Just about everyone is watching the financial markets, either wondering how much better they can get, or when the AI financing bubble is going to crash. Dough Dawson, from POTs and PANs, joins Tom and Russ to talk about the logic of a coming crash, and consider where engineers should be looking for potential future career fields to weather the storm. https://media.blubrry.com/hedge/content.blubrry.com/hedge/hedge-322.mp3 download
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
- Join us as we gather around the hedge, - where we dig into technology, business, and culture - with the finest minds in computer networking. - Hello, Tom. Hello, Audrey. - And - I'm glad to see there is stuff on - your whiteboard, Tom. - Yeah. And Yeah. The rocket is gone. Was - it fun? Did it go off?
- It's just off screen. - It's still there in its box and its - potential energy. - You just moved it to fool us. - That's that's what you did. Alright. - And today, we have Doug Dawson from the - other side of the mountains. - He's on the he's on the right side - of the mountains. I'm on the left side - of the mountains. - Well, depends where you're standing. So, yeah, - Asheville pretty much thinks they're on the left - side of everything. So
- They do. But... Yeah. So, yeah, Doug is - just over the mountains for me, and we - were gonna talk today about - AI. - And, - you know, it's an interesting... Not not AI. - We've been talking a lot about AI, but - everybody has. So whatever. - But, - about the potential for a crash, - and I may have to move this show - up in the schedule a little bit to - beat the crash - before it happens. - Yeah. This this show becomes worthless the week
- after the crash. Yeah. Yeah. - You predicted it after the crash. - Impressive. - That was quite brave of you. Yeah. So... - Yeah. - So let's talk about it because you wrote - an article about this, I think, right, on - your pots and pans blog, which I did. - By the way, you're not following pots and - pans, you should. - So - talk to us a little about what you're - seeing or what you're thinking. - I think we can add more fuel to - the fire over the course of the discussion.
- But what are you seeing or what are - you thinking? - Well, first of all, let's put it in - context what I'm talking about. I mean, AI - as a technology is gonna be just fine - in the long run. I mean, it's a - wonderful technology. It does all sorts of pretty - amazing things. - What we're talking about crashing - is the big data center model. - I mean, the big data center model looks - to be unsustainable. - So there's lots of other ways to do - AI that doesn't take these giant data centers. - So, - and so... And they're... Interestingly, you're... Before we - got on here, we... You said there was
- a lot of articles coming out. There was - a... There's probably been... I probably read 40 - reports from various Wall Street analysts. - One just came out on Tuesday from Bain, - if you know who they are, if you - remember the Bain guys. Yeah. Yeah. And that... - So that's the that's the newest one. It - very much says what the other one does. - What's interesting is Bain is very - big - pro on AI, but but they look at - the numbers and they just shake their head. - So here's what they said. - They said by 2031,
- the current... The big data center industry would - need $6,000,000,000,000 - of revenue - a year to - sustain itself. - And - so... And now to put that number into - context, - The US - whole federal budget right now is 7,400,000,000,000. - Wow. Now that... Now the 6,000,000,000,000 is a - worldwide number, - but that doesn't really matter. That's just... You - know, that... It's like, well, we need the - second biggest economy after The United States just - to run AI. I mean, it's like, this
- is not gonna work. Yeah. And and what - Bain did, Bain was very optimistic - more than the other folks. So here's what - Bain estimated. - They estimated the consumer - AI by 2031 - would generate 200 to 400,000,000,000 - a year. - So from, you know, from people like me - and you buying the $200 subscriptions. - And then they said that corporate, - and government AI would generate 1 to $1,400,000,000,000 - a year, meaning that even if... And those
- numbers, which I'll talk about a second, sound - both ridiculously high to me. But even if - they achieve that, there's still 4,200,000,000,000 - short - a year. - It's like, so not... So let's talk about - why they need that big number. - And and the answer is fairly easy. It's - not people. It's the cost of running those - data centers. Those things are expensive as hell. - And it's not just the operating cost. They - do suck a lot of power. They do - suck a lot of water. - But the worst thing is the cards in
- the data centers have gotten ungodly expensive. - I mean, to do a rack... Just to - do a shelf on on one of their - data centers is, like, a $100,000 - now. - And so by the time you put all - the components in there... And and these things - go on and on and on. I mean, - they're building these things, you know, the size - of three Super Bowlmarts. You know, these are - just monstrously large buildings. - In fact, one one of the ones that's - being built right now is the size of - of, you know, of Central Park in in - New York City. It's like, this is an
- insane building. - Yeah. And so and so so they they - made billions of dollars worth of hardware - and thing... And those things are only lasting - three and four years, so they have to - keep buying. That's what I was gonna say. - Yeah. Then... Right. Exactly. - That's what I was gonna say is that - the life cycle is so short, and yet - they're trying to make it sound like it's - almost a permanent investment. Like, they go to - a community, and they say we're putting in - this data center. And we're gonna add a - 100 jobs to the community, - and they don't say how long. But the
- fact is is that data center buildings themselves - are not eternal things. - Right? Yeah. They're they're... - You you you have a change in power - requirements or a change in cooling requirements, and - it's often cheaper to build a new building - than it is to retrofit the one you - have. - And Yeah. And that becomes... You end up - with abandoned stuff all over the place. - Now there are scientists and engineers working on - making the cards last better because the problem... - The reason they're burning out is they - they just let them run. And so when...
