Nvidia GTC– Flavour of the month

Jensen executes perfect timing.

• With perfect timing, Nvidia has launched some new products and services that would have barely caused a ripple a year ago but in this environment, Nvidia is likely to end up with more demand than it can handle but for how long?
• Nvidia focused almost exclusively on AI at its 1^st GTC of 2023 and launched new products that aim to make it easy for anyone to develop their own ChatGPT-like chatbot for which I am sure it will experience huge demand for now.
• Three new cloud services were launched which are Nvidia Nemo for large language models (ChatGPT), Nvidia Picasso for image and video-related applications (Midjourney, DALLE) and BioNeMo which is for scientific and medical applications like drug discovery.
• This sounds great but it is going to be brutally expensive as DGX Server boxes which have 8 H100 or A100 GPUs and 640GB of memory will cost $37,000 each per month to rent.
• This includes storage, software as well as technical support from Nvidia to help get the generative AI model up and behaving the way intended.
• Based on Tom’s Hardware’s testing (see here), one of these is probably enough to run a very large LLM like ChatGPT but whether it is enough to train the model or deal with millions of queries is another matter entirely.
• Consequently, it is not hard to see how generative AI quickly becomes incredibly expensive but as always during a craze, no one cares.
• This is because the crowd thinks that the revenues generated will be so huge and so profitable, that the cost to deliver this technology is almost irrelevant.
• This is true until companies have to answer to their shareholders as to why the service is not as good as promised, meaning revenue delays and emergency capital raises.
• It is at this point that the bubble will burst and reality will return with a vengeance causing many start-ups to go bankrupt as well as many forced consolidations and fire sales.
• This has already happened with autonomous driving and crypto, and given that generative AI does not solve any of the pressing problems of AI today, I think that this outcome is very likely.
• Nvidia makes a very convincing case for its accelerated computing model over regular CPUs in many use cases but it does not address the cost of generative AI more broadly.
• For example, it does not answer the question of whether migrating search to a generative AI-based system from the systems in use today would be cheaper.
• Given Google’s gross margins, I suspect it would end up being far more expensive, giving another reason why I think that generative AI will not take over the search business anytime soon.
• At the moment, this is immaterial as generative AI start-up valuations remain ludicrously high and these companies are having money thrown at them.
• This means that there is plenty of money available to spend on Nvidia services and GPUs regardless of how much Nvidia is charging for them.
• Furthermore, everyone now wants to make a chatbot of their own for the second time (remember the Alexa craze of 2015 and 2016?) and this is a perfect product for anyone who doesn’t know how to get started.
• Hence, for as long as the craze lasts, Nvidia is going to have more demand than it can handle which bodes well for financial results and guidance over the next 12 months.
• Nvidia also announced that Microsoft is moving to support Omniverse which is a tacit admission that it is giving up on the Metaverse.
• Microsoft Azure will now host the Omniverse and Nvidia showcased automakers who are using Omniverse to model their factories as well as design and test their new vehicles.
• Microsoft will also integrate Omniverse with Office 365 apps to enable all aspects of collaboration for the enterprise metaverse to occur in one place.
• Completely absent was any mention of the consumer metaverse which further supports RFM’s view that the metaverse will take off in the enterprise long before it takes off for the consumer segment.
• Nvidia has given the market exactly what it wants to hear and while no one is giving a second thought to the cost of generative AI, Nvidia is likely to see a strong pick-up in demand for its products.
• This is why I continue to think that Nvidia still has further to run despite the fact that I think the valuation looks stretched whichever way you look at it.
• Make hay while the sun shines but be ready to get out when pesky reality turns up as it always does in the end.