As the market cap of the top 10 mega-cap stocks reaches an all time high, coupled with 45% YoY increase in AI capex spend from the Big 4, when will investors start demanding a return on AI investment?
We tend to over invest and overbuild when the next game changing technology shows up. Railroads are a good example. Also we had the exact same monetization questions about the internet. Early on the only ones monetizing were porn outfits. If I were to guess we see the same dynamics. We will overbuild leading to a capex pause, but will eventually figure out generalized monetization. Longer term the biggest winners will be on the software/data side as infrastructure is a smaller market and cyclical. Right now I think they need to solve the data accuracy problems to make it more commercially useful. People will look at googles put glue in pizza sauce fiasco, and question whether I want to run my enterprise on this stuff. AI is more than LLM’s of course, but that’s what can replace humans so I’d bet that’s where the really big money is once it’s solved. Good piece as always.
I had to share this one on Notes, Brett! Really interesting take on my post. You mentioned the railroads example. I think that's also a great anecdote to use here as firms started hiking capex cycles laying the "railroads" - the foundational infrastructure for the world to be connected. Later capex budgets moved to data centers whereas some investments remained to just upgrade those "railroads" - the cables, switches and routers that the internet runs on. I see this phase of AI as the early phase of the internet and LLMs are kind of like a foundational step in AI which includes the infrastructure capex needed to train those LLMs. As accuracy improves trust should return. But I think a re-rating in expectations is probably due before people eventually realize the secular growth story still lies in software and data like you mentioned.
I agree with this. One of the things I really internalized from software engineering is patterns. I feel like the kid in the sixth sense.
“I see patterns”
It’s never a guarantee, but patterns are like fractals they tend to repeat although unlike fractals with changes and distortions. The better bet is usually to assume that if you look at things as abstractions to identify why they are the same in a general sense there is a good probability they will behave the same way.
Couldn't agree more with you. We are yet to enter the mass deployment of genAI applications and adoption phase, and when the marginal cost of creation goes to almost 0, as is this is the case with AI tools, it will vastly disrupt and create new industries and jobs. Definitely think there will be short term pain as the world is not ready with the scale of creative destruction coming our way, but once companies are able to build durable business models on top of genAI, I think it will be gamechaninging.
I’m not sure if I’ve said this before or if it’s sitting in one of my 20 half written drafts of Substack posts.
I have an interesting vantage point in that I somehow managed to wiggle my way into a software engineering career, and I was also a pretty early adopter of the internet all the way back in 1994.
I was doing what was then some fairly cutting edge internet related development in the late 90’s.
My answer to will this progress at hyper speed is yes and maybe.
I’ve seen the software industry mature in that time frame, and what’s different between now and say 1995 is that everything we needed for the modern internet had to be more or less built from scratch. There was no such thing as a web server. We had to create them. We didn’t have angular or JavaScript or even PHP and ASP. The very first dynamic web pages weren’t dynamic. They were basically static pages served from Perl scripts server side.
What delayed the explosion of the internet was creating all that stuff. But where we sit today is in a place where the software infrastructure itself is orders of magnitude more advanced. The plumbing’s been built for us not only around the web, but around software in general.
That means developing integrating and deploying AI will be a much simpler and faster process than we saw for the internet in 1995, and that exploded in a very short time frame.
So yes it will change our world quickly
The maybe comes in because I’m not 100% sure on when that starts. Neural networks are already well established, and proven. LLM’s still have these problems that are not trivial.
I can provide a specific example in that I was using chatgpt to get some background on a microcap stock I was looking at last year, and it claimed that the CEO was previously the CEO of another massive company. This was laughably wrong.
I think It will move quickly from day zero. Day zero is when they work out these problems. I don’t know when that will happen. At this point I still see LLM’s as somewhat of a proof of concept not a polished product.
This is awesome Brett! I agree with you on this.. There is still lots of work that needs to be done on accurately operationalizing LLMs. Lots of glaring showstoppers like the one you just described about ChatGPT and i think you had earlier also alluded to the Use-Glue-Fix-Pizza bug by Google. It sure looks like AI is not ready for unsupervised yet and we already have Nvidia and the financial market gurus already calling for market expansion on model inference chips. I feel there is that significant lag between expectations and reality.
