Nvidia GTC 2024 Day 2 – Picks and shovels

An exhibition of where the real money will be made.

• The exhibition floor of GTC is all about bringing products and services (all powered by Nvidia obviously) to market which is a reflection of Nvidia’s approach to provide the tools to bring whatever works now to market and worry about the rest later.
• The COVID pandemic conclusively proved that online exhibitions do not work, and this is the first time in 5 years that there has been a physical exhibition at GTC, and 5 years has made a big difference.
• The last time GTC was held, the show floor took up one hall in the San Jose Convention Center, but this year it takes up all of them and the keynote was so large it had to be hosted in an ice hockey stadium.
• Given the hype levels around AI at the moment one might have expected the exhibition to be full of fluff, slideware and hype but the reality was quite different.
• The floor is full of companies who are bringing stuff to market imminently and the chatter on the floor was all about execution and driving revenue and very little pie in the sky.
• This was refreshing and reflects some of the commentary given by Jensen Huang during one of the Q&A sessions when he was asked about artificial general intelligence (AGI) which is the time when machines become more intelligent than humans and take over most economic tasks.
• Nvidia’s position on this is practical in that it intends to bring the best technology that is available today to market and leave the super-intelligent machine development to the academics.
• This is why Nvidia is already making billions of dollars in profits from AI and the generative AI companies are burning money like there is no tomorrow.
• There was hardly anyone on the floor from the generative AI companies and this part of the AI sector was tucked away in one corner of the hall.
• The rest was full of tool companies, hyperscalers and service providers all of whom are looking to generate revenues now from companies integrating AI into their systems to improve their productivity.
• These are the picks and shovels of the current AI gold rush and just as it was 170 years ago, it is likely to be the tool makers that make most of the money.
• Nvidia remains uniquely well positioned here because it makes the wood and the metal from which the picks and shovels are made and, try as the others might, no one wants to use the wood or metal from anyone else at the moment.
• This may change with time as the AI ecosystem evolves, which is why Nvidia is moving to lock up the ecosystem now.
• This combined with Nvidia Inference Microservices (NIMs) is how Nvidia aims to preserve its market position when the market is more mature and the competition is more effective.
• This is the real threat for Nvidia because if the control point moves from silicon to foundation model, Nvidia is at risk of facing much harsher competition as it has no foundation model of its own.
• This is why NIMs are important because if developers can be persuaded to use NIMs rather than foundation models, then Nvidia will have migrated its control point up through the AI technology stack.
• NIMs will be optimised to run on Nvidia hardware giving a strong incentive for developers to continue to demand Nvidia hardware even if they are no longer developing to CUDA.
• This strategy is dependent on NIMs becoming popular with companies that need to use AI to boost their productivity and despite good early interest, this is far from proven.
• In the meantime, there seems to be very little to challenge Nvidia’s position and so for the short-term at least, Nvidia looks to be secure.
• Nvidia’s share is now so high that revenues are effectively out of its control leaving it hostage to whatever the market decides to do.
• For the moment, the market looks like it will continue to demand AI in ever-increasing quantities meaning that the outlook for Nvidia should continue to improve.
• With this backdrop, the shares can easily continue to move up although I am wary of what might happen should the AI bubble suddenly burst.