저자: ParkevTatevosian,CFA · 외국인
요약 보기
AMD MI300가 출시 두 분기 만에 누적매출 10억달러를 넘어 최단기간 성장 제품이 됐고, 마이크로소프트·메타·오라클 등에 배치됐다. 파트너사들이 H100 대비 토큰/초 처리량 개선을 보고해 추론(inference) 시장에서 비용 우위 가능성이 있다. AMD는 2024년 데이터센터 GPU 매출 전망을 35억→40억달러로 상향했다. 다만 엔비디아가 여전히 선두이며 AMD 게이밍 부문은 감소 중이다. 발표자는 2024~2025년 내 엔비디아 성과를 따라잡기는 어렵다고 보고, 투자 선택에선 엔비디아가 낫다고 결론지었다.
다듬은 글
hey everyone of course you all know about the success Nvidia is having in the AI Market but could AMD catch up to Nvidia and could AMD become the next Nvidia to thrive and prosper and experience a stock price appreciation similar to that of Nvidia well one of the ways AMD can do that is with a GPU that can compare with nvidia's for artificial intelligence so I'll elaborate on that here in this video so let's take a look at the detail AMD launched its rival the Mi 300 to Nvidia a couple of quarters ago and AMD is giving us an update on how the progress is in the marketplace for amd's gpus AMD saying that the MI I 300 became the fastest ramping product in AMD history passing $1 billion in total sales in less than two quarters in the cloud the mi30 production deployments expanded at Microsoft meta platforms Oracle to power generative AI training and inferencing for both internal workloads and the broad set of public offerings so you're already starting to see the beginnings of a AMD experiencing a surge in Revenue as a result of demand for its AI GPU that's a great start but it's not yet there so there's more to know let's take a closer look so AMD saying that several of our partners are seeing significant increases in tokens per second when running their Flagship large language models on Mi 300X compared to h100 and the h100 is NVIDIA chip so what AMD is saying here is what they're hearing from their Partners is that the AMD chip is more efficient to run compared to the Nvidia chip at least for inferencing so when you're thinking about artificial intelligence there's two main areas to think about training the model and inferencing or running the model so those those two in general terms to simplify you could think think of creating the model and then operating the model and what AMD is saying is not that it's cheaper to train the model with amd's chip but it's cheaper to operate the model with amd's chip that's what he that's what AMD is saying here several of our partners are seeing significant increases in tokens per per second when running their large language models on the Mi 300X and keep in mind it's much more expensive to develop a large language model than it is to operate a large language model so AMD is seeing a competitive advantage in the Lesser important part of the AI Market the running of the large language model regardless that's still an important Milestone and this is coming from AMD reporting on this I haven't seen confirmation from other companies to verify this report but if it is true that's a really big deal that means many companies are going to use nvidia's h100 to develop large language models and then once they're developed to operate them they're going to want to run them on amd's chips because it's more efficient to operate every time you search for something on a large language model that takes computing power and that computing power cost the company that's providing you that large language model that cost them money and so if they can deliver that answer to you that you're looking for at a lower cost that's beneficial for them so they're going to look to do that AMD can provide that better operating price to those companies so really interesting there and if it's true if more companies if I see this from more companies reporting this that's going to be a really big competitive Advantage for AMD in that one area at least right you're not going to beat Nvidia overall but if you can find the category that you can outperform Nvidia then you can carve out a piece of this Market that's absolutely massive it's an absolutely massive Market opportunity if AMD can carve out this segment that's going to be big for AMD and AMD stock investors so far based on their expanding customer engagements they now expect data center GPU Revenue to exceed 4 billion in 2024 that's up from 3.5 billion that they thought in January so just in the last three or four months or so the conversations that they're having with their customers has led AMD to boost their forecast for Revenue in gpus for 2024 by 500 million 500 million more Revenue expected now than it was expect Ed just three three four months ago that's great news for NVIDIA it shows they're gaining traction faster than they initially expected to their client segment Revenue was 1.4 billion which was an increase of 85% year-over-year driven by strong demand for their latest generation rise in Mobile and desktop processors with oems and in the channel and there is strong demand for their latest generation ryen processors in the first quarter so again more evidence from AMD that they Innovation hitting the marketplace and experiencing strong demand they're behind envidia still right Nvidia is long benefiting hey've been you know several quarters in a row now Nvidia is demonstrating absolute surging Revenue growth profit growth and cash flow growth AMD is just now catching up and while these two segments are performing extremely well for AMD the overall business still not performing all that well because AMD still has parts of its business that's declining notably the gaming segment is in Decline as the console you know those latest generation consoles are now in their fifth sixth year now and that's hurting AMD so parts of its business are still not thriving whereas Nvidia all of its business is thriving it's like firing on all cylinders and so to answer that question you know I posed in the headline could AMD be the next Nvidia maybe maybe there is a chance there is a nonzero chance that it can be certainly in one sliver of the artificial intelligence Market it's demonstrating that it is up there with Nvidia perhaps right competing with the inference Market providing a solution that's lower cost to serve than Nvidia so while it can approach Nvidia well it can sort of mimic the success of Nvidia I don't think excuse me I don't think it will copy or exceed nvidia's performance in any time at least not in 2024 probably not in 2025 either so Nvidia still the clear leader in this category still the clear leader I mean I've done several videos comparing Nvidia to AMD and my conclusion in all of them nearly has been if you're an investor and you're trying to choose between the two Nvidia is the better choice the valuation is actually comparable to Nvidia you're paying a similar price to the stock but you're getting better performance with Nvidia still it's nice to see AMD catching up demonstrating that it's providing the marketplace with a viable competitor to Nvidia so at least companies have a choice they don't have to only go with Nvidia and there is actually more Supply in the market that desperately needs more computing power
Comments 0