This is such a fantastic and analytical way to describe the parallel. We just can't help ourselves. The greed keeps causing us to repeat the same things over and over again.
I lost a ton of money on GC. I believed in the long term value of the fiber and that GC would survive. Back then I had no idea about Capacity Swaps, IRU transactions and suppliers loaning customers the money to buy their products. That experience and the various semiconductor cycles taught me, if you hear someone say "This time it's different", then run and hide, because its highly likely not different at all. I am now a Michael Burry disciple. And I am shorting against this run. Ask me in a year. Good luck all.
A most enlightening, edifying, and entertaining essay—thank you. Net-net: the hammer must necessarily fall on all this tomfoolery. Continued success in your good endeavors.
The strongest point here is not simply that today “looks like dot-com.”
It’s that the market may be confusing circular funding flows with independent demand.
That distinction matters.
If Company A funds Company B, Company B buys from Company A, and both sides then point to the transaction as evidence of demand, the question is no longer just revenue quality.
It becomes system quality.
Real demand comes from customers outside the loop.
Reflexive demand comes from participants inside the loop validating each other’s numbers.
That is the part investors need to separate very carefully in the AI infrastructure cycle.
The modern version of this trade is happening in AI and its hiding in the same accounting seam.
Microsoft invests $13 billion in OpenAI. OpenAI spends a significant chunk of that on Azure compute. Microsoft books it as cloud revenue. The cash went out as an investment and came back as revenue. The net economic transfer is circular but the income statement reads as organic growth.
Google does something similar with AI startups through its cloud credit programmes. Fund a startup, the startup buys Google Cloud compute with the funding, Google reports the compute spend as cloud revenue growth. The startup gets to report funded usage as traction. Both sides have a number to show investors.
The mechanic is identical to what you describe with the IRUs. Revenue recognised immediately on one side, cost amortised or buried on the other. The congressional quote about the gap between projected and actual revenue determining the size of the swap could have been written about quarterly cloud revenue targets today.
Your series is building something important here. The playbook survives because the product changes but the incentive structure stays the same.
Awesome comparison to the Dot Com bubble! I suspect this will have the same outcome as it did in the past. Same pig with a different name and clothing!
They’re rinsing and repeating the whole laundering operation. There’s an excellent book on the dotcom bubble, Dot Con, by John Cassidy, written as immediate history just after the crash. He has a fascinating table of share prices, showing just how many dotcoms went bust.
The most interesting comparison with the AI bubble is this insistence that we are in a new paradigm, you can’t evaluate “productivity” in the same old ways.
Here's a quote from April 7, 2000, a big meeting at the White House just as the big dotcom crash was beginning:
"America's economic elite gave a boastful endorsement of the new economy at a White House conference yesterday and said they remain mostly optimistic that the good times and record-breaking economic expansion can continue.
"Speaker after speaker in the gilded East Room of the White House said the information revolution and technology investment were transforming the economy and had led to a sharp increase in productivity. ... No-one believed that the business cycle was gone for good, although economists said the newly productive economy might be less prone to recession than ever before."
Note the phrases: "the new economy", "the newly productive economy". We're just so incredibly productive now that you simply can't measure it. Those insane stock evaluations are quite valid, if you only have the vision to understand how fantastically everything is being transformed.
We’re hearing exactly the same rhetoric now. One human will supervise a thousand bots, it’s all going to be so productive, until that human starts trying to explain why their bots are behaving in a certain way, like deleting databases and engaging in crypto speculation with stolen tokens.
This time, though, there’s really going to be a big bang when the bubble bursts. The worldwide backlash against AI is a signal. Sell. Quickly.
Creative accounting at best, recording transactions as revenue while spreading costs over decades to sustain favorable narratives. Similar dynamics may exist in AI. GPU makers, hyperscalers, and neoclouds invest in one another, buy each other’s services, and lean on alternative metrics to keep confidence and long-term momentum alive.
I’m not quite sure why anyone who was over the age of 20 in 2000 is being suckered into the current bubble but the pull of the herd is strong. Elon says AGI, I say Pets.com.
hmm, what about the argument that during the Dot-com era the fiber was dark, i.e. we did not have demand for it. Today, on the other hand, there are no dark GPUs. All GPUs are hard at work.
This is such a fantastic and analytical way to describe the parallel. We just can't help ourselves. The greed keeps causing us to repeat the same things over and over again.
