Thank you for writing this highly topical and well-researched piece on the hype vs reality of AI. At the Bloomberg Technology Summit today, I heard the future direction would be training specific industry/firm data using smaller models rather than LLM, and there are so much data out there to be trained still (away from the internet). At the same time, tech diffusion also can take up to 20 years so the AI adoption and diffusion throughout each and every industry is only at the beginning. So it is still an exciting future to see how our life and productivity will truly be impacted and what type of AI consumers are willing to spend on.
Hi Marianne, Good to see you here. You bring up an excellent point. The next step would be verticalized LLMs that are trained on very specific data sets rather than broad-based parameters. Its just like how cloud computing evolved. First there was just the general cloud led by the big hyperscalers, then you had some specific cloud verticals for healthcare, sales&marketing etc.
That could be another tipping point too that ties into the theory of user maturity. The one caveat is that startups and firms may have already anticipated this and would have started work on building out vertical LLMs for example, Google launched AlphaFold AI last week that assists in pharmaceutical drug discovery.
There are a lot of benefit to AI, but it can't do what many told in would be doing in the near future. We saw a major leap wit ChatGPT and it has improved since and will. I guess for many their mind is still the limit, but on the same way AI is still not "thinking". I look at it like this. It emphasized who or what you already are and what you know. But the Gatner Hypecycle says it all. For me it's undoubtedly a big party bubble and the next morning is right around the corner. Most will realise that it can help a lot, but not as much as people got sold.
The AI "Scam" is how the re-distribution of wealth occurs here with Generative AI super-charging the Cloud and Ads growth of monoopoly firms. While the vendors get peanuts outside of the winners that aren't even determined in a meritocratic way, i.e. OpenAI eating most of the ALL of the profits made by Generative AI startups today.
So the Generative AI narrative is really a scam that accelerates winners takes all Capitalism in America. It also allows the elites to get a lot more wealthy on the stock market pumping their winners. It's an age-old playbook but with a stark lack of regulation, transparency and even factual information.
Great read, thank you for the detailed analysis and historic perspective. Having lived through all the tech bubble cycles in modern times, I believe what we are experiencing is layers of bubbles and benefits. Some smaller bubbles have already burst, the larger ones have yet to burst, some will deflate and remain/transform over a long time (so many still believe technology fundamentally benefits all people—a bubble that was invented and inflated in the 1970s). The financial / investment bubble you’re describing here is definitely leaking already and it could be argued from January 2023 has already burst—it is less than half in volume and total capital. Given the freedom we humans will always overbuild, overcapacity, oversupply, over-inflate in mass before balance is restored. It’s the way of the herd. We murmurate our way to find a / the direction(s) to go.
Thanks for adding your thoughts, Dean. I've definitely not seen as many tech bubbles as you have but I've spending a lot of time reading and researching about them - and that comes from me, an ex-Silicon Valley tech person. That is an interesting way of putting it with smaller bubbles bursting. Great perspective
Hey Uttam. Market cap. I would think Nvidia with Blackwell. Given AMD earnings are telling a story of a one horse race.
I think Nvidia will continue to go higher by 20-30% over the next year and could be the number 2 biggest company. That’s my best guess. Surely could stall out here but I can’t imagine demand for Blackwell won’t be through the roof. Makes the current gen seem tame in comparison. What has worried me is a lull in demand since Nvidias conference and announcement of Blackwell.
Correct. For now I think there could still be room for stocks to party. My best guess would be to look at the rate of capex growth for companies and compare them to the nomina/real growth of each company vs nominal/real growth of semis. It's also unclear sometimes what justification companies use to capitalize expenses so commentary would be another thing to peel some more layers of the capex side of things.
About Blackwell - we'll get to know in a couple weeks. Jensen's messaging around the timelines for shipping GB200/B100 etc.. will be key to understand future demand i guess
I got several Email replies about how well this piece was written. Credit to the authors and their Newsletter as well: take a look: ~ Want to connect the dots in macroeconomics, technology & culture to help you understand the "big picture": https://amritaroy.substack.com/
Thanks for adding your thoughts in Daniel. I think you have quite a reasonable set of assumptions driving your hypothesis. If you go back to the hype cycle chart, rising expectations is always fueled by enthusiasm of the innovation, especially if there is a definite practical use case like the set of use cases that AI presents today. I see your point of consistent innovation, new LLMs, verticalization and product launches can lead the momentum but to sustain that momentum users need to pay up for companies to sustain that momentum. So far, enterprises have been leading the way paying up for Copilots and AI credits which has encouraged hyperscalers and other AI Titans to scale their capital expenses by a +30% average annual growth rate when most consumer electronic companies are still registering single digit sales. Even Small enterprises appear to be choosy about how they spend on AI at the moment.
So, while the long-term gains are definitely there for AI and I do believe in the efficiencies and productivity that AI can deliver, I believe this will go through some kind of a reset first before correcting and advancing just like how the internet evolved. But unlike the internet, I think the correction and reset in expectations will be significantly quicker than the internet, purely because of the productivity gains that AI has already started to demonstrate. Hope that answers your question.
https://en.wikipedia.org/wiki/Telecommunications_Act_of_1996
Thank you for writing this highly topical and well-researched piece on the hype vs reality of AI. At the Bloomberg Technology Summit today, I heard the future direction would be training specific industry/firm data using smaller models rather than LLM, and there are so much data out there to be trained still (away from the internet). At the same time, tech diffusion also can take up to 20 years so the AI adoption and diffusion throughout each and every industry is only at the beginning. So it is still an exciting future to see how our life and productivity will truly be impacted and what type of AI consumers are willing to spend on.
