We attended AMD’s industry analyst day event this week. The company came in force with updates to its laptop CPUs, its server CPUs and a formidable new GPU tailored for AI. So or course its stock fell 5% on the news.
Most likely, the stock fall can be chalked up to a case of “Buy the rumor, sell the news”, as much of their product announcements had been foreshadowed elsewhere. And no one we spoke with seemed to have any problems with the performance of the new products which all looked very well positioned versus the competition. Some on the Street lamented that the newest product, especially the AI accelerator, did not come with customers announcements, but we think that is a bit much for a product that has not really begun sampling in large volumes yet. Moreover, they had executives from Microsoft, AWS and Facebook on stage with glowing praise for other AMD products – so how much of a hint do we really need?
AMD’s prospects are squeezed between its two major competitors – Intel and Nvidia. They have great products, but even those are bound by the laws of physics, so there is no Silver Bullet out there to radically alter the market position. This means their products have some advantages over the competition for the moment, but those competitors will eventually launch new products, and the great wheel will keep on turning.
On the one side, Intel retains a strong incumbency position in both the Server and Client CPU markets. AMD’s products have some clear performance advantages here, which are enough to further erode Intel’s share in those markets but not totally displace them. Intel still exerts considerable channel control in the PC market, that plus a healthy dose of price cuts allow them to tread water in the segment for now. On the data center side, Intel’s share loss is more noticeable and painful. But data center customers make purchase decisions based on Total Cost of Ownership models, so if the Cost side of the Price/Performance ratio comes down a bit, the math still sort of works out in their favor, even if the performance side slips. Many investors we speak with remain cautious about AMD’s ultimate potential share in the data center market. It still has room to grow, but not to infinity.
And of course, the real focus of attention is on AI and Nvidia. AMD maintains that customers are frustrated by Nvidia’s growing grip on this corner of the market. This is a reasonable argument, but Nvidia has 11 billion contrary arguments this quarter, so it is not clear when or if customers will express their frustration with actual orders.
That being said, we think there is a real opportunity for AMD in data center AI. The training market is lost to them for the foreseeable future, they barely mentioned the topic during their presentation, a sensible omission. On the other hand, the market for data center inference is going to be much larger. All those generative AI models are currently hampered by the cost of user queries, a bottleneck for which AMD now has a solid solution. The question remains will customers take advantage of their offering, and the answer is unclear. There is no doubt that customers would like to buy something other than Nvidia. For one thing, those are in pretty short supply, and getting incredibly expensive to name another. Set against this are two factors. First is customer inertia – the hyperscalers do not like to have too many vendors and sticking with Nvidia is often the path of least organizational resistance. (We would say no one gets fired for buying Nvidia, but we can imagine what the CFOs have to say about that assumption.) The other factor is Nvidia’s CUDA software. Put simply, CUDA makes it easier to optimize Nvidia GPUs, and those optimizations are material when working at AI scale. We got into a fairly deep debate with AMD execs on this subject. They make the credible point that CUDA is not the final arbiter of AI software. Note that AMD had leaders on stage from both PyTorch and Hugging Face, two critical AI software projects. Despite this, CUDA still has its advantages. In a market where software requirements are changing incredibly rapidly, the known solution is an easy default on which to fall back. We accept that CUDA may not be a permanent competitive moat for Nvidia, but there are few practical signs of that on the ground.
Despite all this, AMD looks well positioned. We can debate the right share price, but from a strictly business perspective they continue to execute incredibly well. Their CEO, Lisa Su, is one of the best in the industry. And they have a product line-up to reflect those factors. So the debate about the stock remains very much about how much more share can they grab and how much of this brave, new AI market can they peel away. Overall, they are in a fairly solid position, and it comes down to how much upside they can pry away.