Guest Post by Karl Denninger
… and the market crash from it is going to be record-setting.
Both Google (Alphabet) and Microsoft have softened their forward spend on “more AI.” In addition Microsoft has announced a large round of “voluntary separations” including roles in their AI-linked projects such as CoPilot. Other firms are telling their coders to stop “tokenmaxing”; that is, emphasizing AI use (some of which were stupid enough to make the amount of AI use a formal part of employee performance reviews!)
The question isn’t really about “does AI work?” It “works” like any other program works; the question is whether the insane cost and inherent problems generated when you quite-literally steal everything it “trained” on produce a result, assuming you get away with the stealing, that is worth the cost.
Business, for all its complexity and the voodoo arm-waving that various “analysts” (and CEOs) do comes down to just one question: Will the customer, whoever that might be, see sufficient value in the goods and/or services you sell to buy them from you at your demanded price, and is that demanded price sufficient, all-in, to return a profit?
If the answer is yes you make a profit. If the answer is no then you don’t, and if you can’t adjust to the “no” so as to make it a “yes” your firm ultimately fails.
All size does in the latter situation is make the failure take longer, but fail you shall unless you can turn that “no” into a “yes.”
Many firms, as they get bigger, start looking for ways to make the “yes” compulsory in some form or fashion. This is basically impossible when you’re a small business because there is always another supplier and it is essentially never that whatever you sell can be turned into a compulsory purchase. As you get larger the means of compulsion become more-available and in particular you can enlist government actors to try to drive that as well; an act that is under the laws illegal but no fine in such a circumstance deters the act because the directors and officers don’t have to pay personally out of their wealth and nobody ever goes to prison despite the law often having that written as either potential or even mandatory penalty makes leaning on that path a time-honored way of resolving the problem. But even that only works as long as people can pay; when they can’t even an attempt at compulsion fails.
The 2000-2002 crash (and the nearly ten year period before the former market levels were regained) happened because corrective action was not to be had when it came to corporate actors who got very full of themselves. Their pronouncements then, just as now, were risible. It was not that the Internet didn’t work; of course it did. It was that the cost being dumped into various business claims was not reasonably matched against what they could deliver and thus all manner of “investment” was made yet there was no return on same.
The same thing is happening here except in much-larger size and, just like then, we have politicians trying to take credit for it. Remember Al Gore and his claim that he “built the Internet”? He did not; he was a Senator, not a technologist. He did champion the NSF funding bills but he never plugged a single cable into a single device, nor did he program a single computer, nor did he design a single protocol used on the Internet either then or now. Vint Cerf, Robert Kahn, Dennis Ritchie and a handful of others can reasonably be said to have “created” the Internet and a whole bunch of people, Rick Adams (UUNet), Marty Schoffstall (PSINet) and myself included, can be said to have built it. Al Gore did no such thing at any level. The placard in a TN rest stop honoring him as a Senator for his “contributions” deserves to be melted down as scrap.
In the 1998 to early 2000 time-frame as the claims expanded and became more-ludicrous the circular financing games used to keep liabilities off the formal balance sheets where people could see them and react in the market expanded dramatically. Accounting regulations are a real thing and any firm with over $5m in revenue at the time (now much higher, but still you’ll cross it quite rapidly if you’re a larger business) must use accrual accounting. Accrual accounting recognizes events rather than cash and many people are unfamiliar with it. When a sale is a committed act revenue is recognized and likewise when a purchase is certain so is expense. This is true even if the money hasn’t changed hands yet. Your personal accounting is not done this way; you accrue income when your paycheck hits your hands or bank account rather than on an hourly or daily basis as you work and you incur expense when you pay the credit card, mortgage or insurance bill. A business on accrual accounting incurs the expense when a business credit card has the charge hit it, not when the check is cut to the credit card company. Likewise if I sign a 5 year lease on date-certain and non-cancelable terms the obligation is on my books now every month forward until the term expires, not when I cut the rent checks.
