Amazon AWS re:Invent – Show stolen.

I would like to pay tribute to Charlie Munger who passed away last night. His acerbic wit and common sense had a profound impact on the way I analyse companies, themes and investments. He will be missed.

Jensen steals the show.

• AWS is holding its annual re:Invent conference where its own silicon chips played second fiddle to Nvidia’s in just another sign of just how firm Nvidia’s grip on AI training is.
• The most interesting announcements at re:Invent were Amazon Q a chatbot that I think could be very useful, updates to its in-house silicon and an appearance by Jensen Huang, CEO of Nvidia.
  • First, Amazon Q: which on the surface sounds like a me too chatbot but is actually a bit more interesting.
  • This is because it has been trained on 17 years of AWS knowledge meaning that Amazon may have finally produced a chatbot that can be useful.
  • The idea here is that Q is like an IT support function that specialises in AWS and provides straight answers, solves problems and makes suggestions.
  • Given how good LLMs have been found to be at indexing, storing and retrieving relevant data, this chatbot could prove to be very useful for AWS clients and also reduce support costs for AWS.
  • Amazon has disappointed on AI for many years but perhaps this one will be different.
  • Second, silicon: where Amazon announced an update to Graviton and Tranium but the real attention was on Graviton which is in its fourth generation.
  • Crucially, Graviton 4 moves away from Arm Neoverse N-Series and uses Arm’s V-series instead.
  • This is a sign of a shift away from focusing on efficiency with the priority moving to performance.
  • This will enable Graviton 4 to compete more directly with Intel and AMD when it comes to raw performance although it will inevitably consume more power.
  • Graviton is already well established in AWS and all of its top 100 customers are already running workloads on it.
  • The intention is to encourage clients to increase their share of the workloads that they run on it.
  • Tranium 2 should deliver 2x more energy efficiency (performance per watt) and 4x more performance than Tranium 1 but no one seems to care very much.
  • This is because the limelight was stolen by Jensen Huang who made an appearance during the event to tout the “expansion of the partnership” but I can’t help thinking that Amazon would prefer that he was not there and Nvidia chips were not in its servers.
  • The problem that Amazon (and Googe and Microsoft) has is that its customers want to train their AIs using Nvidia silicon and if AWS (or Google Cloud or Azure) don’t provide it then the clients will go elsewhere.
  • While developers continue to prefer CUDA and while Nvidia maintains its rapid product release cadence, not much is going to change.
  • Nvidia’s chips are currently the best available for training and everyone wants to use its CUDA platform which is why no one really paid much attention to Trainium.
  • It is going to take a long time and probably the bursting of the generative AI bubble before this stranglehold is broken.
• The net result is a steady set of announcements from Amazon but there is not much here that is going to make customers jump from Azure or Google Cloud back to Amazon.
• Azure has the momentum at the moment with its unique relationship with OpenAI which now seems to be back on track, and I expect that it will continue to grow faster than its competitors for a few more quarters.
• Hence, there is not much that draws me to Amazon as the shares are expensive and AWS is no longer the hottest place in the cloud industry.
• All of the cloud players look fully valued and so I would continue to focus on the value end of the technology sector where the stocks I hold are Alibaba, Qualcomm, Ouster, Peloton and Nokia.