Summary
NVIDIA leads in training, while AMD focuses on inference.
AMD's open-source approach, cost-effectiveness, and energy efficiency provide competitive advantages over NVIDIA's closed system.
Recent acquisitions of Xilinx and ZT Systems enhance AMD's AI inference and datacenter capabilities, positioning it to compete against NVIDIA.
The AMD Investment Thesis
Advanced Micro Devices, Inc. is one of those companies that, I think, is being held back by market sentiment and has a chance to rally strongly when sentiment turns. And because the market is currently pricing in less growth than I think is appropriate, I think the stock is undervalued, and I maintain my rating even though it has underperformed the market on a one-year basis. But in the long run, I think AMD will outperform the market again because the fundamentals are just too good.
NVIDIA vs AMD
At the end of December, Semi Analysis published a comparison between the MI300x and the H100 and H200. And here they found that NVIDIA (NVDA) continues to lead the way in training. In my opinion, this is not surprising, since AMD specializes in the area of inference. Nevertheless, this often quoted article has led to a renewed focus on NVIDIA's software competitive advantage.
And here it must be said that NVIDIA's focus on software has resulted in CUDA currently being ahead of AMD's ROCm and HIP solution. While many applications designed for CUDA can be converted using HIP, they will not run as smoothly as they do on CUDA. So it has to be said that AMD is still behind in the software area. But since AMD has more memory capacity and more memory bandwidth, they have a clear focus on inferencing, and training is, in my opinion, rather secondary for them at the moment. Especially when you consider that CUDA has been in development since 2007 and AMD only released ROCm and HIP in 2016, NVIDIA has a big head start. But in the supercomputer space with Frontier and El Captain, which is not treated as a secondary project by AMD, the ROCm software is powerful and serves its purpose.
So I think it will be interesting to see what the results will be in the next parts of the Semi Analysis AI series and if AMD will do better because, in my opinion, training is the area where it was already known that NVIDIA was ahead. And I think it is also known that NVIDIA has a better usability because their products are out-of-the-box, whereas with AMD, you have to tinker a little bit to get the best results.
However, AMD also has its advantages, such as its focus on open source and the fact that Fidelity Super Resolution is compatible with competing graphics cards, while NVIDIA is more of a closed system. And in many cases, open source solutions will prevail in the long run because they are usually more effective for markets and their participants. In addition, AMD is currently more cost-effective and offers more energy-efficient solutions.
FinChat.io
I sometimes feel that AMD's achievements in the datacenter segment are not being recognized. After all, as we can see in the chart above, AMD has managed to scale its datacenter business to $11 billion in revenue. And this is a segment with good margins, which will in my opinion lead to higher overall margins for AMD in FY25.
Data by YCharts
Over the past 12 months, we have seen margins begin to improve as investments begin to bear fruit, and this should continue. So a double-digit profit margin in FY25 would not surprise me.
AMD Investor Presentation
In the third quarter, Data Center revenues grew 122% year-over-year and 25% sequentially, while margins were significantly higher than in the Client and Gaming segments. In general, the datacenter segment currently accounts for just about half of all revenue, and I think will likely be closer to 75%+ of revenue by the end of FY25. And because the margins are strong in that area, I could see $18 to $20 billion in revenue and $8 billion to $10 billion in operating income in FY25 just from the data center segment. And that should more than compensate for the client and gaming business, which is currently at the weaker end of the cycle in their industry.
But gaming will also see higher revenues in the future as the next generation of consoles come out. Maybe the PS 5 Pro will even have a small impact on revenues, although I, personally, do not see it as a major update to the PlayStation 5 Slim. If the slump in gaming sales had come at a different time, AMD might be in a different place in terms of stock price and market sentiment.
NVIDIA, albeit from a higher base, grew its data center revenue 17% sequentially and 112% year-over-year, so we can see that the growth rates are relatively similar and in the last quarter AMD actually had a slight edge in terms of growth rate. So I think the sentiment towards AMD is far too negative and not focused on the core, which is that their datacenter segment is growing strongly. Everything else is noise.
Intel, AMD, and ARM
As many of you may know, AMD has been successful in the past in dislodging the once unassailable titan of the CPU industry, Intel (INTC), as I believe the Ryzen CPUs are ahead of Intel's offerings. And one of the reasons for AMD's success was their focus on chipset technology, which gave them a competitive advantage. And now it will be interesting to see if the change at the top of Intel will lead to the company getting back on track.
But not only Intel, but also ARM, could become a serious competitor in the CPU market. The fact that AMD and Intel have formed a joint x86 Ecosystem Advisory Group speaks to the growing influence and power of ARM solutions.
After ARM's success in smartphones, I see more and more success in notebooks and small desktops. Google, for example, has unveiled ARM-based Axion processors, and Amazon's Graviton processors are also based on ARM. But with the focus on data centers, it will be interesting to see if there are any compelling ARM solutions in the near future. Apple chips in data centers could also be something to watch here.