This is a more-fair way to look at a firm’s set of books — or yours as a person, for that matter, but it also is what leads to these sort of circular financing games. If I can set up a “captive” LLC that I then “fund” on a monthly basis but the affiliated and separate entity is the one with legal responsibility to pay then I’ve shifted what is a certain obligation off my balance sheet and into the appearance of a monthly “discretionary” expense. It still has to be disclosed however it doesn’t show up as an immediate liability all the way out because the “captive” firm is not me and it can go bankrupt separately. Or can it? Many of these “captive deals” have terms in them that, in theory, return recourse upon the parent — but then the parent doesn’t recognize it “because, in their opinion, it is very-unlikely they’ll get tapped for it.” That’s legal up until the “very unlikely” becomes something less-favorable but it is deceptive and nobody ever reads the footnotes or takes them seriously, particularly during the enthusiasm part of a bubble.
This is the reasons for the first bullet point in my post on data centers. If a data center is effectively “captive” to META, for example, then hiving that off into a separate firm not only permits META to “divorce” the data center if it becomes uneconomic it also effectively deceives the market as to METAs intended forward spending because while META has to put that in a footnote it converts what is a certain expense into a monthly price which implies it can be walked away from at-will. Technically perhaps it can, and its considered “unlikely” that will change thus its legal to present it this way but that deceives people in the market and forces other people (specifically, unrelated citizens in the area where the center is) to eat the cost if something goes wrong most of the time rather than the company for which it was built and operated.
As soon as you see this sort of pattern develop your hair should stand up. Yes, there are reasonable explanations for it on an individual basis but when the market is filled with this sort of chicanery the reason for it is to deceive those who would otherwise throw a flag on what is a rapidly-expanding bubble in that this is effectively a credit transaction where evasion of the balance sheet impact is a material part if not the entire reason for structuring it that way.
In the end the bottom line is that nobody ever hides good news; they only hide risk or bad news. If you have good news you trumpet it from every available rampart and at every opportunity for obvious reasons. When you do not want the market to look at the totality of what you’re claiming you will spend and then apply an appropriate revenue model to it you hide the liabilities via any means you can devise so as to make it look more-likely than it really is that you’ll be able to actually pay for it.
The problem with all technologies that emerge into the marketplace is that they always become better, faster and cheaper over time. Witness the whining Comcast is now putting up about how fiber to the home providers are operating at an “unreasonable” price. No they’re not; I believed the fiber people had a cost problem originally until I discovered that they had perfected passive optical splitter/combiners which means their operational cost for said equipment, once purchased, is an effective zero other than to cover physical damage as they are not subject to electrical disturbances, semiconductor junction failure due to wear-out or temperature, nor do they have an ongoing expense to power them. I wrote several articles on this over the last few years in the context of both the cable companies and Starlink. The cable company not only has a failure rate for the electronics in their distribution amplifiers and pedestals they have to pay for the power to run them every month as well since they contain active electronics. This means their business model has costs in it (and they’re not small) that the fiber provider does not. Never mind that the cable companies all did this very same thing to everyone else moving data in the early 2000s timeframe including the DSL firms when they built out their plant for Internet service because they cross-subsidized Internet service with pay-per-view movies and events which rode on the very same 2-way infrastructure they had to install. Indeed that much smaller upstream bandwidth is why the cable modem DOCSIS protocol was designed the way it was originally. You see the real price now for Internet service from Comcast because the sales of said PPV movies and events have fallen to an effective zero. Rather than build out fiber plant with that cost-shifted profit for the last 20 years they collected the funds and laughed at everyone else since they had an effective monopoly on said service in many areas.
Oops.
The “oops” in the AI space is going to be much bigger simply because of its current size. There is absolutely nothing that these large firms can do to prevent the proliferation of “open weight” (free) models and as time goes on both cost and data security considerations will drive processing to local machines both in large companies and individuals. Data security considerations are a cost and as the breaches and “misbehavior”, if you will, continues and accelerates (and it will) more and more consumers and firms will either discontinue said use or move it to local processing to contain both cost and risk.
The problem with committed forward expenses in such an environment ought to be obvious. If I’ve committed to spend $500 billion and instead of accelerating demand as loads move into my firm, which would allow me to increase pricing and thus pay for it, I have no “moat” and customers instead move the load to the computer in their house or phone, or even abandon it entirely, I’ve just lit the $500 billion on fire and the revenue with which I was going to pay the carrying costs for that debt also goes up in smoke.
That’s exactly what happened in 2000 and it is starting right now once again — for the same reason.
Tweet
Continue reading...