- So those... These things are running twenty four - seven at full capacity, and they generate a - lot of heat. - You know, there's smart strategies to say run - them at 60% - load, - which takes... - It's a really complicated set of programming to - make every chip do that. - So it's just not like you can't just - go to the data center and go slow - my chips down. But but that would probably... - That might even double their shelf life, but - then it also cuts the computing power now. - Right. So... But, you know... And and - the... Now...
- So so the money is just not making - any sense. This makes absolutely no sense. We... - Now let's let's talk about the current revenue. - So - this year in 2026, - Anthropic - and OpenAI together will make one point... 105,000,000,000 - in revenues, total revenues. - So somehow they have to grow from a - 105... - And and then I thought I don't... I - didn't go look up the number for, for - Grok, but it's not very big. So when - you add all those together, there... It's a
- really long way to get the 6,000,000,000,000. - It's like that... You know, it's just really - not there. - So so now we can just talk about - what's wrong even if they had a path - to get there, which they don't. Let's talk - about their big number. Their big number from - Bing was 1,000,000,000,000 to 1,400,000,000,000 - in corporate revenues. - There's a whole lot of these studies I - read in the last few months that say - corporations have soured on the big AI model
- because they're too expensive. - And they're... You know, to ask the major - corporations in the world to ship... To - to spend a trillion dollars a year on - AI, - it's like, you know, again, - most of them all added altogether don't even - have that much revenue. That's a lot of - money. That's a lot of money for me. - What's the ROI on that? Is that... That's - part of the problem is what's the ROI - on that? If I'm if I'm a big - company and I'm spending, - you know, 5,000,000 a year, 10,000,000 a year, - am I getting 5 or 10,000,000 back? And - then the other question is, just like when
- you look at cloud versus on premises, - this whole thing might actually drive - a movement back into on premises networking - and on premise data centers because - the cost - of them running these huge data centers and - spreading it everywhere. - And then not only that, but, like, all - the data you're giving them - to do this Right. That Right. That's... I - think that's very problematic. - If I Well... And and you just hit
- on you just hit on the problem. - The reason it's so expensive to run these - is they've tried to load all the data - of mankind has with these data centers. - So just for the just for the data - center to find what it is you ask - it, - they are now... You know, the public here, - as we talked today, unless the guys tune - and use AI, they don't get it. But - the current OpenAI model uses 400,000 - tokens just to get - started, just - to find the data - to try to answer your question. And that
- includes... If that's my mother going, - you know, could you tell me if the - bus schedule changed for tomorrow? It's gonna cost - 400,000 tokens because it looks through the whole - world of data. It doesn't know how to - go straight to that. And so, you know, - so that I... They... They've overtrained them. They... - This idea that it has to have all - the information in the world in there just - to do that stuff is crazy - because companies want stuff. Bank of America just - wants to run stuff on Bank of America - data. They don't care about, you know, they
- don't care about Charles - Darwin's books and and, you know, statistics about - volcanoes and everything that's in these data centers. - Right? So so the the way they... They're - gonna have to they're gonna have to back - off of that model, - because it's just - incredibly inefficient, and it's getting more inefficient. - The smarter the models get, the worse they - are at at getting started. So the older - ones were actually way more efficient than the - new ones are. So So, Greg, I wanna - I wanna Yeah. The math is so bad. - Yes, sir.
- I I just wanted to ask, have you - noticed any studies about, - impact on revenue for companies for large corporations - spending lots of money? Like Yeah. Is there - is there any... Yeah. What's the evidence there? - Well, here's a great story. So Uber - Uber had a budget for AI this year, - and in June, - they they had a $40,000,000 - spend - plan for Uber headquarters. - They hit that number in June, and they - just turned it off. They go, we can't - afford this. - So they hit their annual spend in half
- a year, and now they're not gonna use - the big open AI anymore. - Because the whole point of them using the - 40,000,000 - was to replace people, and they weren't replacing - them. So not only did they spend the - 40,000,000, they they have nowhere near $40,000,000 - of savings. - And so the the head of... And this - is a tech company, and they're just going, - you know, our analysts are just doing it - to to save them time on their work. - It's like, hey. That's... We pay them to - do that work. They can do the damn - analysis themselves without spending $40,000,000.
