Nvidia vs. AMD – Game on!

AI developers will decide the outcome

• Amazon has said that it is considering using AMD’s latest AI training chips, but I suspect that it will be the developers and potentially Open AI that decides whether or not AMD makes any headway against Nvidia.
• Nvidia currently has around 84% market share in the market for silicon chips that are used to train AI algorithms which is currently enjoying a boom in demand thanks to the ongoing generative AI craze.
• On top of its virtual monopoly of this market, Nvidia is now enjoying gross margins of around 70% and so it is natural that others are keen to get in on the action.
• The problem is that Nvidia has its market position because it has been the AI training standard for over 10 years and developers are demanding that their algorithms are trained on its silicon.
• This is why both Google Cloud and Microsoft recently signed deals with Nvidia to put Nvidia systems in their cloud infrastructure.
• Google already has training silicon of its own and Microsoft is also developing its own chips meaning that they had no incentive to use Nvidia unless they were likely to lose their customers.
• This means that it is the developers who will determine whether AMD sees any traction with its silicon and this will come down to ease of use, performance and cost-effectiveness.
• Amazon should be considering implementing all available AI training chips into its infrastructure and it should do this in line with what its customers tell it what they want.
• This is why Amazon, Google, Microsoft or any other cloud provider will be thinking about implementing all of the available offerings meaning that this news from AWS means almost nothing in reality.
• The only cloud that I see on the horizon for Nvidia is the possibility that ChatGPT or other very popular generative AI systems become platforms upon which developers create their functionality.
• In the last few months, at least 140 plugins for ChatGPT have been developed to extend its function into all sorts of areas and if this continues, then the control that Nvidia has on the market may weaken.
• This is because if developers start using ChatGPT or other generative AIs as the starting point for their development, they will care less about what training or inference silicon is being used further down the stack.
• Furthermore, I am pretty sure that ChatGPT and other popular AI systems will be made available on Nvidia, AMD, GraphCore and others undermining Nvidia’s position.
• This is typical of all ecosystems as the control point originates in hardware at the very beginning when the ecosystem first comes into existence then creeps further up through the value chain as it matures.
• This is exactly what has happened in smartphones and consumer devices in general and there is no reason why it should not also happen here.
• This means that Nvidia’s challenge is to ensure that developers continue to demand its silicon which is why we see Nvidia offering more and more in terms of complete systems based on its silicon as well as more tools and platforms for development.
• Nvidia is extremely good at this sort of thing but the more successful it becomes, the more pressure it is going to have to endure from both rivals and the maturing of the ecosystem.
• Nvidia continues to be way out of my league when it comes to valuation, but in this climate where narrative is clearly driving the company, the shares can easily continue to go up.
• There are plenty of far more reasonably priced semiconductor companies such as Qualcomm, MediaTek and TSMC but for the moment the wind remains strongly in Nvidia’s sails.