just this week they acknowledged that they were only going to make $1 billion in GPU revenues this year because they tried to use the same drivers on their discrete GPUs as they did their integrated GPUs and will now have to delay shipping GPUs because they need to go back and write new drivers from scratch. Still, no one has provided actual evidence that they are having problems with their 7nm process. With the 14nm and 10nm processes, Intel had already announced problems by now. Granted there are still two big hurdles to clear – yield manufacturing then validating the result – but no one is going to know the results of either for weeks at the earliest (Intel stated that mass production on Intel 4 would start “2H 2022”.)

Even with that very large brain of hers (BS+MS+PhD in EE at MIT) I think it will take Lisa Su more than one day to count all of this! More so in upcoming quarters as the Xilinx+Pensando get paid off (if I read you well). Pat Gelsinger (MS in EE at Stanford) on the other hand is known for huuuge industry experience (eg. 486 design) and so the high-tech “boxing bout” between the red corner and the blue corner should prove fascinating (much to our benefit in terms of ever better hardware). Not to be left out, in the other blue corner, Arvind Krishna (PhD in EE at NCSA pioneer UIUC) is sure to keep swinging hard as well … after all, Summit beats LUMI in HPCG (while LUMI beats Summit in HPL) … great times ahead! With these three, it is hard-work all around (none of the MBA spin).

In November when AMD was at $160/share, they topped Intel then too. Then price halved, but then they increased shares 25% when Xilinx rolled in. So back to topping Intel as Intel’s price has gone down. Now, it’s investors paying up for growth in AMD, and discounting Intel’s shrinking revenues/profits. If you really want to see ‘stretched’, check out Tesla vs. all other car makers, and Nvidia.

I’m sure analysts will start to pester AMD about datacenter product splits again as Xilinx FPGAs, Pensando DPUs, and Instinct GPUs gain traction in DC. Right now all but about $150-250m is Epyc.

I meant for this one day only.

The scales are not always the same–that is just coincidence for this quarter. That cash pile will grow the axis on the right, let me assure you.

The percents are not summed, they are operating income as a percent of revenue for each group.