I lost a ton of money on GC. I believed in the long term value of the fiber and that GC would survive. Back then I had no idea about Capacity Swaps, IRU transactions and suppliers loaning customers the money to buy their products. That experience and the various semiconductor cycles taught me, if you hear someone say "This time it's different", then run and hide, because its highly likely not different at all. I am now a Michael Burry disciple. And I am shorting against this run. Ask me in a year. Good luck all.
Brilliant article
A most enlightening, edifying, and entertaining essay—thank you. Net-net: the hammer must necessarily fall on all this tomfoolery. Continued success in your good endeavors.
Excellent piece.
The strongest point here is not simply that today “looks like dot-com.”
It’s that the market may be confusing circular funding flows with independent demand.
That distinction matters.
If Company A funds Company B, Company B buys from Company A, and both sides then point to the transaction as evidence of demand, the question is no longer just revenue quality.
It becomes system quality.
Real demand comes from customers outside the loop.
Reflexive demand comes from participants inside the loop validating each other’s numbers.
That is the part investors need to separate very carefully in the AI infrastructure cycle.
Fantastic and timely series.
Thank you. I really enjoyed my lunch reading this piece. Or vice versa, I enjoyed this article while eating my lunch
The backlogs are so suspicious. Only time will tell. But why take the risks.
Well done!
The modern version of this trade is happening in AI and its hiding in the same accounting seam.
Microsoft invests $13 billion in OpenAI. OpenAI spends a significant chunk of that on Azure compute. Microsoft books it as cloud revenue. The cash went out as an investment and came back as revenue. The net economic transfer is circular but the income statement reads as organic growth.
Google does something similar with AI startups through its cloud credit programmes. Fund a startup, the startup buys Google Cloud compute with the funding, Google reports the compute spend as cloud revenue growth. The startup gets to report funded usage as traction. Both sides have a number to show investors.
The mechanic is identical to what you describe with the IRUs. Revenue recognised immediately on one side, cost amortised or buried on the other. The congressional quote about the gap between projected and actual revenue determining the size of the swap could have been written about quarterly cloud revenue targets today.
Your series is building something important here. The playbook survives because the product changes but the incentive structure stays the same.
Awesome comparison to the Dot Com bubble! I suspect this will have the same outcome as it did in the past. Same pig with a different name and clothing!
They’re rinsing and repeating the whole laundering operation. There’s an excellent book on the dotcom bubble, Dot Con, by John Cassidy, written as immediate history just after the crash. He has a fascinating table of share prices, showing just how many dotcoms went bust.
The most interesting comparison with the AI bubble is this insistence that we are in a new paradigm, you can’t evaluate “productivity” in the same old ways.
Here's a quote from April 7, 2000, a big meeting at the White House just as the big dotcom crash was beginning:
"America's economic elite gave a boastful endorsement of the new economy at a White House conference yesterday and said they remain mostly optimistic that the good times and record-breaking economic expansion can continue.
"Speaker after speaker in the gilded East Room of the White House said the information revolution and technology investment were transforming the economy and had led to a sharp increase in productivity. ... No-one believed that the business cycle was gone for good, although economists said the newly productive economy might be less prone to recession than ever before."
https://www.afr.com/world/us-elite-exudes-optimism-20000407-jl5i0
Note the phrases: "the new economy", "the newly productive economy". We're just so incredibly productive now that you simply can't measure it. Those insane stock evaluations are quite valid, if you only have the vision to understand how fantastically everything is being transformed.
We’re hearing exactly the same rhetoric now. One human will supervise a thousand bots, it’s all going to be so productive, until that human starts trying to explain why their bots are behaving in a certain way, like deleting databases and engaging in crypto speculation with stolen tokens.
This time, though, there’s really going to be a big bang when the bubble bursts. The worldwide backlash against AI is a signal. Sell. Quickly.
Creative accounting at best, recording transactions as revenue while spreading costs over decades to sustain favorable narratives. Similar dynamics may exist in AI. GPU makers, hyperscalers, and neoclouds invest in one another, buy each other’s services, and lean on alternative metrics to keep confidence and long-term momentum alive.
I’m not quite sure why anyone who was over the age of 20 in 2000 is being suckered into the current bubble but the pull of the herd is strong. Elon says AGI, I say Pets.com.
hmm, what about the argument that during the Dot-com era the fiber was dark, i.e. we did not have demand for it. Today, on the other hand, there are no dark GPUs. All GPUs are hard at work.
GPUs that are installed. there are lots, lots of GPUs collecting dusts in warehouses.
Don't know the extend of that. There are some, for sure, but Nvidia guidance promises huge sales. Why would they lie about that?
Political circle of hell gets funded meanwhile... depression, radicalisation and further corruption.