Also - you mentioned it took quite a while dynamic web pages to be operationalized.... I wasnt around at the time but I remember coming across some notes where all these product concepts had already been made available in the late 90's... Probably took 10-15 years for the internet to be a need-to-have from a nice-to-have.
But I feel it might be much faster in the age of AI once we get past zero-day launch errors since ML models would be in a better place to expand from there.
Great piece, covers several converging themes ... Magnificent 7 dominanc, LLMs diminishing utility (at the moment), AI hype from an insiders and societal perspective, and technology revolutions more generally. Thank you.
AI is more than real and it’s here. Short term monetization is also here only in a different shape: Efficiency gains.
Most big companies cut white collar workforce significantly due to the efficiency gains from AI which enhances their bottom line.
Infrastructure players like $NVDA on the GPU side, $SMCI $DELL and $HPQ on the sircer side companies like $PWL on the data center electrification side already saw significant gains.
What we want to see, of course, monetization of models. That market is too diffused now for big monetization but we all saw how capable GPT-4o. Two more developmen cycles on it and I believe it’ll be something that people will happily pat to have in their daily lives.
Thanks for your insight Oguz. You are right about the efficiency gains that we are seeing where companies across the spectrum are leveraging AI tools to augment and/or automate various tasks, which is probably one of the most biggest contributing factors to this so-called "white-collar recession".
I think we are in the early innings of the phase in genAI when it comes to increasing the size of the economic pie (innovation vs. automation), that is when we will see broad economic value/prosperity, disruption/new industry creation and new jobs created. Like you said, the market is too diffused for big monetization, and I expect the AI investment cycle to properly pick up, especially in the application layer, where over the course of time, we will see which companies can build durable business models vs. those who can't. Personally, I am quite excited for genAI applications in the vertical AI/SaaS space, especially with AI assistants set to become integrated part of our lives.
Like many I’m trying to consciously avoid irrational exuberance, but whenever I read comments that the latest AI craze is entirely hype I imagine the people who felt the same about big data analytics (which transformed all businesses over the last 15 years) and the internet in the early days of those technologies.
I’m know I’m not just a hopeless technophile because I’m a serious skeptic on self driving tech, cheap clean energy, and other “world changing” tech that hasn’t panned out as hyped.
But if you are thinking today’s chatbots, and other generative AI tools are “nothing new”, I encourage you to try them out for what you do. Not just to prove that they can’t do it as well as you yet. That’s not the point. The point is, as Amrita and others are always saying, to accelerate knowledge work and net increase productivity and efficiency for people who create and analyze.
Correct, Dan. Even if.... even if... someone still ignores the broad range of capabilities, AI has demonstrated over the past 15 months from a textual chat assistant in likes of ChatGPT, to video capabilities of Sora to general abilities of other AI tools.... if watching ChatGPT talk in the May event does not change someone's mind, I dont know what will. Yes, it is scary, but the wide array of use cases that AI's voice capabilities now open up offers a new dimension to the productivity/efficiency discussion. I think my husband would have some interesting things to say about your example on big data since he worked in that space.
AI is a game changer in nearly every industry and business process. Besides big beneficiaries like $NVDA and $MSFT, some lesser-known legacy companies could also gain significantly if they manage things correctly.
A major challenge for these legacy businesses is data governance, which involves like how data is gathered, stored, and processed. As this is crucial for any AI model and application, I believe it will take some more time to fully leverage AI’s potential.
Nice take, Oktay! I noticed you mentioned 'lesser-known legacy companies'. Im curious which are you're favorite picks among legacy companies who stand to be beneficiaries here?
One prime example is Teleperformance ($TEP), a leader in business process outsourcing. Following Klarna’s successful chatbot implementation announcement in Feb. of this year, the stock plummeted to all-tome lows.
For sure, AI will have an impact, but Teleperformance’s biggest cost driver is labor cost. With decades in the business and recent acquisitions, their available customer data volume has grown significantly. So, In short, If AI and chatbots improve, they will be one of the biggest beneficiaries, by training their models with massive amount of industry specific data, reducing labor costs and improving margins.