[ H/T The Burning Platform ]
… and the market crash from it is going to be record-setting.
Both Google (Alphabet) and Microsoft have softened their forward spend on “more AI.” In addition Microsoft has announced a large round of “voluntary separations” including roles in their AI-linked projects such as CoPilot. Other firms are telling their coders to stop “tokenmaxing”; that is, emphasizing AI use (some of which were stupid enough to make the amount of AI use a formal part of employee performance reviews!)
The question isn’t really about “does AI work?” It “works” like any other program works; the question is whether the insane cost and inherent problems generated when you quite-literally steal everything it “trained” on produce a result, assuming you get away with the stealing, that is worth the cost.
Business, for all its complexity and the voodoo arm-waving that various “analysts” (and CEOs) do comes down to just one question: Will the customer, whoever that might be, see sufficient value in the goods and/or services you sell to buy them from you at your demanded price, and is that demanded price sufficient, all-in, to return a profit?
If the answer is yes you make a profit. If the answer is no then you don’t, and if you can’t adjust to the “no” so as to make it a “yes” your firm ultimately fails.
All size does in the latter situation is make the failure take longer, but fail you shall unless you can turn that “no” into a “yes.”
Many firms, as they get bigger, start looking for ways to make the “yes” compulsory in some form or fashion. This is basically impossible when you’re a small business because there is always another supplier and it is essentially never that whatever you sell can be turned into a compulsory purchase. As you get larger the means of compulsion become more-available and in particular you can enlist government actors to try to drive that as well; an act that is under the laws illegal but no fine in such a circumstance deters the act because the directors and officers don’t have to pay personally out of their wealth and nobody ever goes to prison despite the law often having that written as either potential or even mandatory penalty makes leaning on that path a time-honored way of resolving the problem. But even that only works as long as people can pay; when they can’t even an attempt at compulsion fails.
The 2000-2002 crash (and the nearly ten year period before the former market levels were regained) happened because corrective action was not to be had when it came to corporate actors who got very full of themselves. Their pronouncements then, just as now, were risible. It was not that the Internet didn’t work; of course it did. It was that the cost being dumped into various business claims was not reasonably matched against what they could deliver and thus all manner of “investment” was made yet there was no return on same.
The same thing is happening here except in much-larger size and, just like then, we have politicians trying to take credit for it. Remember Al Gore and his claim that he “built the Internet”? He did not; he was a Senator, not a technologist. He did champion the NSF funding bills but he never plugged a single cable into a single device, nor did he program a single computer, nor did he design a single protocol used on the Internet either then or now. Vint Cerf, Robert Kahn, Dennis Ritchie and a handful of others can reasonably be said to have “created” the Internet and a whole bunch of people, Rick Adams (UUNet), Marty Schoffstall (PSINet) and myself included, can be said to have built it. Al Gore did no such thing at any level. The placard in a TN rest stop honoring him as a Senator for his “contributions” deserves to be melted down as scrap.
In the 1998 to early 2000 time-frame as the claims expanded and became more-ludicrous the circular financing games used to keep liabilities off the formal balance sheets where people could see them and react in the market expanded dramatically. Accounting regulations are a real thing and any firm with over $5m in revenue at the time (now much higher, but still you’ll cross it quite rapidly if you’re a larger business) must use accrual accounting. Accrual accounting recognizes events rather than cash and many people are unfamiliar with it. When a sale is a committed act revenue is recognized and likewise when a purchase is certain so is expense. This is true even if the money hasn’t changed hands yet. Your personal accounting is not done this way; you accrue income when your paycheck hits your hands or bank account rather than on an hourly or daily basis as you work and you incur expense when you pay the credit card, mortgage or insurance bill. A business on accrual accounting incurs the expense when a business credit card has the charge hit it, not when the check is cut to the credit card company. Likewise if I sign a 5 year lease on date-certain and non-cancelable terms the obligation is on my books now every month forward until the term expires, not when I cut the rent checks.
