AI is about Hardware — stupid

rolex replika

Intel or Nvidia, who will own the AI stack? The question might sound surprising to some, since AI is considered the domain of the likes of Google or Amazon or Baidu or even Tesla. But no matter how much hype there is about the filed, the only way it will actually happen is if hardware catches up to the hype. In other words, the hardware needs to be developed and engineered for all the hype to actually become reality.

Now, it’s our job to make money with investments. We operate in publicly listed securities so the theme of AI needs to match the performance of stocks. Hence the big question; which stocks will benefit from AI? Our answer; the hardware stocks first. (we missed out on Amazon and don’t want to chase it). We chose Intel as the main vehicle since the company has low valuation and high ambitions. That’s a good combination. Risky, but good. Nvidia has high ambitions, too, but the valuation and hype around the company feels uncomfortable. AMD has a great equity story but we don’t trust the company, so we stay away. So, Intel it is.

At the recent Intel AI Devcon, a conference about Intel AI, the company outlined their AI strategy. There are three key components to it.

Build on the huge installed base of data center CPUs by offering a software layer between the hardware and the various AI applications. Intel calls it the nGraph, which is a compiler that takes a model running a specific application, such as a Convolutional Neural Network for example, running an image recognition application and optimizes the model for an Intel Xeon processor. The idea of nGraph is to utilize existing hardware in the most optimal way for AI.Develop solutions for inference at the edge. Two examples are FPGAs and Movidius. In both cases hardware is developed to offer low power-high performance solutions for edge devices such as cameras or sensors, to perform machine learning at the device level.Build a purpose chip for AI training. Intel bought Nervana, which is a chip company that is trying to build a purpose built chip from scratch for AI training. According to Naveen Rao, head of Artificial Intelligence Products at Intel and founder of Nervana, the chip is particularly good at dealing with the conflict of compute and memory. How do you make sure that the compute function of the chip is not constrained by the memory bus, or in other words, what’s the use of fast compute if you can’t get data fast enough to the processor. This problem is well know in academia and is considered key for the development of AI optimized chips. Nervana is Intel’s answer to the problem.

Intel is trading at at PE ratio of around 16. That is very low. The reason for the low PE is that the core business of Intel, which is the PC business , is stagnating. But the potential in AI plus the potential in 5G are two growth drivers that can overcome the stagnant PC business. We believe Intel has the potential to break out and follow a similar path like Nvdia, which broke out thanks to its role in AI and gaming.

Edit
Pub: 03 Mar 2024 08:40 UTC
Views: 2