- Right? And so so they so they so - they candid it. And I forget who did - the study, - but - someone recently did one that said something like - 80% of corporations are double thinking whether they - can use this at the corporate level, and - it's just not panning out at the bottom - line for them. - So, you know, where AI comes useful, - like, universities - love this. I mean, if you're doing drug - research or or something, as long as someone - will give you the money, - you know, for the research, it's really wonderful - for stuff like that. But for, you know,
- for your for your - regular analyst who's doing budgets in a corporation, - you can let them use AI to do - that. It's, like, costing way more than the - cost of than the cost of... It's like, - what? Be cheaper to hire another analyst or - to or to go get one that actually - knows what they're doing. - So, you know, so now here's the here's - the other math because... - I mean, a friend of mine is an - AI scientist, and I've known him forever. And - and, - he's been he's been doing AI since the - old old days of the seventies when it
- was something very different than now. And and, - he says that when you... That the average - person who gives them a $200 a month - for their home subscription is costing OpenAI between - 30 and $50,000 - a month in cost to give you your - stuff. - I mean, that's how unsustainable this is. Obviously, - they're not gonna charge you 30 to 50,000, - but they're off... They can't charge the corporations - that either. So... - And that's that's the gap to get to - the 6,000,000,000,000. - It's just an... It's a ridiculous - mess.
- Now - why are we there? - That... That's the second big problem. - We have, - you know, y'all... We... I... Russ is an - intro. Are you in the telecom industry, Tom? - I assume you are. Yes, sir. Yeah. Yeah. - Are you... We all remember what happened in - 2000. - Exactly the same thing. We had we had - we had Lucent and Nortel and a couple - other big companies - giving... - The vendors were giving money to the ISPs - to turn around and give back to them
- to buy their equipment. And so they were - booking huge amounts of sales, - but but then they... You know? And it - just was a circular financing, - and that's really what broke. That was the - number one item that broke the the the - the burgeoning sea like world, - because they were just handing out money, and - it was ridiculously - spent - way too much money. People were told to - go fast, so they did, and that's very - wasteful. And so that's where we're at again. - I mean, - there's just this round robin. I
- I should have saved it somewhere, but there's - this round robin diagram that shows - Microsoft going to OpenAI - going to NVIDIA, and the money just goes - around and around, and they're all booking it - to revenue. - So it gets booked three or four times - as industry revenue. - And it's like... And none of it's revenue - because - what happened was the basis of that money - was somebody borrowed it. It's like when you - borrow money, that's not revenue to anybody. So - even if you turn around and give it - to somebody else, you still have to pay - those loans off. So so that's the other
- one of the... Yeah. Go ahead. There was - another story this morning about... I believe it - was Amazon - who was trying to raise $7,000,000,000,000 - to... Maybe maybe it was 7,000,000,000. - I don't know. Whatever. Some stupid number. And - maybe it was 7,000,000,000. - They're trying to raise $7,000,000,000 - to sell off, - a huge chunk of their GPUs - and then lease them back so they can - continue using them as a way to, like, - build their their their their bottom line, build
- their their sheet. Right? Their financial sheet. And, - like, - guys, - who... First of all, the market is not - made of infinite amounts of money. - We cannot... - People cannot invest this kind of money - in these things forever. It's already dried up. - I was reading QTR the other day, and - they were saying... QTR was saying that... Quote - the Raven, and he was saying, look. - If you take the big seven out of - the S and P 500 right now, it's - been losing money for the last
- however long... A long time. It's only the - big seven floating it. And part of that - is just simply people - not investing in other things so they can - invest in the big seven. - Yeah. Well, that money's gonna run out. I - don't know. Like, - who has another 7,000,000,000 in the bank? - Let let alone let let alone settle. Now - The US spending alone for data centers this - year is 600,000,000,000. - Yeah. The world... The worldwide spending on data
- centers, just 2026 - is over $10,000,000,000,001,000,000,000,000. - Next year, if you could look on paper, - the plans that they've all announced are 1,300,000,000,000. - Now - people are backing off of data centers now, - so it's starting to cool down, which is - why the crash is probably coming. So they're - never gonna spend the 1.3, but but these - are really unbelievably - large numbers. So... Yeah. Now what you described - a minute ago was called... I... On top - of being a tech guy, I also... And - what... Gosh. Now it's about fifty years ago, - I passed the CPA exam. So I'm also
- an accountant even though I never I never - actually did that for a living, but but, - you know, while I have that background, what - you're talking about is called the backstop clause. - It's a funny money game that they're playing. - What they're doing now - is is they are - booking - capacity, - and they're putting it on the books as - an asset. - And and everyone in in the AICPA world - who are the guys who are - on the county and the audits of the - world are going, that's not an asset.
- Capacity is not an asset. Money is an, - you know, is an asset. The cards you - buy are an asset. That's what you're talking - about Microsoft doing. They're trying - to invent an asset because they want to - make their balance sheet look better. Yeah. So - they're taking they're taking a cost, - not even a cost, a theoretical cost - that... From that big circular chain of stuff - we're talking about, and they're trying to book - that to make their books look better. - Now - here's something. Some of the reports I read, - this is really interesting - because you talked about the big seven.