CES 2025
On the GPU side, AMD introduced FSR 4, which should result in a better image quality. Other than that, the focus was on PC chips, and only NVIDIA made a major announcement in the GPU space with the introduction of the GeForce RTX 50 series. NVIDIA's 5090 is expected to be the best model for gaming and AI, with NVIDIA once again providing the highest performance model. But in the mid-range price class, AMD seems to remain an alternative.
It will be interesting to see how the market for AMD's AI CPUs develops. They have released the Ryzen AI 300 at the end of 2024, and with the presented Ryzen AI 5, Ryzen AI 7 and Ryzen AI Max more powerful models will follow soon.
AMD's Acquisitions
The two most prominent recent acquisitions were probably Xilinx and ZT Systems. Xilinx provides AI inference and FPGAs, while ZT Systems provides data center architecture and engineering. Therefore, I see both acquisitions as value adding, as they are exactly what AMD needs to compete or gain market share against NVIDIA. Most exciting is the fact that they will probably sell the manufacturing part of ZT Systems and could use ZT Systems engineers to develop ROCm to attack NVIDIA's competitive advantages. With the addition of Silo AI's top-notch AI scientists and engineers, they have also taken a step in that direction.
AMD's Balance Sheet
AMD Investor Presentation
Cash is down 15% sequentially, but I think that will change over the next couple of quarters as the outstanding receivables are paid and cash comes back up. Total debt remains flat quarter-on-quarter and continues to be more than covered by cash, as cash is nearly three times total debt. Therefore, in my view, the balance sheet is still relatively safe and not a cause for concern.
Peer Comparison
Data by YCharts
I think the Forward PE comparison is a good reflection of the current sentiment towards the companies. NVIDIA is seen as the best, while AMD and Intel are seen as less good. I think a higher multiple for NVIDIA is appropriate, but I think the gap is too large, and I also think AMD is much better positioned than Intel. Therefore, I see multiple expansion potential for AMD.
If we look a little bit closer, we can see that AMD has the advantage over Intel in that they make the better CPUs right now, and they are much better positioned in the GPU market than Intel is. So AMD should definitely trade at a premium to Intel. The AMD vs. NVIDIA comparison is more difficult, as NVIDIA currently dominates the market and their GPUs are considered the best on the market, but the growth rates seem to indicate that AMD may be gaining market share.
Data by YCharts
Also, the margins and the ROIC clearly show that NVIDIA is the best quality company right now. But you always have to remember that we are investing in the future, and these are historical numbers, so it is more important how these numbers are going to evolve. And this is where those metrics, margin and ROIC, are going to develop positively for AMD as the high-margin datacenter segment becomes a larger part of the business, which will drive better numbers. While they won't quite reach NVIDIA's numbers because AMD's CPU and gaming businesses are low margin compared to the datacenter segment, the profit margin will still increase significantly from the current 7.5%.
Because the datacenter segment is growing as a percentage of AMD's revenue, I think margins and ROIC will improve, and because Intel is lagging badly in this area, I think Intel will continue to weaken and probably take more hits in the multiple. However, as AMD catches up to NVIDIA, I see potential for margin expansion.
Where could AMD's EPS be in the future?
Seeking Alpha Earnings Estimates
FY26 earnings estimates currently stand at $7.03, down slightly from the $7.06 projected in March. However, the highest estimate in March was $8.36, although there is now one analyst who sees $10. And I, personally, had given a $7.5 estimate in March, which I would now raise to $8 because costs have come down faster than I thought they would, and the datacenter segment has also done better than I thought it would.
This means that at a 30x multiple and an EPS of $8, the stock would trade at $240, a nearly 100% upside from the current price of ~$120.
What are the expectations of the market?
Author
Basis: $3 EPS for FY24. The estimate for FY24 is even slightly higher, as the chart in the last section shows.
My favorite way to see what the market is pricing in is reverse engineering. Here, I look at which EPS growth rate is priced into the current stock price. And here is where the market is currently pricing in EPS growth of 15% annually for the next 5 years and 12% annually thereafter. Such an EPS growth rate would be a strong performance for a normal company.
But AMD is no ordinary company, they have one of the most sought-after products on the market. And they have also delivered much higher growth rates in the past. The 5Y EPS CAGR is ~42% and the 10Y is ~32%. I do not expect them to continue to grow at such a high rate, but it will most likely be higher than what the market is pricing in. Especially if gaming comes back and data centers continue to grow at this rate.
Conclusion
The fact that AMD's overall revenues have not exploded as many had hoped due to the sharp decline in gaming revenues keeps AMD underappreciated. However, the company has an advantage over NVIDIA in AI inference and will benefit from margin expansion.
As the fastest growing and highest revenue division will be the most profitable in the future, which is rare for a company of this size, I see AMD as undervalued. We still don't know how cyclical the market will be in the future, but right now, I don't see any signs that the AI party won't continue in 2025 and 2026. So I think the growth rates that the market is pricing in are lower than the growth rates that AMD is actually going to achieve, and therefore I think the market is going to be wrong about AMD and does not recognize what an outstanding company AMD is.
Comments