Btw, I’m finalizing my deep-dive on $TEP and will share more details soon.
Another lesser-known business is Shift Inc., a fast growing software testing company from 🇯🇵, which will further benefit from developments in AI by incorporating more automated test case generation, execution, and analysis, for instance.
However, there is a definite risk that some companies may be slow to adopt new technologies, even if they possess extensive industry-specific data and knowledge. This lag could leave them vulnerable to disruption by larger tech comapnies or new market entrants.
"AI" is 100% overhyped. I don't understand what changed that got everyone talking about AI. Probably the release of ChatGPT? The "AI" we see today is just a continuation of stuff that already existed and has for a while. AI Chabots existed 10 years ago. Auto-complete has existed for longer. There is nothing technically significant about the latest wave of AI. I think it's just a deliberately hyped up bubble. I would avoid investing.
Great read, Amrita. I'd be more careful as we approach the elections. Mr. Biden is interested in keeping the stock market stable. What comes after is a big unknown for me.
You are right, I think at a macroeconomic level, liquidity conditions will remain strong, especially with the Fed easing on their QT, and with better liquidity conditions, I don't expect any major mayhem in the stock market. Like you said, after the elections are over or next year, I have no idea.
We clearly see the market's disappointment with AI over the last earnings release in the software sector. Investors were convinced that AI would generate real revenue in FY24 and not next year.
Absolutely, especially when most of these software companies are investing in building their AI capabilities and not see that translate into revenue growth, coupled with missing their revenue and earnings projections, investors were definitely not kind. I think that investors are still pricing in growth reacceleration into the future quarters and while that may take place, I think we are going to see an airpocket of valuation squeeze if investors start shifting their valuation models from revenue to earnings/fcf based.
Excellent post! NVidia and hyperscalers are the winners in this first round. However we’re so early in this race that it’s not a good predictor of future success, even if I think they’ll fare well in the AI market as long as AI takes off. App providers are struggling but it is expected IMHO. It is much harder and takes time to build an AI-augmented app vs offering an AI platform (Hyperscalers) with services to build such app. Eventually if SaaS apps fail, hyperscalers and semi-conductors will follow suit. Can’t wait for inning two!
Exactly. You hit a lot of key points. SaaS apps are one of the two most crucial success/failure points in the evolution of GenAI and the promised augmented productivity it is expected to add to end users and eventually to GDP. SaaS apps are the portal for consumers to interact with AI, and UX is a critical factor to harness that interaction between AI and end users.
The other success/failure point, IMO, is the data accuracy issues that some models have today. There are some issues now, but Im optimistic that some key issues will be solved and eventually models like Gemini wont recommend gluing cheese to pizza slices 😁. The other part is UX like I mentioned - and I think ChatGPT's simple interface was a great way to abstract related complexities to interacting with the underlying model - whether its text or voice.
Thanks for the double shout out in your post, Amrita. I am behind on my Substack reading and just got through your latest writing here. I could not agree more with your POV. As soon as this AI carousel ride (market ATHs) ends or slows down, investors will be quick to become more discerning and demand profits and cash flows instead of hype. In fact this was coincidentally the topic of my latest Market Whispers episode. Hope you enjoy it. Cheers!
Thanks Beachman! Your previous podcast on the topic was really interesting - had to put it in to tie my views on AI monetization. I am going to bookmark your latest podcast and cycle through my to-do's as the weekend progresses.
We tend to over invest and overbuild when the next game changing technology shows up. Railroads are a good example. Also we had the exact same monetization questions about the internet. Early on the only ones monetizing were porn outfits. If I were to guess we see the same dynamics. We will overbuild leading to a capex pause, but will eventually figure out generalized monetization. Longer term the biggest winners will be on the software/data side as infrastructure is a smaller market and cyclical. Right now I think they need to solve the data accuracy problems to make it more commercially useful. People will look at googles put glue in pizza sauce fiasco, and question whether I want to run my enterprise on this stuff. AI is more than LLM’s of course, but that’s what can replace humans so I’d bet that’s where the really big money is once it’s solved. Good piece as always.