- We think most of them have actually shielded - themselves from a whole lot of the loss - because they know that crash is coming. - So when the crash does come, they're gonna - get hurt. No doubt about it. Nvidia probably - is the one who takes it the worst - in the shorts. - But what they've done is they're not the - ones pay... We say there's 600,000,000,000 - going into data centers. They're not paying for - that. - Do you wanna know where that money's coming - from? You're you're not... You're gonna hate this. - Oh, I'm saying private - equity and,
- It's not it's not even come... It's retirement - accounts. It's not even private equity. It's coming - from retirement funds, like state retirement funds. This - is Enron all over again. - The little guy is gonna get killed in - his 04/2001 k when this stuff crashes. Now - these... The big seven, - Microsoft, NVIDIA, you know, all those guys, they're - gonna get hurt, but they're not gonna take - the brunt of it except that none of - us will ever trust them again - Yeah. For ten years. You know? We'll have - an... After AI crashes, there'll be a ten - year AI desert where people like, I won't
- touch that stuff with a 10 foot pole. - Just like happened to telecom. - I mean, boy, there was no new investment - after 2000 for quite a while. So Yeah. - So we're gonna see that again. So so - these these guys have had... They're... They pegged - their money back, so they're not even gonna - get hurt. A lot of these companies are - actually... - Or they're being accused. We don't know that - they're being accused of buying shorts of their - own stock so they're gonna crash as it - evens out. - That's crazy. - That is crazy. - So so from an individual
- networking... For the individual engineer's perspective, - the the first thing that that you might - think is, number one, - I think there's a lot of things that - go through my head, you know, if I'm - if I'm working in this space, which by - the way, I do work in this space. - So I do think about this. - One of the things I think that runs - through my head is, - oh, no. People won't be building these big - networks. The prices are gonna go down, - and the demand is gonna go down, and - therefore, there will just be less jobs
- in the world of engineering. - What what do you think? Is that right, - or do you think that's that's kind of - off base? - I don't think so because almost all this - much money... I mean, here's how the money - is being spent. - There's certainly jobs created when they build these - data centers. You know, the guys who do - construction and bulldozers and all that stuff are - making money, but that's really not a big - pile of people to do. It's it's not - a whole lot different than building a giant, - you know, UPS warehouse. I mean, it's just - it's just a giant warehouse.
- The real cost is in the card. So - the real the real cost is in the - electronics on the inside, - and then the real operating cost going on - is - for electricity and and water. - And so - when we're not spending that money anymore for - that, I don't see how that hurts the - other jobs. Because right now, - one of my friends - owns a phone company in a county - where the county next door, they built one - of these data centers in Arkansas. - He can't find an electrician
- because... Because they didn't have that many to - start with, and all the electricians went to - work as a full time job at the - data center. That's one of the two kinds - of jobs they create. They take care of - the - they take care of the generators and all, - you know, all the stuff and and the - wiring and all that stuff. So the... So - so, you know, those folks will go back - to the economy, and we'll actually have we'll - actually have electricians who could build houses again. - Yeah. - Yeah. So I I don't think these jobs - will be lost. - AI is not generating jobs.
- The actual AI industry is not generating jobs - at all. The typical data center has between - twenty and fifty employees. It's not a big - number. So Yeah. - Yeah. And so the other thing that as - a network engineer or as an engineer in - general, I might think or worry about is - all this automation and all these systems I've - built, - they - run on all of these things, - all these public models. - Like, what am I gonna do when all - of that... What's the disruption
- when all of that starts to either go - away or just... It won't go away. Let's - be honest. It won't go away. It will - stop being maintained. - That's that's what will happen. There will be - no new features added essentially - over time, and the quality of the answers - will become worse and worse - as, you know, as it degrades. Because these - systems, they degrade over time. They... They're like - any other software. They require constant maintenance to - keep them up. And when the money goes - away, the maintenance goes away.
- And people, like, no big vendor is just - gonna come along and say, okay. Well, we're - sorry you bought our AI network management product. - It's going to be shut down. That'll never - happen. They want the revenue from those rentals - and stuff. But I do think that over - time, the quality will reduce. - So - from that perspective, do you see any, - like, - may I actually wanna dis... I wanna disagree - with you. Okay. If if we have an - AI crash, - software that runs directly in the big data - centers is gonna stop overnight. It's just gonna
- be shut down - because because the AI companies are gonna shut - down a whole bunch of the data centers. - And and if you have a - software - that actually works in those data centers, - you're you're out of business. And now - this is a crazy number. There's four... 40,000,000 - apps have been written for AI. That that - was, like, a month ago, so now it's - probably 45. - None of them make any money, but the - fact is there is all these programs. - You know? Talk... You know?
