원문 보기
Open AAI just announced GPT5, the most powerful large language model yet. But what they really showed us was where the entire AI revolution is headed next. And that direction could be big trouble for AMD stock if you don't know what to look for. everything I'm going to cover. AMD's current AI business as of their latest earnings call.
Open AAI's biggest announcements for GPT5. what those announcements mean for the future of AI and of course what that means for AMD stock as a result. But let me start by saying that this video is not for everyone. I did my best to keep it high level, but it gets pretty technical because AI workloads and the chips that power them are technical topics by nature.
This also isn't a video about AMD being a bad company or a bad investment since there's so much more to AMD than just their data center business. Instead, my goal is to help you understand the real investment thesis around AMD and their place in the enterprise AI market as it evolves with new use cases, new AI models, and new hardware and software stacks to support them.
So, let's take a look at where AMD is today by going through their latest earnings call. AMD has three major business units: embedded systems, client and gaming, and data centers. AMD's embedded segment was slightly down for the quarter, but it's only about 10% of their overall revenue, so we can ignore it for the sake of time.
AMD's client and gaming segment grew by 69% year-over-year thanks to record CPU sales and strong demand for their Radeon GPUs and console gaming products. If you didn't know, AMD makes the semi-custom chips for the Xbox and PlayStation 5, and Microsoft partnered with them on their next generation of consoles, PCs, and handheld devices.
And while I'm sure these PCs and gaming devices will have some AI capabilities in the future, that's not really the enterprise AI market that most investors want exposure to. AMD's data center segment reported an operating loss of 5% for the quarter due to export restrictions on their MI308 Instinct accelerators that get shipped to China.
I'm not worried about this loss though since it was a onetime $800 million write-off and their data center business would have been profitable without it. What I'm actually worried about is their data center revenues which came in at $3.2 billion. That's up only 14% year-over-year. Let me point out two important things about AMD's data center revenue.
First, half of AMD's data center revenue comes from their epic line of data center CPUs, not GPUs, networking, or any other data center products. And second, if we add back in the billion or so dollars that AMD would have made from China, their data center revenue would be up by around 50%. Which sounds much better until you compare it to Nvidia.
Nvidia's data center revenues were 39.1 billion last quarter, which was up by 69% year-over-year. So, Nvidia's data center revenue is 12 times bigger and growing five times faster than AMD's based on the actual earnings results and the actual loss chip sales to China for both companies. But let's take this comparison one step further.
Here's a table that I made comparing AMD's and Nvidia's latest data center revenues broken down by revenue source. like CPUs, GPUs, and networking equipment, as well as by geography. 89% of Nvidia's total revenues come from AI accelerators, infrastructure, and software specifically for AI data centers.
And if we add back in the $4.5 billion that Nvidia lost to export controls, 22% of their data center revenues would have come from China. But like I just said, half of AMD's data center revenues come from CPUs, not GPUs. And if we add back in the billion dollars that AMD lost to export controls, almost 40% of their data center revenues would have come from China.
So AMD's data center revenue is much less focused on AI accelerators, much more exposed to China, 12 times lower than Nvidia's, and still growing slower. Even though we're almost 3 years into the AI revolution, that doesn't mean AMD is a bad company or a bad investment. It just means we need to stop comparing it to Nvidia.
I'm a PC power user and I've been using AMD's CPUs forever. In fact, AMD is now the king of the desktop CPU market after overtaking Intel's market share last summer. Don't forget, AMD's client and gaming segment is still their biggest business unit today, contributing 47% of their total revenues. And it's not just consumer PCs.
AMD is making serious strides in the server CPU market as well. Their market share is at an all-time high of 30%. And I wouldn't be surprised to see AMD overtake Intel as the number one data center CPU maker over the next few years. The traditional data center chip market is expected to triple in size over the next 8 15% through 2033.
So, if they can keep growing their market share in these quickly growing CPU markets, then AMD will keep making serious money. Speaking of which, changing your digital location. But did you know that VPNs can also save you some serious rself and your entire family. They also have add-ons like Surf Shark Alert, which alerts you if your email, credit cards, or personal information get leaked in a data breach.
Talk about a total lifesaver. Also, many . But with Surf Shark, you can sh with Surf Shark's hopefully it's clear that AMD's data center business is fundamentally different from Nvidia's. But the big thing that investors need to realize is that their AI chips are too. AMD's chips really shine when it comes to large language model inference, specifically for massive context windows.
AMD's MI355X accelerators have 50% more memory than Nvidia's Blackwell B200 GPUs, which lets them process bigger models and longer contexts on a single GPU or split bigger workloads across fewer chips. This also gives AMD a big advantage for advanced reasoning models that think through problems step by step, use a mixture of experts approach, or reflect on their answers before outputting them.
As a result, AMD's chips perform up to 35% better than Nvidia's on massive models like Llama 405B, which means 35% lower costs per token. But there's a big catch here. This is only true for texttoext and only for smaller clusters of GPUs. Nvidia has around 20% better performance for workloads involving images, audio, video, graphics rendering, or agentic tool use, which are fundamentally different processes than just predicting the next word in a chain.
