At the current stage, I think it is more important with MAG 7 (and the entire AI complex) with valuation multiples alreay quite evelvated (especially on a CAPE Basis) to focus more on the underlying technology. In a gold rush, the picks & shovel companies (TSMC, NVDA, AVGO, ETC.,) always making a killing for awhile whether THERE IS real gold under the ground OR NOT. As long as the dream of the gold being there exists,the picks and shovels will continue to be bought. What the key is (especially at this stage and given current valuations) - is there actually gold under the ground or in other words will LLMs/LRM's get us to AGI or even remotely close?
As I have suggested before, if you have access to Gary Marcus Substack (Marcus on AI Substack) and are not religiously reading his posts, you are doing yourself a disservice (expecially if other Substack Authors have access to each others work for free, if not it is only $80-$90 per year) Marcus is well schooled in AI, was a Professor at NYU and started and founded and sold 2 AI companies. While admittedly, he comes acress as fairly arrogant, I have been a subscriber since the beginning. He has been dead right all the way along and is vehment that LLMs/LRMs to get to any high level of AI where ROIC > WACC is decade(s) away. And if have noticed some of the largest early proponents of LLMs/LMRs (even former Turing award winners, including Richard Sutton who wrote "Bitter Lesson" which has served as large catalyst of money pouring into LLMs/LRMs ever since it was written. Sutton recently reversed course and said he was wrong and Tweeted to Marcus that he was correct all along that you can't just keep thowing more compute (i.e. TSMC manufactured chips designed by NVDA) after more compute to reach powerful AI and it is unproductive capex - essentially scaling laws are BS and hallucinations; factual inaccuracies, lack of reasoning capability, etc., means LLMs/LRMs hit a wall and you end up thowing more bad capex against bad capex under the false percention that all you need to do is just thow more computational power at your model. A very expensive endeavor that generates negative ROIC and burns large FCF and turns you from a free cash flow maching to a borrower of debt.
If you started reading his notes since he first reviewed GTP-5 (8-9-25-GPT-5-Overdue, overhyped & underwhelming) to current (even though he has some many prescient notes in years before), I think you would have quite a different perspective on AI. Of course the picks and shovels and the sexiness of AI can carry the story a long ways - and it has done just that. However, now investors are starting to ask what types of ROIC's are we getting on this Trillions of Dollars of capex and it is not looking pretting. Thus, how willl growth in the picks and shovels segment look once AI shareholders start realizing ROIC<WACC and the MAG7 evolved from capital light to capex intensive industry. Take a look at the blowout in CDS spreads for MAG 7 stocks (especially META and ORCL (ORCL more of a MAG 7 Proxy). Throw on top of this Scam Altman and his CFO begging for some type of Goverment backstop last week along with NVDA CEO essentially saying we are close to losing the AI race to China. Bottom line, in this BUBBLE, the key is whether the technolgy (LLMs/LRMS) works and the evidence of the past 18 months has not been encouraging to say the least.
Lastly, while we are rapidly approaching the 3-year anniversary of ChatGPT. According to The Information on 9/30/25, OpenAI’s First Half Results: $4.3 Billion in Sales, $2.5 Billion Cash Burn. Even more revealing and scary, backing into OpenAI's Q3 results from MSFT's SEC Filing (as they account for their OpenAI position with the Equity method of accounting: MSFT's net income and diluted EPS for just reported Calendar Q3 were negatively impacted by net losses from investments in OpenAI, which resulted in a decrease in net income and diluted EPS of $3.1 billion and $0.41, respectively. Based on MSFT's recently disclosed 27% ownership of OpenAI (and thus the reason for equity accounting method) and using the above impact to MSFT in CAL Q3 from OpenAI, this suggests OpenAI lost about $11.5 billion during the calendar third quarter. These are huge losses and burn rates on an absolute basis and especially relative to revenues. No wonder Scammy and his CFO were probing at a Goverment backstop last week!!