- Medical research, talk to your vet, you know, - schedule your bus rides. All these people have - these functions running on data centers. And if - they're running into made data centers and they - crash, the next morning they go to use - those, they're not there anymore. They're gonna go - completely dead because they're gonna turn off the - power at the data centers. What's gonna crash - is they can't afford to pay the bills - with data centers. Again, anthropically, these guys don't - own these data centers. The data center guys - are gonna go no revenue, - no no computing. - The other funny thing about all this is
- we keep talking about all these data centers. - Over half of them don't have the chips - in them yet. They're still waiting. There's a - two year backlog on these chips. We have - a whole bunch of empty data centers that - don't even have chips in them yet. But - but... No. I think... Now - what you're talking about that we keep going, - I have three friends just in this area - where we live who have built their own - AI data centers, which just means one rack - of of the cards that they run for - their own business. - And so those guys will keep running forever,
- but the problem will be - if if the anthropic truly crashes, when those - cards wear out, they may never find any - replacements - because that's not the same cards. There's gonna - be a flood of the market of the - big cards from the data centers, but that's... - Those are really not practical for me and - you to put into our own little rack - of equipment because they suck too much power. - So these guys are sort of running, - you know, three or four generations of NVIDIA - old. And so they're... And they're not even - making those now anymore. So so, you know,
- the little data centers... The new industry, if - the big IA crashes, little AI, somebody has - to step in and make those cards again. - In the long term - model is there will be hundreds of thousands - of these little data centers in the world, - and and whoever supplies them will make good - money. - Kinda like when NVIDIA made - cards for game - game. Yeah. But that's what... But now they're - they're gonna go from big big Nvidia back - to little Nvidia, and it's a permanent business - if they don't if they don't die in - this crash. Right? There's a good chance they
- die. So - now in our industry, here's an interesting set - of people who have a lot of risk. - Zayo - and - and Lumen are... And, and and the old - CenturyLink - are are building... They're building massive amounts of - fiber in an... On the interstate networks - to carry all this traffic. - Because when you make a request to a - data center, it literally... You might use seven - or eight or 10 or 20 of these - data centers. So the amount of traffic bouncing - between them is gigantic. So they're spending a - fortune
- on these new routes. - And if if there is an AI crash, - their traffic's just gonna go away and and, - you know, to some much lower level. And - they're and they're... I mean, that's all... That - might be all that it takes to to - finally put CenturyLink under. - They're they're on the edge anyway. - So... Yeah. I... Yeah. I remember, - at a place that I worked a long - time ago, there was, there was a data - center shell that was built out, - in... It was like '99 or something. Somebody - took an old coke bottling plant, and they're
- like, everybody's gonna be building, you know, they... - So they... So they built the shell and - then everything crashed, and it was just a - shell. And then the place I worked went - and bought it at a nice discount, and - it was like the property was already developed. - It was awesome. It was perfect. - And, - and they were able to take that asset, - and and they would never would have been - able to build it themselves. - But because of, you know, the the aftereffects - of the crash, right, - they they found a lot of utility. Do - you see anything similar to that happening here - with these assets? I mean, yeah, we can't
- run them in our garage or whatever. - But is there, do you think there's a - secondary effect where that same sort of thing - happens? - I don't know that the buildings are ever - gonna be useful. Have you been to one - of the day centers, either of you? Just - on just on the just on the outside - of the building, they have, forty and fifty - tractor trailer size generators. - You have to sit... You have to see - one of these to believe them. And so - when they don't need those generators anymore, - GE Capital and whoever holds the lease on
- those generators is gonna come and drag them - away because they will be reusable somewhere in - the world. Right? - And so - so that... It's a very weird building. These - were... These are being built, and they have - this gigantic water circulation system and all. They're - really not friendly - towards a normal use. - I I I kind of... I'm at the - Northern Virginia at the two thousand crash. - I remember the row of dead sea legs, - and it took ten to fifteen years before - people moved into those buildings. - I kinda see the same thing. So, I
- mean, some of them will get reused, but - but, you know, when you have one of - these giant data centers in rural Arkansas, there's - no use for it. It's - never gonna get used again. So some of - them will get used, but they're they're not... - They can't easily... It'd be... Cost more money - to convert one of those into a Walmart, - you know, shipping center than it would be - to build a new one, I think, because - they're not they're not laid out for any - of those functions. - So... Yeah. - I I mean, it's a very wasteful, weird - thing that we're doing here. So... Yeah. Yeah.
- Interesting. - Yeah. So I know I saw some statistics - recently, and I was looking for them, and - I can't find them. But something like a - 119,000,000, - a 190,000,000 - miles, something crazy of new fiber needed in - The United States - in the next - ten years or something. - Like... So the fiber companies that are building - up for this are going to be hit - if this goes down. That's going to be - something that's going to happen. What? - Yes. Well, let's talk about Corning.
