But the thing is, all other things are NOT equal, and the fact that AMD’s drivers suck, that popular AI software libraries don’t run as well on AMD GPUs, that you can’t find really good GPU experts who specialize in AMD GPUs outside of the gaming world (why would they bother when there is more demand in the market for CUDA experts?), that you can’t wire thousands of them together as effectively because of lousy interconnect technology for AMD— all this means that AMD is basically not competitive in the high-end data center world, and doesn’t seem to have very good prospects for getting there in the near term.
My experience with google map is, it’s not updated that frequently for such remote site. At one site which i track since over last ~2 years, it took more than 8 months to update image of changes at site. Funnily, after 1.5 years the site shows images which are older than the intermediate changes. Like barren site (1.5 yrs back) → construction started → Nearly after 8 months after construction started , maps showed the images of construction start phase → Now it’s been nearly 1.5+ years and full construction is complete → The image has got updated to show barren site again.
I think looking at openai is simply looking in the wrong direction. Openai or not, the value of amd compute hw was waiting to be unlocked. Is there any other company that can compete nvidia in gpu compute? NO. it was always as simple as that. Rest was monitoring what the management was doing. SW improvement was already going great. All the other businesses are firing on all cylinders (leaving embedded).
If the hw(gpu) did not exist, openai would not even be in the deal. As jensen said gpus is way you go when algorithms are changing so fast. Not asics. If not for openai, it would have been some other csp. Is there any other gpu maker out there? I have been suprised by the blind side of analysts last year.. I kept loading up amd. Most analysts are bean counters with advanced supply chain data. What the author of the above article said needs no effort. Just state the things as they are and say they will stay that way. Done.
But openai may be a better partner than say msft since msft really does not have a great track record with third party silicon. They short changed x86 for arm last year by giving exclusive copilot+ branding to qc for no technical reason when others were giving > 40 tops.
Interesting. Did some research across other platforms and seems Tata had made some traction on the site. keyword “some”. Will they be starting manufacturing at the end of 2026 at this pace I doubt it. There are other projects being built in Sanand which seems to be a larger city and I think construction progress will be faster on these sites
My understanding about this topic is well below your knowledge on the matter. What Ive read from other sources though sounds like they are using AMD for inference functions and Nvidia for training the large language models . I am guessing every time I punch a prompt into Chat GPT I am inferencing or running an inference function?
Overall dont know how long this exuberance will run for the chip sector. Safer to get exposure in the Nasdaq100
Regarding exuberance.. I would say a correction would be good for cos that have strong balance sheet. It would wipe out the competition. If you believe in the long term story of ai (I can already see unmatched benefits in productivity) then investing in this area is a no brainer. This is not a 1 year 2 year “market crash will kill ai” story. It is here and it is going to change the way we interact. Not even browsers may remain (jensen alludes to this in one of his talks).
I came across this really good article on economics of semi-conductor industry I believe it’s must read for everyone who wants to understand or invest in semiconductor industry:
My prospective about the semiconductor industry completely transformed after reading this article, hope it will help others also.
Have been reading about edge computing as I feel I am late to the the ai server and cloud compute economy where everything has been up more than 200% or more in the last 5 years. What opportunities are out there in a humanoid future? I see a lot of Qualcomm being mentioned but they are down currently and have not risen like the rest of the pack.
Unfortunately my experience comes from dc/consumer cpu/gpu. For edge, you are talking volume and less margin it is a slog for any new player. Checkout the incumbents. I may be totally wrong..
–
Coming to dc cpu gpu.
Even now it is amd that is the real alternative to nvidia and they are executing just fine and taking bolder steps.
In the ai world, the recent bottleneck is memory and next is cpu .. this is because agentic ai demands cpu. There have been reports that the cpu gpu ratio can go back to near 1:1. In this scenario, as far as dc cpu is concerned amd is the real deal. I think jensen is trying to front run amd in deals and that is not going to work when the tco shows up. It also appears nvidia vera is rushed because it is sampling now compared to amd Venice which was sampling last year. A lot has happened… but for me amd still remains the undervalued stock.
I have seen AMD has done a lot of deals where they have also offered stock options to potential customers. How does that affect them as an investment. Currently from my non industry opinion I take Nvidia AMD ASML and TSMC seem to be unique players that still have great potential in the future. I might buy some Qualcomm too as they seem to be one of the few players still down. With margins of 56%+ are those low for this category?
There simply is no way for amd to get arround what nvidia can do with existing monopoly. I think amd learned the lesson in dc cpu vs intel. They have had superior cpu for 7/8 years and only now are they at 40% dc revenue share. Amd did the right thing with the deal. They are not diluting now. They are diluting based on technicals(GW installations) and stock price targets in the future. They are trading future value for getting AI racks experience . The other side would also not make this deal unless they think amd has the right solution in mi450 and further ahead. The management is pretty clear they are delivering h2 26.
All companies your mentioned (except qualcomm) are good bets as long as you believe ai is really a long term transformational play. Valuations are left to you since I am dumb there.
Qualcomm, i do not believe the hype they do with their marketing. What happened to their x elite laptops that got pushed like arm windows has finally arrived? In general, it is very very hard for a new player to start from the high end market downwards. It is prudent to first get some foot hold in low end and learn the market and then make your way up.
Considering the compute feeding frenzy in inference, Qualcomm’s ai dc product is also a good candidate for assymetric bet.a
The dc gravitational pull gives new player a window in client space. Because intel is prioritising dc. Apple neo I think will be clear winner. Qcom client can also shine as intel moves supply to dc.
Lot of abbreviations here @kenshin Whats DC and QC? Data Centre and Quantum compute?
I bought some Broadcom and Qualcomm since i believe edge computing is going to be the next big thing in the future. Broadcom is on a tear any given day with volumes up there with QQQ. Qualcomm on the other hand has been a laggard and is negative for the year. Don’t understand why the market does not love it
As jensen said gpus is way you go when algorithms are changing so fast. Not asics.
Well, today my point (derived from jensen’s) about asics got invalidated. With the right flow, asics can be a great solution for inference. And that flow is Taalas foundry now. The fact that they can ‘bake’ an asic at couple of layers (instead of the costly base layer) and combined with the proven tokens/sec numbers of 14000, this is a compelling sell.
Demo chat bot with their asic is here: llama 3.1 8B. Speed demo of this model.