Why is OpenAI so important to the entire AI complex: 1) ChatGPT started the AI cycle and AI hysteria along with Scam Altman's hyperbolic proclamations, 2) OpenAI is most aggressively looking for the gold and acquiring the picks and shovels (i.e. compute power) and 3) they are at the forefront of the majority of overwhelming levels of circular financial relationships we have seen over past months!
Thanks Russell! My default (retail here) bubble indicator is a parabolic move in the price chart, FOMO buying. MAGS has been totally orderly. SOX pretty orderly and in a nice channel back to 2009. What you are saying is helpful interesting to me as I think with "the stock market follows pricing, and pricing normally follows supply" you are talking about what is going on the inside.
I recall months back you had anti-trust/ break-up concerns, I remember at the time thinking kinda opposite ...TINA, but that was pre-tariffs I think. I guess your view has shifted somewhat here?
Certainly there is a bubble in Google Trends for 'tech bubble', I wish the data went back pre-2004, be we are still early in thing whole internet thing. Searching just 'bubble', looks like a trailing indicator for GFC rather than predictive.
Maybe here too:
“If everyone is worried about something, then the chances of it happening are close to zero”
Back in 2021, Chinese government effectively moved against its tech companies. At the time they were very explicit - but no one was worried about it, so stocks fell 50 to 80%.
In this case, everyone is calling bubble - but I dont see either the Fed or the government moving against it... my only caveat is that if AI is driving inflation, then we could see a surprise on the monetary side - that is higher interest rates - and that is something people do not seem to be worried about.
The shift from price comptetion to investment competition is a critical insight. TSMC's ability to sustain $40bn capex while generating cash sets them apart from Intel's struggles. This "invest to compete" model might actually be more sustainable in a tight labor market since it priortizes scaling production over racing to the botom on pricing.
It also fits in my view with that we are heading to capital scarcity - if you need capital to invest, then making capital more scarce will be a competitive advantage
At the current stage, I think it is more important with MAG 7 (and the entire AI complex) with valuation multiples alreay quite evelvated (especially on a CAPE Basis) to focus more on the underlying technology. In a gold rush, the picks & shovel companies (TSMC, NVDA, AVGO, ETC.,) always making a killing for awhile whether THERE IS real gold under the ground OR NOT. As long as the dream of the gold being there exists,the picks and shovels will continue to be bought. What the key is (especially at this stage and given current valuations) - is there actually gold under the ground or in other words will LLMs/LRM's get us to AGI or even remotely close?
As I have suggested before, if you have access to Gary Marcus Substack (Marcus on AI Substack) and are not religiously reading his posts, you are doing yourself a disservice (expecially if other Substack Authors have access to each others work for free, if not it is only $80-$90 per year) Marcus is well schooled in AI, was a Professor at NYU and started and founded and sold 2 AI companies. While admittedly, he comes acress as fairly arrogant, I have been a subscriber since the beginning. He has been dead right all the way along and is vehment that LLMs/LRMs to get to any high level of AI where ROIC > WACC is decade(s) away. And if have noticed some of the largest early proponents of LLMs/LMRs (even former Turing award winners, including Richard Sutton who wrote "Bitter Lesson" which has served as large catalyst of money pouring into LLMs/LRMs ever since it was written. Sutton recently reversed course and said he was wrong and Tweeted to Marcus that he was correct all along that you can't just keep thowing more compute (i.e. TSMC manufactured chips designed by NVDA) after more compute to reach powerful AI and it is unproductive capex - essentially scaling laws are BS and hallucinations; factual inaccuracies, lack of reasoning capability, etc., means LLMs/LRMs hit a wall and you end up thowing more bad capex against bad capex under the false percention that all you need to do is just thow more computational power at your model. A very expensive endeavor that generates negative ROIC and burns large FCF and turns you from a free cash flow maching to a borrower of debt.