- So why... I was talking about, you know, - Zayo and Interstate Fiber. That's the small piece - of the fiber. - The data center that they're building - at Louisiana that's almost finished has 1,000,000 feet - of fiber in it. - That... That's where all the fiber's going. It's - going between racks. - All these racks are not... - You know, in the old days, we would... - Inside the data center, we would have had - a lot of CAT six cables. Right? These - are all fiber. - Card to card to card to card, there's - shelf to shelf, rack to rack. It's all
- fiber everywhere. - And so - so that... They're... They are sucking all of - the - fiber capability - from people like Corning because they're paying a - premium price. And so now there's a shortage - of... If you're an ISP wanting to build - fiber now, there's... We just went from... We - finally got over the we finally got over - the whole COVID - supply chain stuff. We got back to normal, - and now the now the the wait times - are back up again. We're up to six - to nine months to get fiber, and they're - warning it's gonna get worse because all the
- fiber is going to the data centers. Now - Corning might take it in the shorts. They're - building three brand new factories to manufacture fiber. - At those quantities, if AI crashes, no one's - gonna want that fiber. - Yeah. Yeah. That was that was my other... - Like, that is another area that I worry - about. Now Yeah. On the on the other - side, - right, if I'm if I'm an individual engineer - looking at this, - it might... Like I... Like you said, you - know, there are all these... The small AI, - which I do have one one problem or
- one thing I don't necessarily... I'm I'm a - little bit. But - the the average engineer also has to think - on the positive side - that if people are putting in, again, if - there's a repatriation - or if there's a little... We're gonna start - building little data centers for each company that's - gonna run AI or each of these access - providers. Right? Like, WISP and stuff like that - start saying, well, I wanna sell AI to - my customers. Right? I wanna sell bus schedule
- AI to my customer or whatever it is - because that adds value, and I can charge - a little bit more. It adds stickiness because - I know, - like, a lot of the problem with access - providers is there's no stickiness. - People roll constantly. - There's a constant churn. And so if if - if you had somebody like... - I don't know, - big name ISPs used to - provide voice mail because it was a sticky - thing. Once my voice mail and email and - everything else, - maybe this becomes a service like that. So
- on the positive side, - engineers have to be thinking - maybe the big stuff goes away, - but there's a lot of little stuff to - be built - if I'm in the right market. And if - I do the right thing, - maybe I can get into that market - and build something out of that. - Is that - And and and that's what... I... I'm thinking - that takes a decade because right now, again, - I said I know three guys who have - built their own. - The learning curve just to get it working - for a guy who really knows how AI
- works is, like, a year and a half - or two years. - And and there's no off the shelf anybody - to help you do that. So the... And - industry has - to grow to say, I can come and - just give you a preloaded little - data center that works. These guys have to - figure out everything. They gotta figure out where - to get the cards, how to train the - model, how to get rid of the hallucinations. - I mean, they have a lot of work - to do before they can actually do anything - useful with it. - So so I I believe that is the - future model. You just... I believe that's the - ongoing survivable model.
- Now the other thing that's happening we don't - really know about. - I... We know Bank of America has their - own AI. We, you know, we know Ford - Motor Company probably has their own AI, and - we don't know how they're doing it because - they're not talking. - So so they're doing this on a bigger - scale than my three buddies, - but they're not doing it on the scale - of the big data centers. And the reason - they're all doing it is one, cost. But - two, they're not about to get their data - to open AI. Like, no. You're you're not - gonna get my stuff. Because once you put
- it in the cloud, it's there for everybody. - So nobody wants their stuff in the cloud. - You know, why I think they're all gonna - be surprised that one of these days is - find out that when all their analysts went - home in the evening and did their spreadsheets - on on their home OpenAI, - all their stuff is in the cloud anyway. - Yeah. Oh, yeah. Yeah. Definitely. - But but, you know, but I... But, you - know, so these... There is there is a... - That industry of people building their own who - are not mom and pops. They're bigger now. - And and so we don't know how many - of those there are. There's no statistics,
- and none of them are talking about it. - I mean, you... I can't find any - literature articles, mench... People don't even talk about - it in their annual reports. - They have an AI department, and you don't - know if they're building their own or they're - buying it from Entropic. We just don't know. - So... But I I think there's a lot - of those. - I - mean, my butt... One buddy who's... How... Always - been in that business, he was pretty sure - that that, the Bank of America, as an - example, jumped into a great way to do - their own. But but but we don't have - a group. So there will be survivors of - the current model. But, again,
- Bank of America probably can run their whole - business on three racks of these cards. I - mean, we don't need this giant data center - when it's just... Let's track all the money - in and out of my customers' accounts. - These these chips are amazingly fast, - so they can do a lot of work - with a fairly small setup. So Yeah. So - they might have a million doll... They might - have a million dollar data center, which is - all they need. So Yeah. I was gonna - say it, but but still, it's - a set of racks sitting in - an existing data center
- that has to be connected, - managed, dealt with, - etcetera, etcetera. - And and - there will be some value or a lot - of value - in engineers who learn - how to - manage and maintain such systems at that scale. - It seems to be Absolutely. - That's... Because there's not that many AI experts - today that actually know that kind of stuff. - So there... That's a real growth area, I - think. - Absolutely. - So, yep. - So... Yeah. Interesting.
