This is funded by 1) semiconductor companies and 2) hyperscalers. The hyperscalers burned through cash reserves and are now taking on debt (some off balance sheet like Meta). The private debt markets are imploding. The semiconductor companies are priced to perfection and facing a looming helium crisis that will hit Taiwan and South Korea especially hard. It seems like the money fueling this cash burn is running out. It looks like fabs are not going to hit production targets. That will impact revenues and earnings for the semiconductor companies. How many more quarters until this bubble dies? Paul Tudor Jones says we have a couple more years. I hope not.
Meanwhile..... Whirlpool said appliance demand hit 2008 lows and suspended its dividend. Planet Fitness paused price hikes because members are too cash-strapped. Shake Shack posted an operating loss as foot traffic dried up. The NY Fed survey showed inflation expectations rising and household finances deteriorating. Mortgage rates climbed back toward 6.5. These are different companies in different industries all saying the same thing.
The IPO route really needs to be studied. It raises questions about how much will they raise from an IPO; what those proceeds will be used for (possibly to retire some debt); how long and in what amounts before insiders can sell their shares to cash out; and what the market price of the stock does after going public (if it falls quickly, then scary!). So, a tough question now must be when to go public? And, will the market absorb several of these companies coming to the market in the same time window?
Something tells me that everyone in the room thought that somehow ChatGPT would attain AGI, superintelligence or somehow become economically viable and wave a magic wand and solve everything.
Because that basically summarizes what investors think and how all the tech bros have been hyping up their products.
Great stuff and another reason for people to follow the reporting by Bloomberg, especially since Bloomberg also covers the financing side of this. Several points are becoming clear: (1) junk bonds (aka high yield riskier) and asset-based financing (repayment is tied to the performance of the asset and its residual value, not the earnings of the company) are being used. (2) The power companies are not running to build more power plants to serve the projected demand on their balance sheets alone, so there is another layer of fancy financing that will be called on there (but with approval of local public utility commissions). (3) The federal government is up to its eyeballs in debt and will not be in a position to back a lot this investment upfront or if it goes bad. (4) The big money from the Middle East sovereign wealth funds is getting nervous about all of this, so their willingness to lend is getting pulled back. We have to keep an eye on all of this as it will unravel in phases--then a big crash when creditors panic.
THE TMT PLAYBOOK, 25 YEARS LATER (that's TECHNOLOGY, MEDIA & TELECOM, for you young folk)
What OpenAI and Broadcom announced tonight looks like the telecom financing loop that helped drive the dot-com crash. Recall, billions of dollars and tens of thousands of jobs lost, and a decade to recover the ‘promise’ of the internet.
HOW THE TELCO LOOP WORKED
Vendor lends money to customer. Customer uses the loan to buy the vendor's equipment. Vendor books revenue. Vendor's stock rises. Vendor uses the higher stock price to fund more loans. The money stays inside the circle.
The structure reported tonight fits the pattern. Broadcom finances the chips. OpenAI buys them. Microsoft rents the infrastructure back to OpenAI.
WHO RAN THIS PLAYBOOK LAST TIME
Vendor financiers included Lucent Technologies, Nortel Networks, Cisco Systems, JDS Uniphase, and 3Com. Each extended enormous amounts of vendor financing during the telecom boom.
WHAT HAPPENED TO THEM
Lucent collapsed from $65 per share to under $1. Sold to Alcatel in 2006.
Nortel went bankrupt. Peak market cap of $398 billion went to zero.
Cisco survived, but the stock fell roughly 90 percent.
JDS Uniphase lost about 99 percent of its value.
3Com disappeared into acquisition by HP.
THE CUSTOMERS THAT WENT BUST
WorldCom. Global Crossing. 360networks. Winstar Communications. PSINet. XO Communications. Williams Communications. ICG Communications. McLeodUSA. Covad. NorthPoint. Rhythms NetConnections.
47 telco carriers went bankrupt between 2000 and 2003. The fiber networks they built sat mostly unused for years. The industry called it dark fiber. Empty data centers anyone?
THE MEDIA VERSION
AOL Time Warner became the defining merger disaster of the era. The combined company wrote down roughly $99 billion in 2002. Largest goodwill impairment in corporate history.
WHAT THIS MEANS
All told, $2 trillion in market value was destroyed in the telecom and internet collapse. Roughly 500,000 jobs disappeared. The financing structure behind that collapse resembles what has now emerged around AI infrastructure.
THE BIG DIFFERENCE, MAYBE?
Microsoft, Broadcom, and today's hyperscalers generate real cash flow. Their balance sheets are far stronger than the telecom operators of 2000.
THE SIGNAL TO WATCH
Microsoft is the canary in the coalmine. Literally. If it flies the game is up, if it slows spending or walks from commitments. The Information reported tonight that Broadcom wants Microsoft to commit to 40 percent of the chips before financing the first phase. Microsoft has not signed. Watch this space…
Your ears burning Gary lol? Caitlin is on the left but I have been reading her for years. I’m a right wing nut job I reckon. Anyhow I have come to learn we aren’t so different and people should work together instead of letting ourselves be divided.