If you started reading his notes since he first reviewed GTP-5 (8-9-25-GPT-5-Overdue, overhyped & underwhelming) to current (even though he has some many prescient notes in years before), I think you would have quite a different perspective on AI. Of course the picks and shovels and the sexiness of AI can carry the story a long ways - and it has done just that. However, now investors are starting to ask what types of ROIC's are we getting on this Trillions of Dollars of capex and it is not looking pretting. Thus, how willl growth in the picks and shovels segment look once AI shareholders start realizing ROIC<WACC and the MAG7 evolved from capital light to capex intensive industry. Take a look at the blowout in CDS spreads for MAG 7 stocks (especially META and ORCL (ORCL more of a MAG 7 Proxy). Throw on top of this Scam Altman and his CFO begging for some type of Goverment backstop last week along with NVDA CEO essentially saying we are close to losing the AI race to China. Bottom line, in this BUBBLE, the key is whether the technolgy (LLMs/LRMS) works and the evidence of the past 18 months has not been encouraging to say the least.
Lastly, while we are rapidly approaching the 3-year anniversary of ChatGPT. According to The Information on 9/30/25, OpenAI’s First Half Results: $4.3 Billion in Sales, $2.5 Billion Cash Burn. Even more revealing and scary, backing into OpenAI's Q3 results from MSFT's SEC Filing (as they account for their OpenAI position with the Equity method of accounting: MSFT's net income and diluted EPS for just reported Calendar Q3 were negatively impacted by net losses from investments in OpenAI, which resulted in a decrease in net income and diluted EPS of $3.1 billion and $0.41, respectively. Based on MSFT's recently disclosed 27% ownership of OpenAI (and thus the reason for equity accounting method) and using the above impact to MSFT in CAL Q3 from OpenAI, this suggests OpenAI lost about $11.5 billion during the calendar third quarter. These are huge losses and burn rates on an absolute basis and especially relative to revenues. No wonder Scammy and his CFO were probing at a Goverment backstop last week!!
Why is OpenAI so important to the entire AI complex: 1) ChatGPT started the AI cycle and AI hysteria along with Scam Altman's hyperbolic proclamations, 2) OpenAI is most aggressively looking for the gold and acquiring the picks and shovels (i.e. compute power) and 3) they are at the forefront of the majority of overwhelming levels of circular financial relationships we have seen over past months!
Thanks Russell! My default (retail here) bubble indicator is a parabolic move in the price chart, FOMO buying. MAGS has been totally orderly. SOX pretty orderly and in a nice channel back to 2009. What you are saying is helpful interesting to me as I think with "the stock market follows pricing, and pricing normally follows supply" you are talking about what is going on the inside.
I recall months back you had anti-trust/ break-up concerns, I remember at the time thinking kinda opposite ...TINA, but that was pre-tariffs I think. I guess your view has shifted somewhat here?
Certainly there is a bubble in Google Trends for 'tech bubble', I wish the data went back pre-2004, be we are still early in thing whole internet thing. Searching just 'bubble', looks like a trailing indicator for GFC rather than predictive.
Maybe here too:
“If everyone is worried about something, then the chances of it happening are close to zero”
Back in 2021, Chinese government effectively moved against its tech companies. At the time they were very explicit - but no one was worried about it, so stocks fell 50 to 80%.
In this case, everyone is calling bubble - but I dont see either the Fed or the government moving against it... my only caveat is that if AI is driving inflation, then we could see a surprise on the monetary side - that is higher interest rates - and that is something people do not seem to be worried about.
The shift from price comptetion to investment competition is a critical insight. TSMC's ability to sustain $40bn capex while generating cash sets them apart from Intel's struggles. This "invest to compete" model might actually be more sustainable in a tight labor market since it priortizes scaling production over racing to the botom on pricing.
It also fits in my view with that we are heading to capital scarcity - if you need capital to invest, then making capital more scarce will be a competitive advantage
How do you say gold is trading well against Mag7?
On a relative basis, Mag 7 peaked against gold at the beginning of the year, and the 200MDA of the relative is favouring gold, but only just.
I have just been look at SOX instead of Mag7 v gold. v similar, longer price series
A lot of overlap between SOX and Mag 7... Nvidia is a big component of both, and Broadcom is big AI play in the SOX
ok thanks for explaining. If it breaks below the beginning of the year level, you will start to worry?
Yes