- Yeah. So I don't know. I mean, I - I see the I see the QTR and - the Hedgie, and I follow Hedgie on x - and all these other people. - And they're all, like, - just explaining - the financing and what's going on - and what it's... How it's warping the market. - And and to some degree, this worries me - as well as the market warpage - and the the negative fallout - of... - What was the .com? It was, like, 1% - of the American economy - or 2%. So it wasn't... That wasn't that
- much. - AI - right now is, like, three to 5% of - the American economy. Yeah. Somewhere three to somewhere - three to five. That's what everyone says. And - that's a lot. Yeah. That's direct. That's not - electricians, - and that's not - concrete, - and that's not... You know, that's direct spend. - And so that's - crazy. - That's a crazy amount of - of stuff that's that's going down, - that could go down
- if there's if there's a a financial - fallout - in the economy. - Well, now now, you know... - So now I ask you the question. When's - it gonna crash? - I'm Nobody nobody knows. None of the none - of the pros know. None of the analysts - know. They're all saying - between this weekend and 2028. - That's, like, that's the best prediction - I I see it. - Because right now, it's this house of cards, - and they're putting a lot of money in
- to keep the house from sinking. So, you - know, they they... It's a matter of how - long they can keep the game going before - they just can't keep the game. Yeah. I - mean, we're watching we're watching - billionaires - who - who are - pledging their their their entire fortune - to keep this game going. - And you have to Couldn't happen. - It couldn't happen to a nice aesthetic. - Yeah. I mean,
- it's just... - It just can't go forever. I mean, I - saw... I'm like, oh, this company is seeking - another 7,000,000,000. This company is seeking another 10,000,000,000. - I know. It's like, where The numbers are - the numbers are mind boggling. Yeah. Yeah. Who - has the money? Who is going to invest - this kind of money? And how are you - not draining investment - in every other sector of the economy when - you do this? - And, like, really, is the return on investment - for AI - that much better - than the return on investment for, like, building
- rural broadband - or building - SATCOM networks that can do broadband - or, like, something that... Even car car plants. - Like, is it... Like, these things produce physical - stuff that people can use. - And here we have this ephemeral thing. - I can go ask the AI questions. - Okay. - Really? That's more of a return on value - than... - Like, I Right. I... Our return on investment, - I don't I don't... I'm not seeing Well, - right now, the return on investment is so - gigantically negative. They have a long way to
- go. They even get to zero. - Yeah. So, yeah, building building building anything else - that people will actually buy is a way - better investment. So... Yeah. Yeah. So Well, these - guys... Let... Let's be honest. These guys - are benefiting from the gigantic - stock - price increases - of the of the big companies. - And, again, I think the smart ones have - hedged their bet. They're not gonna get killed - when it crashes. They're already ready for that. - So... Yeah. - Some of them will because not all not
- all rich guys are smart guys. You know? - Yeah. - Well, and not all not all not all - hedges work - in the long run. You know? The big - big fans of mice and Well, if this - is a really ugly crash, none of the - hedges may work. This may be just so - ugly that this blows past the hedges. Right? - Yeah. Exactly. And that's kind of part of - the impossible. Part part of the question or - part of the problem. - I don't know the answer to. Alright. So - some negatives, - some thoughts about, like, you know, I didn't - wanna just go into this and talk about,
- oh, everything's bad, doom and gloom. I was - trying to figure out, like, ways of what - should I as an individual engineer think about, - like, how do I position myself? What do - I think about? - What's what's the likely? - And there's probably gonna be a ten year - drought, like you said, and then - things will pick back up. I I... That's - what I think. I think they'll They absolutely - will. I mean, if I'm a young engineer - today, - and I want it - to really be part of that coming back.
- Because the coming back, even during the ten - years, it's a small level. I think there's - gonna be a ton of work. And the - and the the only way for you to - do that today is you gotta scrape together - money... Enough money to go out and buy - a couple of these cards, - put it in your garage, run it, and - learn it because that's... You... The only way - to learn small AI is to do small - AI because there's nobody to tell you or - teach you. So you have to just do - it yourself. And if you do that, you've - just built a business. I believe that you'll - do fine. I mean, I think you could - go to a 100 other businesses in the - next five years and then put their data
- centers in for them and and make a - really good living at that. So I think - it's a massive growth area. - The other thing we didn't mention, the one - I... Everyone I know still scratches their head - about why these guys gave all their software - away for free. - Yeah. It doesn't make any sense. If this - is such a valuable thing, why are you - publishing the code for the software? Because all - my friends who have these small data centers, - they went and got the software for free. - They didn't even have to pay for it. - It's like, well, thank you. - Yeah.