Hi Marianne, Good to see you here. You bring up an excellent point. The next step would be verticalized LLMs that are trained on very specific data sets rather than broad-based parameters. Its just like how cloud computing evolved. First there was just the general cloud led by the big hyperscalers, then you had some specific cloud verticals for healthcare, sales&marketing etc.
That could be another tipping point too that ties into the theory of user maturity. The one caveat is that startups and firms may have already anticipated this and would have started work on building out vertical LLMs for example, Google launched AlphaFold AI last week that assists in pharmaceutical drug discovery.
There are a lot of benefit to AI, but it can't do what many told in would be doing in the near future. We saw a major leap wit ChatGPT and it has improved since and will. I guess for many their mind is still the limit, but on the same way AI is still not "thinking". I look at it like this. It emphasized who or what you already are and what you know. But the Gatner Hypecycle says it all. For me it's undoubtedly a big party bubble and the next morning is right around the corner. Most will realise that it can help a lot, but not as much as people got sold.
The AI "Scam" is how the re-distribution of wealth occurs here with Generative AI super-charging the Cloud and Ads growth of monoopoly firms. While the vendors get peanuts outside of the winners that aren't even determined in a meritocratic way, i.e. OpenAI eating most of the ALL of the profits made by Generative AI startups today.
So the Generative AI narrative is really a scam that accelerates winners takes all Capitalism in America. It also allows the elites to get a lot more wealthy on the stock market pumping their winners. It's an age-old playbook but with a stark lack of regulation, transparency and even factual information.
Great read, thank you for the detailed analysis and historic perspective. Having lived through all the tech bubble cycles in modern times, I believe what we are experiencing is layers of bubbles and benefits. Some smaller bubbles have already burst, the larger ones have yet to burst, some will deflate and remain/transform over a long time (so many still believe technology fundamentally benefits all people—a bubble that was invented and inflated in the 1970s). The financial / investment bubble you’re describing here is definitely leaking already and it could be argued from January 2023 has already burst—it is less than half in volume and total capital. Given the freedom we humans will always overbuild, overcapacity, oversupply, over-inflate in mass before balance is restored. It’s the way of the herd. We murmurate our way to find a / the direction(s) to go.
Thanks for adding your thoughts, Dean. I've definitely not seen as many tech bubbles as you have but I've spending a lot of time reading and researching about them - and that comes from me, an ex-Silicon Valley tech person. That is an interesting way of putting it with smaller bubbles bursting. Great perspective
Agree to disagree. Good read, but I would bet Apple is sitting at number 3 before any kind of implosion.
Hi Eric! Nice to see you here too. No problem with disagreeing - but what do you mean by Apple at number 3?🥸😁
Hey Uttam. Market cap. I would think Nvidia with Blackwell. Given AMD earnings are telling a story of a one horse race.
I think Nvidia will continue to go higher by 20-30% over the next year and could be the number 2 biggest company. That’s my best guess. Surely could stall out here but I can’t imagine demand for Blackwell won’t be through the roof. Makes the current gen seem tame in comparison. What has worried me is a lull in demand since Nvidias conference and announcement of Blackwell.
Correct. For now I think there could still be room for stocks to party. My best guess would be to look at the rate of capex growth for companies and compare them to the nomina/real growth of each company vs nominal/real growth of semis. It's also unclear sometimes what justification companies use to capitalize expenses so commentary would be another thing to peel some more layers of the capex side of things.
About Blackwell - we'll get to know in a couple weeks. Jensen's messaging around the timelines for shipping GB200/B100 etc.. will be key to understand future demand i guess
I got several Email replies about how well this piece was written. Credit to the authors and their Newsletter as well: take a look: ~ Want to connect the dots in macroeconomics, technology & culture to help you understand the "big picture": https://amritaroy.substack.com/
Thanks so much, Michael!
Really interesting read, thanks for the analysis
Thanks! Glad you found it interesting!
Thanks for adding your thoughts in Daniel. I think you have quite a reasonable set of assumptions driving your hypothesis. If you go back to the hype cycle chart, rising expectations is always fueled by enthusiasm of the innovation, especially if there is a definite practical use case like the set of use cases that AI presents today. I see your point of consistent innovation, new LLMs, verticalization and product launches can lead the momentum but to sustain that momentum users need to pay up for companies to sustain that momentum. So far, enterprises have been leading the way paying up for Copilots and AI credits which has encouraged hyperscalers and other AI Titans to scale their capital expenses by a +30% average annual growth rate when most consumer electronic companies are still registering single digit sales. Even Small enterprises appear to be choosy about how they spend on AI at the moment.
So, while the long-term gains are definitely there for AI and I do believe in the efficiencies and productivity that AI can deliver, I believe this will go through some kind of a reset first before correcting and advancing just like how the internet evolved. But unlike the internet, I think the correction and reset in expectations will be significantly quicker than the internet, purely because of the productivity gains that AI has already started to demonstrate. Hope that answers your question.