This is a more-fair way to look at a firm’s set of books — or yours as a person, for that matter, but it also is what leads to these sort of circular financing games. If I can set up a “captive” LLC that I then “fund” on a monthly basis but the affiliated and separate entity is the one with legal responsibility to pay then I’ve shifted what is a certain obligation off my balance sheet and into the appearance of a monthly “discretionary” expense. It still has to be disclosed however it doesn’t show up as an immediate liability all the way out because the “captive” firm is not me and it can go bankrupt separately. Or can it? Many of these “captive deals” have terms in them that, in theory, return recourse upon the parent — but then the parent doesn’t recognize it “because, in their opinion, it is very-unlikely they’ll get tapped for it.” That’s legal up until the “very unlikely” becomes something less-favorable but it is deceptive and nobody ever reads the footnotes or takes them seriously, particularly during the enthusiasm part of a bubble.
This is the reasons for the first bullet point in my post on data centers. If a data center is effectively “captive” to META, for example, then hiving that off into a separate firm not only permits META to “divorce” the data center if it becomes uneconomic it also effectively deceives the market as to METAs intended forward spending because while META has to put that in a footnote it converts what is a certain expense into a monthly price which implies it can be walked away from at-will. Technically perhaps it can, and its considered “unlikely” that will change thus its legal to present it this way but that deceives people in the market and forces other people (specifically, unrelated citizens in the area where the center is) to eat the cost if something goes wrong most of the time rather than the company for which it was built and operated.
As soon as you see this sort of pattern develop your hair should stand up. Yes, there are reasonable explanations for it on an individual basis but when the market is filled with this sort of chicanery the reason for it is to deceive those who would otherwise throw a flag on what is a rapidly-expanding bubble in that this is effectively a credit transaction where evasion of the balance sheet impact is a material part if not the entire reason for structuring it that way.
In the end the bottom line is that nobody ever hides good news; they only hide risk or bad news. If you have good news you trumpet it from every available rampart and at every opportunity for obvious reasons. When you do not want the market to look at the totality of what you’re claiming you will spend and then apply an appropriate revenue model to it you hide the liabilities via any means you can devise so as to make it look more-likely than it really is that you’ll be able to actually pay for it.
The problem with all technologies that emerge into the marketplace is that they always become better, faster and cheaper over time. Witness the whining Comcast is now putting up about how fiber to the home providers are operating at an “unreasonable” price. No they’re not; I believed the fiber people had a cost problem originally until I discovered that they had perfected passive optical splitter/combiners which means their operational cost for said equipment, once purchased, is an effective zero other than to cover physical damage as they are not subject to electrical disturbances, semiconductor junction failure due to wear-out or temperature, nor do they have an ongoing expense to power them. I wrote several articles on this over the last few years in the context of both the cable companies and Starlink. The cable company not only has a failure rate for the electronics in their distribution amplifiers and pedestals they have to pay for the power to run them every month as well since they contain active electronics. This means their business model has costs in it (and they’re not small) that the fiber provider does not. Never mind that the cable companies all did this very same thing to everyone else moving data in the early 2000s timeframe including the DSL firms when they built out their plant for Internet service because they cross-subsidized Internet service with pay-per-view movies and events which rode on the very same 2-way infrastructure they had to install. Indeed that much smaller upstream bandwidth is why the cable modem DOCSIS protocol was designed the way it was originally. You see the real price now for Internet service from Comcast because the sales of said PPV movies and events have fallen to an effective zero. Rather than build out fiber plant with that cost-shifted profit for the last 20 years they collected the funds and laughed at everyone else since they had an effective monopoly on said service in many areas.
Oops.
The “oops” in the AI space is going to be much bigger simply because of its current size. There is absolutely nothing that these large firms can do to prevent the proliferation of “open weight” (free) models and as time goes on both cost and data security considerations will drive processing to local machines both in large companies and individuals. Data security considerations are a cost and as the breaches and “misbehavior”, if you will, continues and accelerates (and it will) more and more consumers and firms will either discontinue said use or move it to local processing to contain both cost and risk.
The problem with committed forward expenses in such an environment ought to be obvious. If I’ve committed to spend $500 billion and instead of accelerating demand as loads move into my firm, which would allow me to increase pricing and thus pay for it, I have no “moat” and customers instead move the load to the computer in their house or phone, or even abandon it entirely, I’ve just lit the $500 billion on fire and the revenue with which I was going to pay the carrying costs for that debt also goes up in smoke.
That’s exactly what happened in 2000 and it is starting right now once again — for the same reason.
Tweet
Continue reading...
[ H/T The Burning Platform ]