I had to share this one on Notes, Brett! Really interesting take on my post. You mentioned the railroads example. I think that's also a great anecdote to use here as firms started hiking capex cycles laying the "railroads" - the foundational infrastructure for the world to be connected. Later capex budgets moved to data centers whereas some investments remained to just upgrade those "railroads" - the cables, switches and routers that the internet runs on. I see this phase of AI as the early phase of the internet and LLMs are kind of like a foundational step in AI which includes the infrastructure capex needed to train those LLMs. As accuracy improves trust should return. But I think a re-rating in expectations is probably due before people eventually realize the secular growth story still lies in software and data like you mentioned.
I agree with this. One of the things I really internalized from software engineering is patterns. I feel like the kid in the sixth sense.
“I see patterns”
It’s never a guarantee, but patterns are like fractals they tend to repeat although unlike fractals with changes and distortions. The better bet is usually to assume that if you look at things as abstractions to identify why they are the same in a general sense there is a good probability they will behave the same way.
100%. Spot on. Are you aware about EWT? Trading? lots of fractals 😄
Who could miss Avi on seeking alpha and his EWT missives? Tbh I haven’t looked at EWT because my instincts say won’t work. Do you have a take on it?
"Who could miss Avi"..... 😆😆😆 It's like watching a financial Reality Show.
Should have said unlike some fractals because really some fractals are exactly like that, similar with distortion.
Interesting thoughts!
Personally I don’t think we can even comprehend how fast the AI space is going to move though.
We’re heading to the point where we can accomplish things in 1 month that once took us 1 year.
Industries will change quicker than we’ve ever witnessed before in human history.
100 years ago it was easy to project what the world would look like in 5-10 years. These days that is getting much harder to “see.”
Couldn't agree more with you. We are yet to enter the mass deployment of genAI applications and adoption phase, and when the marginal cost of creation goes to almost 0, as is this is the case with AI tools, it will vastly disrupt and create new industries and jobs. Definitely think there will be short term pain as the world is not ready with the scale of creative destruction coming our way, but once companies are able to build durable business models on top of genAI, I think it will be gamechaninging.
Great thoughts, thanks for sharing!
So I agree with this with a but
I’m not sure if I’ve said this before or if it’s sitting in one of my 20 half written drafts of Substack posts.
I have an interesting vantage point in that I somehow managed to wiggle my way into a software engineering career, and I was also a pretty early adopter of the internet all the way back in 1994.
I was doing what was then some fairly cutting edge internet related development in the late 90’s.
My answer to will this progress at hyper speed is yes and maybe.
I’ve seen the software industry mature in that time frame, and what’s different between now and say 1995 is that everything we needed for the modern internet had to be more or less built from scratch. There was no such thing as a web server. We had to create them. We didn’t have angular or JavaScript or even PHP and ASP. The very first dynamic web pages weren’t dynamic. They were basically static pages served from Perl scripts server side.
What delayed the explosion of the internet was creating all that stuff. But where we sit today is in a place where the software infrastructure itself is orders of magnitude more advanced. The plumbing’s been built for us not only around the web, but around software in general.
That means developing integrating and deploying AI will be a much simpler and faster process than we saw for the internet in 1995, and that exploded in a very short time frame.
So yes it will change our world quickly
The maybe comes in because I’m not 100% sure on when that starts. Neural networks are already well established, and proven. LLM’s still have these problems that are not trivial.
I can provide a specific example in that I was using chatgpt to get some background on a microcap stock I was looking at last year, and it claimed that the CEO was previously the CEO of another massive company. This was laughably wrong.
I think It will move quickly from day zero. Day zero is when they work out these problems. I don’t know when that will happen. At this point I still see LLM’s as somewhat of a proof of concept not a polished product.
This is awesome Brett! I agree with you on this.. There is still lots of work that needs to be done on accurately operationalizing LLMs. Lots of glaring showstoppers like the one you just described about ChatGPT and i think you had earlier also alluded to the Use-Glue-Fix-Pizza bug by Google. It sure looks like AI is not ready for unsupervised yet and we already have Nvidia and the financial market gurus already calling for market expansion on model inference chips. I feel there is that significant lag between expectations and reality.