Apologies for the political rant. I plugged your page and I believe you speak from the heart.
The only possible conclusion I see is that OpenAI and Anthropic collapse, with the parts getting scooped up by some combination of Google/Microsoft/Amazon/Apple. The real fun would be if one of those four, or a big hardware player like Nvidia, actually gets pulled under the waves by this, too.
Gary: Regarding the good influences of AI: I added this below to your Substack note (Topol) but it didn't show up, so here it is again:
In Nature/Cancer (online magazine), they note that AI has helped in early diagnostic predication in x-rays for later-arriving pancreatic cancer. Apparently, AI can spot clues in the x-rays that humans don’t recognize (or can’t), even when their know via other means that something is "off."
But no one save the very extreme said that AI was all-bad. Here is the note:
"AI detects early signs of pancreatic cancer"
"Computed tomography scans taken a year or more before a person develops pancreatic cancer carry signatures of precancer that the human eye can’t easily spot but artificial intelligence can. An AI tool detected early signs of pancreatic cancer with 82% accuracy in scans taken a median of 475 days before diagnosis. 'We knew, based on the biology of the disease … that the signal was there,' says radiologist Ajit Goenka. 'We just needed to find a way to be able to detect it.' The AI tool, called Radiomics-based Early Detection Model (REDMOD), offers “tangible hope for improving outcomes in this challenging disease', conclude the authors."
This is funded by 1) semiconductor companies and 2) hyperscalers. The hyperscalers burned through cash reserves and are now taking on debt (some off balance sheet like Meta). The private debt markets are imploding. The semiconductor companies are priced to perfection and facing a looming helium crisis that will hit Taiwan and South Korea especially hard. It seems like the money fueling this cash burn is running out. It looks like fabs are not going to hit production targets. That will impact revenues and earnings for the semiconductor companies. How many more quarters until this bubble dies? Paul Tudor Jones says we have a couple more years. I hope not.
Meanwhile..... Whirlpool said appliance demand hit 2008 lows and suspended its dividend. Planet Fitness paused price hikes because members are too cash-strapped. Shake Shack posted an operating loss as foot traffic dried up. The NY Fed survey showed inflation expectations rising and household finances deteriorating. Mortgage rates climbed back toward 6.5. These are different companies in different industries all saying the same thing.
LLMs turned intelligence into a circular economy of GPUs, retraining cycles, wrappers, and infrastructure debt.
Everyone gets paid while the models stay frozen.
When the subsidy ends, someone gets left holding the bag.
Right now, the planned exit is the retail investor buying LLM vendor hype through IPO.
That is risk distribution, not tech disruption.
The IPO route really needs to be studied. It raises questions about how much will they raise from an IPO; what those proceeds will be used for (possibly to retire some debt); how long and in what amounts before insiders can sell their shares to cash out; and what the market price of the stock does after going public (if it falls quickly, then scary!). So, a tough question now must be when to go public? And, will the market absorb several of these companies coming to the market in the same time window?
It's just a twist on a pyramid scheme between the chip makers and the AI mavens. As you say, a circular scam.
What's a round pyramid scheme? A balloon scheme? Donut? Hula hoop?
An AI-picycle?
Well you know it's going to pop.
Something tells me that everyone in the room thought that somehow ChatGPT would attain AGI, superintelligence or somehow become economically viable and wave a magic wand and solve everything.
Because that basically summarizes what investors think and how all the tech bros have been hyping up their products.
I think you're right.
You don't even need a fancy chatbot to reach the conclusion I did. The writing's smeared all over the walls.
Great stuff and another reason for people to follow the reporting by Bloomberg, especially since Bloomberg also covers the financing side of this. Several points are becoming clear: (1) junk bonds (aka high yield riskier) and asset-based financing (repayment is tied to the performance of the asset and its residual value, not the earnings of the company) are being used. (2) The power companies are not running to build more power plants to serve the projected demand on their balance sheets alone, so there is another layer of fancy financing that will be called on there (but with approval of local public utility commissions). (3) The federal government is up to its eyeballs in debt and will not be in a position to back a lot this investment upfront or if it goes bad. (4) The big money from the Middle East sovereign wealth funds is getting nervous about all of this, so their willingness to lend is getting pulled back. We have to keep an eye on all of this as it will unravel in phases--then a big crash when creditors panic.
THE TMT PLAYBOOK, 25 YEARS LATER (that's TECHNOLOGY, MEDIA & TELECOM, for you young folk)
What OpenAI and Broadcom announced tonight looks like the telecom financing loop that helped drive the dot-com crash. Recall, billions of dollars and tens of thousands of jobs lost, and a decade to recover the ‘promise’ of the internet.