They're also better when it comes to workloads that need low latency like real-time chat bots, virtual assistants, AI enhanced multiplayer games, code generation, robotic motion and self-driving cars, or any workload where the use case demands realtime responses. It's important to understand that AMD can't easily close these performance gaps because they come from fundamental differences in hardware and software architectures.
Nvidia has proprietary high-speed chip-to-chip connections like NVLink, which provides much faster data transfer rates between GPUs. They also have tighter synchronization between GPUs and CPUs to route tasks to the right kind of chip and optimize processing performance. They have smart task scheduling which lets urgent jobs cut in line and get handled almost immediately.
Their chips run at higher clock speeds and come with dedicated low latency tensor cores and RT cores. The list of features goes on and on. This is also why Nvidia's ecosystem is so much better at AI training, which requires connecting tens of thousands of GPUs together and is usually bottlenecked by latency and bandwidth, not raw compute power.
On top of that, AMD's Rockom ecosystem is nowhere near CUDA in terms of performance. It's true that Rokom supports major frameworks like PyTorch and TensorFlow, but it typically performs 10 to 20% worse than on CUDA. And by worse, I don't just mean slower. I mean the same exact model can sometimes be less accurate on AMD's framework versus Nvidia's.
I just threw a ton of stuff at you. So here's a table summarizing everything I just said. AMD outperforms Nvidia when it comes to texton, large language model inference with large context windows. They also win when it comes to reasoning as long as that reasoning stays texttoext and the use case doesn't demand immediate answers.
Nvidia outperforms AMD for almost every other kind of inference workload, from generating images and videos to moving robots and self-driving cars based on camera data. Nvidia also wins at real-time reasoning like in chat bots, virtual assistants, and real-time code generation. And of course, Nvidia is the king of AI model training, performing up to 50% better than AMD or any hyperscalers custom chips, depending on the model and the kind of training.
I also included a row for the rest of the ecosystem because optimizations for new models or workflows are usually available on CUDA from day one. While developers can wait anywhere from 2 weeks to 6 months for the same optimization to come out on AMD's Rockom. Kind of like how many apps are built for iPhones first and then eventually make their way over to Android.
Not to mention that AMD's chips usually ship 3 to 6 months after the Nvidia chips that they're supposed to be competing against. The important thing to understand is that every row on this table impacts revenues for data centers because data centers compete by offering the fastest and most cost-effective services.
That means a 10 to 30% performance gap for a given workload translates directly to higher costs and thinner margins for data centers. And when you multiply that gap across tens of thousands of GPUs running 24/7, it means significantly more power, more rack space, and more cooling to get the same output.
The same is true when you're waiting weeks or months for optimizations that are already available on your competitor's hardware. That's why we need to stop comparing AMD and Nvidia. The real question should be how big is the part of the AI market where AMD's ecosystem actually shines. And that's where OpenAI's GPT5 announcements come in.
go down more rabbit holes the AI market for AMD's chips. Here's my best guess at the amount of compute power dedicated to each kind of workload today. These numbers are my personal estimates, and they're not official industry figures, but I really did my best here. According to my research, around 35% of all AI compute time still goes to training.
Another 20% goes to textonly inference and so on. So I think AMD's total addressable market is about onethird of the generative AI market today. Text textonly LLM inference and long context reasoning where response time isn't an issue. But if you watched OpenAI's announcement of GPT5 last week, then you saw two big trends that show us where the AI revolution is headed next.
First, future models will be much more agentic. If they don't know the answer, they'll search the internet or read a file or watch a video or write a piece of code to calculate the answer. OpenAI showed demo after demo after demo of chat GPT compiling code, rendering graphics, listening and responding to voice commands, and using tools and APIs to accomplish tasks and solve different problems.
And second, there was a ton of emphasis on speed. Speed, speed, speed. opening multiple tabs to generate competing answers. Watching code get generated and debugged in real time, asking questions and expecting immediate responses. The instant gratification that we already expect from mobile apps is now being expected from AI apps too.
And going back to my table, both of these trends, complex multimodal inference and real-time interactive reasoning are not in AMD's domain, but they are in Nvidia's. So GPT5 just showed us that AMD's share of the data center GPU market for generative AI will probably never be more than 10 to 15% in the long run.
That's bad news if you're still comparing AMD to Nvidia. But here's the good news. 10 to 15% would still be a massive market share. The global AI accelerator market is expected to more than 28% through 2034. So, if AMD can grow their share of this market from 4% today to just 8% over the next 10 years, they're going to see exponential growth in their data center business.
I think that's very possible for them to achieve and even exceed. On top of that, AMD is eating Intel's lunch when it comes to desktop CPUs, and they're only a few years away from doing the same with data center CPUs as well. Not to mention AMD's insane growth in the gaming market, thanks to their semi-custom processors powering the next generation Xbox, PlayStation, and handhelds.
Don't make the same mistake that so many Wall Street analysts are making when they compare two companies with fundamentally different products. AMD is a competitive company that's well diversified across some great computing markets, making it a great stock to get rich without getting lucky. Not because it competes with Nvidia, but specifically because it dominates the market where it doesn't. And want to see the science behind the stocks I'm buying, .
Comments 0