Also - you mentioned it took quite a while dynamic web pages to be operationalized.... I wasnt around at the time but I remember coming across some notes where all these product concepts had already been made available in the late 90's... Probably took 10-15 years for the internet to be a need-to-have from a nice-to-have.
But I feel it might be much faster in the age of AI once we get past zero-day launch errors since ML models would be in a better place to expand from there.
Great thoughts, thanks for sharing! Cool to hear about your early experiences with the internet.
Great piece, covers several converging themes ... Magnificent 7 dominanc, LLMs diminishing utility (at the moment), AI hype from an insiders and societal perspective, and technology revolutions more generally. Thank you.
Thank you Walter, glad you enjoyed the post. It's always fun to connect the dots across all these areas, glad it made sense, at least somewhat.
AI is more than real and it’s here. Short term monetization is also here only in a different shape: Efficiency gains.
Most big companies cut white collar workforce significantly due to the efficiency gains from AI which enhances their bottom line.
Infrastructure players like $NVDA on the GPU side, $SMCI $DELL and $HPQ on the sircer side companies like $PWL on the data center electrification side already saw significant gains.
What we want to see, of course, monetization of models. That market is too diffused now for big monetization but we all saw how capable GPT-4o. Two more developmen cycles on it and I believe it’ll be something that people will happily pat to have in their daily lives.
Thanks for your insight Oguz. You are right about the efficiency gains that we are seeing where companies across the spectrum are leveraging AI tools to augment and/or automate various tasks, which is probably one of the most biggest contributing factors to this so-called "white-collar recession".
I think we are in the early innings of the phase in genAI when it comes to increasing the size of the economic pie (innovation vs. automation), that is when we will see broad economic value/prosperity, disruption/new industry creation and new jobs created. Like you said, the market is too diffused for big monetization, and I expect the AI investment cycle to properly pick up, especially in the application layer, where over the course of time, we will see which companies can build durable business models vs. those who can't. Personally, I am quite excited for genAI applications in the vertical AI/SaaS space, especially with AI assistants set to become integrated part of our lives.
Completely agreed!
Like many I’m trying to consciously avoid irrational exuberance, but whenever I read comments that the latest AI craze is entirely hype I imagine the people who felt the same about big data analytics (which transformed all businesses over the last 15 years) and the internet in the early days of those technologies.
I’m know I’m not just a hopeless technophile because I’m a serious skeptic on self driving tech, cheap clean energy, and other “world changing” tech that hasn’t panned out as hyped.
But if you are thinking today’s chatbots, and other generative AI tools are “nothing new”, I encourage you to try them out for what you do. Not just to prove that they can’t do it as well as you yet. That’s not the point. The point is, as Amrita and others are always saying, to accelerate knowledge work and net increase productivity and efficiency for people who create and analyze.
Correct, Dan. Even if.... even if... someone still ignores the broad range of capabilities, AI has demonstrated over the past 15 months from a textual chat assistant in likes of ChatGPT, to video capabilities of Sora to general abilities of other AI tools.... if watching ChatGPT talk in the May event does not change someone's mind, I dont know what will. Yes, it is scary, but the wide array of use cases that AI's voice capabilities now open up offers a new dimension to the productivity/efficiency discussion. I think my husband would have some interesting things to say about your example on big data since he worked in that space.
AI is a game changer in nearly every industry and business process. Besides big beneficiaries like $NVDA and $MSFT, some lesser-known legacy companies could also gain significantly if they manage things correctly.
A major challenge for these legacy businesses is data governance, which involves like how data is gathered, stored, and processed. As this is crucial for any AI model and application, I believe it will take some more time to fully leverage AI’s potential.
Nice take, Oktay! I noticed you mentioned 'lesser-known legacy companies'. Im curious which are you're favorite picks among legacy companies who stand to be beneficiaries here?
One prime example is Teleperformance ($TEP), a leader in business process outsourcing. Following Klarna’s successful chatbot implementation announcement in Feb. of this year, the stock plummeted to all-tome lows.
