$Advanced Micro Devices (AMD.US)$ is making serious waves in...
$Advanced Micro Devices(AMD.US$ is making serious waves in the AI market, threatening to snatch a significant chunk of $ $NVIDIA(NVDA.US$ 's dominance. Two key factors are fueling this disruption:
1. A big part of $NVIDIA(NVDA.US$ 's advantage stemmed from CUDA, a software suite that made programming for their GPUs a breeze. This locked developers into the $NVIDIA(NVDA.US$ ecosystem. But the tide is turning. PyTorch, the new deep learning darling, now seamlessly supports $Advanced Micro Devices(AMD.US$ GPUs. This opens a path for $AMD to chip away at NVIDIA's software moat over time.
2. Remember how $Advanced Micro Devices(AMD.US$ 's chiplet-based architecture helped them topple $Intel(INTC.US$ ? They're bringing the same strategy to the AI battlefield. While $NVIDIA(NVDA.US$ relies on monolithic chips, $Advanced Micro Devices(AMD.US$ 's chiplets offer a cost advantage. Imagine building a complex machine – with chiplets, a single faulty part doesn't ruin the whole thing, unlike monolithic designs. This translates to cheaper production for $Advanced Micro Devices(AMD.US$ .
However, there's a catch. Just like with Intel, these chiplet-powered GPUs might not lead in performance initially, resulting in lower margins. This echoes the Innovator's Dilemma – $NVIDIA(NVDA.US$ , with its established business, faces a challenge in disrupting itself by adopting chiplet technology.
But AMD's not giving up. Through continuous iterations, they'll eventually develop a chiplet-based GPU that surpasses NVIDIA in both performance and price. When that happens (beyond the current MI300 generation), expect a major market share shift in AMD's favor. The chips are definitely on the table, and the heat is rising.
Incidentally, $AMD's MI350 boasts 1.5x the memory and 1.2x the AI compute power compared to $NVDA's upcoming Blackwell Ultra. The above is already happening.
Further, what's most interesting is that $AMD doesn't have to disrupt $NVDA to see huge upside in the business. $AMD is now the leader in the FPGA market, which is a technology that enables the most power-efficient AI at the edge.
This is a $200B business in which $AMD has no real competitors at present. So, even if $AMD doesn't disrupt $NVDA, the company can still do very well over the next decade.
$AMD is a highly asymmetric pick at present.
Disclaimer: Community is offered by Moomoo Technologies Inc. and is for educational purposes only.
Read more
Comment
Sign in to post a comment
digimonX : This is a good piece & the 1st time I see someone in moomoo platform who demonstrates understanding of AMD's technology vs competitors.
1. Although all big AI players have to buy nvidia today, all of them are concurrently;
(a). Contributing towards open source software stacks to replace nvidia's cuda. These open source software stack is hardware agnostic ie can be used on AMD, Intel & other AI hardware.
(b). engaging Broadcomm to develop custom AI accelerators, also to replace nvidia's GPUs.
It is clear that major AI players do not want nvidia's continued dominance. The writing is on the wall. If AMD is able to secure more capacity @ TSMC & solve its supply chain issues, AMD will get continued support from these major AI players.
There is a fundamental strategy difference in the direction AMD & nvidia is heading. nvidia is trying to replace everything with itself, via proprietary hardware & software. They are even trying to replace networking standards & equipment with their own proprietary hardware. On the other hand, AMD is trying to embed itself & value add in all ways possible, using open standards that industry participants can have choice over.
digimonX : 2. Amd is the pioneer in using chiplet technology in CPUs & GPUs. u are right there is significant learning curve implementing chiplet technology. It took AMD 3 generations of cpus to finally beat Intel. It will be the same story for GPUs. Although nvidia said Blackwell is chiplet based, it is really just 2 monolithic GPUs.
I believe what Jensen is trying to do is to gobble up all the $ while AI craze is still out there, then worry abt transitioning to chiplet later on. By then he would have lots of $ in the bank account. AMD, on the other hand wants to do things (design & implementation) the right way. It is taking some time but they wanna start off the right foot. We do not yet know which strategy will end up the winner. If the AI bubble pops sooner, then Jensen's idea will turn out right. If AI has a few more yrs to run, then AMD & Broadcomm will emerge the subsequent beneficiaries as major AI players shift away from nvidia.
protraderx : it's the eco system of NVDA had created, not just the Chip
digimonX protraderx : u need to understand what ecosystem mean here. The chip is the hardware & cuda is nvidia's software stack. This hardware-software combination makes up nvidia's ecosystem. The software part is also addressed in the write-ups above.
There is a concerted effort to create open source alternative(s) to cuda. Read an article like below:
How AMD and Intel Are Teaming Up With FAANG to Take On Nvidia | The Motley Fool
Another example, OpenAI, the company that created ChatGPT has proposed a software that made programming easier, citing difficulties doing so in cuda. Altho the article is 3 yrs old now, cuda was alr a mature product then. So, cuda has its issues. Have a read here:
OpenAI proposes open-source Triton language as an alternative to Nvidia's CUDA | ZDNET
All the big techs r buying fr nvidia because it is the only choice they can get at scale today. It doesn't mean they support nvidia wholeheartedly. Remember not too long ago when nvidia tried to buy over arm? Major corporations eg Microsoft, Google, Qualcomm, etc opposed the deal, citing concerns over negative effects on pricing & competition. Ultimately, regulators blocked the deal.
nvidia is known for certain behavior. This is why I said the big techs do not want nvidia's continued dominance. This doesn't mean nvidia stops making $. It just depends on how fast the big techs can bring alternative(s) (hardware + software stack) to market. The slower they r, the more nvidia is going to milk them.
protraderx : Thank you for sharing. There is a reason why all the big tech buying from Nvidia. Businesses gravitate towards the fastest, most innovative eco-systems, it has a leader with vision rather than complex multi-parties where there is no leader. Nvidia pioneered this over a decade ago, and my bet is with NVDA. To profit in stocks, invest where the money flows. Read that Nvidia is also innovating chiplet technology.. https://www.nvidia.com/en-us/data-center/nvlink-c2c/#:~:text=NVIDIA%C2%AE%20NVLink%C2%AE%2DC2C,coherently%20interconnected%20with%20custom%20silicon.
digimonX protraderx : Like I said before, nvidia is the only option that can provide AI hardware at scale. Big tech cannot get so many AI GPUs elsewhere. They don't have an alternative option now.
Open AI is 1 of the early implementors of AI. If it is nvidia's single leadership that u talked abt, open AI would not have bothered to develop the alternative software to cuda 3 yrs ago. Why waste effort.
I wanna be clear of 1 important pt. I m not saying nvidia is not innovating or they will get crushed soon. I m just stating the industry dynamics is not as simple - that big techs all embrace nvidia's ecosystem wholeheartedly. If this is so simple, they wud not have approached Broadcom to develop custom AI accelerators & simultaneously participate in workgroups to come up with alternative software stack.
protraderx : I get what you meant. Last two days AMD surged. Any idea what is happening , any catalyst news?