- And not just open weight models, but actually, - like, the actual source code that you can - The actual source code. Oh, yeah. It's there. - It's all out there, and it's... You know? - So it's it's all setting out on Hugging - Face right now. You can just get whatever - you want out there. I wonder. I wonder. - That's hard because there's a mill... There's a - million different versions now of the AI software - on Hugging Face. - Yeah. I was gonna say... And that that - makes me also wonder, like, this whole panic - over AI doomsters and where it's gonna kill - everybody or whatever - might not be just a
- regulatory - capture - game that we're playing right now. Right? Yeah. - Like, oh, please regulate us. - I... Anytime a company goes to to any - government organization, - any place in the Earth, I don't mean - just The United States. I mean, any place - in the Earth, if If a company goes - to a government and says, - please regulate me, that - to me just means - that just an open admission - that you can't take the competitive
- heat, - and you want the government to come in - and give you a lane to play in - where you can make money. - That's that's just my... Maybe it's... Maybe I'm - wrong on that sometimes. - But immediately when I see a company do - that, that is my initial... - Like, - this is... They are just trying - to build a moat around their business - using government regulation. - That's my immediate response every time I see - that. - I completely agree with you. That's exactly what
- I think is going on. I I think - they're also looking for someone to blame. Well, - well, yeah, we were doing good till they - regulated us. - Yeah. Right. It could be too. Yeah. It - could be too. It... It's so hard to - say. This stuff is, honestly, when you think - about any of this stuff, it makes your - head hurt because this doesn't... This makes no - sense. It's like, you know, this stuff doesn't - make any sense. - Yeah. I don't... Yeah. I I honestly, I - don't know how to - No? Yeah. Anyway... Alright. Cool. Well, I think - we beat that horse to the point where - I think we did. Dismount.
- Yeah. I think so. It's been a good - conversation now. So Yeah. Yeah. It's cool. And - Doug Yeah. And I and I and I - really liked how you didn't predict when it - would crash. So, you know - Very - adroit at changing the subject. - Yeah. You - you did well. You did well. - So, Doug, we need to have you more - on more often. - But, - where can we... Where can people find you? - I've said pots and pans. - Pots and pans by CCG is the actual - name. Pots and pans by CCG. I've written
- I've written a telecom blog for the industry - every day for fourteen years. So - which I added it all up in that - equivalent of writing 40 books or some ridiculous - amount of words. - Nice. So Yeah. Yeah. Cool. So... Yeah. So - I I cover every topic there. Is it - it's either tech, regulatory, or - operational, - all all related to ISP in the telecom - world. So Awesome. Yeah. You should go you - should go read that. I follow - I follow Doug's blog all the time. Alright.
- Tom. - You can find me on LinkedIn. - What? It was more than one word. I - can't I can't hear that. - Well, I'm on LinkedIn as well. But... Yes. - So... But if you got... If you don't... - Yes. But if you go to my blog, - you'll find the LinkedIn. So... Yeah. Cool. Yeah. - I'll try to put a link to the - blog in the show notes here at some - point. Well, I was I was thinking, Doug, - with all of your content that you've produced, - people can just go ask Chad GPT what - you think. Right? - Well, that's funny because I noticed yesterday that
- somebody scraped my - site. So I'll... I had, like, 20,000 - reads yesterday since somebody scraped me. - Oh, my goodness. - So I'm I'm sure I've been into... I'm - sure I know. I met all the models - multiple times. Yeah. Because they're always looking they're - always looking for words. Right? So Yeah. It's - crazy. - Alright. Well, I'm Russ White. You can find - me here at the hedge at rule eleven - dot tech, occasionally on LinkedIn, occasionally on x - route at routing geek. - We know that we live in an attention - driven economy. So thank you for listening all
- the way to the bitter end of this - episode of The Hedge, and we will catch - you next. - Time.
One of the more interesting ways to use AI tools is to use local hardware and open weight models, or even cloud-based closed-weight models, to build skills you can then use a tools to help you accomplish specific tasks....
Brian Trammell joins Alvaro Retana and Russ White to discuss his IETF draft Optional Security Is Not An Option, and why optional security is very difficult to deploy in practice. Brian blogs at http://trammell.ch and...
Social media services have largely become a marketplace designed to capture and hold your attention, rather than places where people share information and talk to one another. A large part of this is because they are...
Who should be held liable when AI fails? Who is liable when an AI system "breaks into" a network or system, causing a major breach? Listen in as Tom and Russ discuss two articles covering these topics on this roundtable...
Network engineers do not need to become full-time coders to succeed--but some coding skills are really useful. In this episode of the Hedge, David Barrosso (you can find David's github repositories here), Phill Simonds,...