HOW THE TELCO LOOP WORKED
Vendor lends money to customer. Customer uses the loan to buy the vendor's equipment. Vendor books revenue. Vendor's stock rises. Vendor uses the higher stock price to fund more loans. The money stays inside the circle.
The structure reported tonight fits the pattern. Broadcom finances the chips. OpenAI buys them. Microsoft rents the infrastructure back to OpenAI.
WHO RAN THIS PLAYBOOK LAST TIME
Vendor financiers included Lucent Technologies, Nortel Networks, Cisco Systems, JDS Uniphase, and 3Com. Each extended enormous amounts of vendor financing during the telecom boom.
WHAT HAPPENED TO THEM
Lucent collapsed from $65 per share to under $1. Sold to Alcatel in 2006.
Nortel went bankrupt. Peak market cap of $398 billion went to zero.
Cisco survived, but the stock fell roughly 90 percent.
JDS Uniphase lost about 99 percent of its value.
3Com disappeared into acquisition by HP.
THE CUSTOMERS THAT WENT BUST
WorldCom. Global Crossing. 360networks. Winstar Communications. PSINet. XO Communications. Williams Communications. ICG Communications. McLeodUSA. Covad. NorthPoint. Rhythms NetConnections.
47 telco carriers went bankrupt between 2000 and 2003. The fiber networks they built sat mostly unused for years. The industry called it dark fiber. Empty data centers anyone?
THE MEDIA VERSION
AOL Time Warner became the defining merger disaster of the era. The combined company wrote down roughly $99 billion in 2002. Largest goodwill impairment in corporate history.
WHAT THIS MEANS
All told, $2 trillion in market value was destroyed in the telecom and internet collapse. Roughly 500,000 jobs disappeared. The financing structure behind that collapse resembles what has now emerged around AI infrastructure.
THE BIG DIFFERENCE, MAYBE?
Microsoft, Broadcom, and today's hyperscalers generate real cash flow. Their balance sheets are far stronger than the telecom operators of 2000.
THE SIGNAL TO WATCH
Microsoft is the canary in the coalmine. Literally. If it flies the game is up, if it slows spending or walks from commitments. The Information reported tonight that Broadcom wants Microsoft to commit to 40 percent of the chips before financing the first phase. Microsoft has not signed. Watch this space…
Dotbomb, anyone?
Wow, I haven't heard some of those names in years.
Ed Zitron has been doing great reporting on the outrageous finances of AI
Your ears burning Gary lol? Caitlin is on the left but I have been reading her for years. I’m a right wing nut job I reckon. Anyhow I have come to learn we aren’t so different and people should work together instead of letting ourselves be divided.
Apologies for the political rant. I plugged your page and I believe you speak from the heart.
https://www.caitlinjohnst.one/p/a-few-more-thoughts-on-ai-and-consciousness/comment/254972273?r=nvvvu&utm_medium=ios
Even a nut job can be right now and then :)
That’s what I was thinking too
The only possible conclusion I see is that OpenAI and Anthropic collapse, with the parts getting scooped up by some combination of Google/Microsoft/Amazon/Apple. The real fun would be if one of those four, or a big hardware player like Nvidia, actually gets pulled under the waves by this, too.
We don't need no stinkin' bankers.
Dude… bro… just have Mayatoshi pay for it. or some other broligarchs…. 🤣😂🤣😂🤣
Meanwhile…. LLMs are being used to Discover hydrogen storage molecules…
https://pubs.rsc.org/en/content/articlehtml/2026/dd/d6dd00102e
But that’s not as important as ranting
Money, money everywhere and not a cent to spend.
Gary: Regarding the good influences of AI: I added this below to your Substack note (Topol) but it didn't show up, so here it is again:
In Nature/Cancer (online magazine), they note that AI has helped in early diagnostic predication in x-rays for later-arriving pancreatic cancer. Apparently, AI can spot clues in the x-rays that humans don’t recognize (or can’t), even when their know via other means that something is "off."
But no one save the very extreme said that AI was all-bad. Here is the note:
"AI detects early signs of pancreatic cancer"
"Computed tomography scans taken a year or more before a person develops pancreatic cancer carry signatures of precancer that the human eye can’t easily spot but artificial intelligence can. An AI tool detected early signs of pancreatic cancer with 82% accuracy in scans taken a median of 475 days before diagnosis. 'We knew, based on the biology of the disease … that the signal was there,' says radiologist Ajit Goenka. 'We just needed to find a way to be able to detect it.' The AI tool, called Radiomics-based Early Detection Model (REDMOD), offers “tangible hope for improving outcomes in this challenging disease', conclude the authors."
NBC News | 5 min read
Reference: Gut paper (28 April)
“AI-picycles”
Ptolemy had naught
On what the tech bros got
AI-picycle fund
Is quite a lot of fun
Thanks for this Gary! The xAI bubble shows Broadcom as text and 1.3T