For sure, AI will have an impact, but Teleperformance’s biggest cost driver is labor cost. With decades in the business and recent acquisitions, their available customer data volume has grown significantly. So, In short, If AI and chatbots improve, they will be one of the biggest beneficiaries, by training their models with massive amount of industry specific data, reducing labor costs and improving margins.
Btw, I’m finalizing my deep-dive on $TEP and will share more details soon.
Another lesser-known business is Shift Inc., a fast growing software testing company from 🇯🇵, which will further benefit from developments in AI by incorporating more automated test case generation, execution, and analysis, for instance.
However, there is a definite risk that some companies may be slow to adopt new technologies, even if they possess extensive industry-specific data and knowledge. This lag could leave them vulnerable to disruption by larger tech comapnies or new market entrants.
Interesting picks, I haven't heard of both of them but Ill be curious to see your deep dive on $TEP. Thanks for sharing your picks, Oktay!
This is so excellently comprehensive--I also appreciate the "2 minute summary"--thank you!
Thanks Diane for your kind words. Glad you enjoyed the post.
"AI" is 100% overhyped. I don't understand what changed that got everyone talking about AI. Probably the release of ChatGPT? The "AI" we see today is just a continuation of stuff that already existed and has for a while. AI Chabots existed 10 years ago. Auto-complete has existed for longer. There is nothing technically significant about the latest wave of AI. I think it's just a deliberately hyped up bubble. I would avoid investing.
Great read, Amrita. I'd be more careful as we approach the elections. Mr. Biden is interested in keeping the stock market stable. What comes after is a big unknown for me.
You are right, I think at a macroeconomic level, liquidity conditions will remain strong, especially with the Fed easing on their QT, and with better liquidity conditions, I don't expect any major mayhem in the stock market. Like you said, after the elections are over or next year, I have no idea.
We clearly see the market's disappointment with AI over the last earnings release in the software sector. Investors were convinced that AI would generate real revenue in FY24 and not next year.
Adobe .... next week....
Absolutely, especially when most of these software companies are investing in building their AI capabilities and not see that translate into revenue growth, coupled with missing their revenue and earnings projections, investors were definitely not kind. I think that investors are still pricing in growth reacceleration into the future quarters and while that may take place, I think we are going to see an airpocket of valuation squeeze if investors start shifting their valuation models from revenue to earnings/fcf based.
Adbe will be fascinating to watch.
Excellent post! NVidia and hyperscalers are the winners in this first round. However we’re so early in this race that it’s not a good predictor of future success, even if I think they’ll fare well in the AI market as long as AI takes off. App providers are struggling but it is expected IMHO. It is much harder and takes time to build an AI-augmented app vs offering an AI platform (Hyperscalers) with services to build such app. Eventually if SaaS apps fail, hyperscalers and semi-conductors will follow suit. Can’t wait for inning two!
Exactly. You hit a lot of key points. SaaS apps are one of the two most crucial success/failure points in the evolution of GenAI and the promised augmented productivity it is expected to add to end users and eventually to GDP. SaaS apps are the portal for consumers to interact with AI, and UX is a critical factor to harness that interaction between AI and end users.
The other success/failure point, IMO, is the data accuracy issues that some models have today. There are some issues now, but Im optimistic that some key issues will be solved and eventually models like Gemini wont recommend gluing cheese to pizza slices 😁. The other part is UX like I mentioned - and I think ChatGPT's simple interface was a great way to abstract related complexities to interacting with the underlying model - whether its text or voice.
Thanks for the double shout out in your post, Amrita. I am behind on my Substack reading and just got through your latest writing here. I could not agree more with your POV. As soon as this AI carousel ride (market ATHs) ends or slows down, investors will be quick to become more discerning and demand profits and cash flows instead of hype. In fact this was coincidentally the topic of my latest Market Whispers episode. Hope you enjoy it. Cheers!
Thanks Beachman! Your previous podcast on the topic was really interesting - had to put it in to tie my views on AI monetization. I am going to bookmark your latest podcast and cycle through my to-do's as the weekend progresses.