Why Hasn't The AI Bubble Popped... Yet?

0:00 Over the last three years, generative AI and large language models specifically

0:04 have simultaneously become the largest investment project

0:06 in the history of mankind and an existential

0:09 shakeup to the way mankind operates.

0:11 Within just the last year,

0:13 the major tech companies have spent the equivalent of 13 Manhattan projects

0:17 or two Apollo moon landings on developing this technology within America alone.

0:21 In the process, some logical questions have been raised

0:25 that nobody really wants to think about too much.

0:27 Like, if AI really does live up

0:29 to the hype and comprehensively replaces all human work,

0:32 who is going to be left to buy anything from these companies?

0:36 Before we even get to that, there

0:37 is the more immediate problem for regular people.

0:40 That if the technology really can do everything it's promised to do,

0:43 then we are going to lose our jobs.

0:46 But if this technology fails to start delivering those results soon,

0:49 it will probably cause a meltdown in financial

0:51 markets that will cause us to lose our jobs.

0:54 The stock market, not to mention the economy itself,

0:57 is almost single-handedly being kept afloat by the hundreds

0:59 of billions of dollars that tech companies

1:01 are piling into data center buildouts with no

1:04 clear indication that these investments will ever pay off.

1:06 It sounds bad, but you probably already knew all of this.

1:10 The thing is though, the people driving this whole industry actually do

1:13 have some good answers to these logical paradoxes.

1:15 The only problem is they may not be what you want to hear.

1:19 Brad, if you want to sell your shares, I'll find you a buyer.

1:23 Meta will invest hundreds of billions of dollars into compute buildout,

1:25 multiple multi- gigawatt clusters in the works.

1:28 As you grow faster, you grow that capex faster.

1:31 I think the investment community is going to demand ROIs on that capex earlier.

1:36 Microsoft shares fell more than 10% in early trade on Thursday.

1:40 Investors are concerned after record AI spending and cloud momentum flows.

1:45 Companies have been trying to cut down on workers

1:48 for as long as those workers have demanded pay and benefits.

1:52 Whether it's downsizing, outsourcing,

1:54 streamlining, understaffing, or automating,

1:55 if there is something a business can do

1:58 to get rid of workers and their salaries,

2:00 you better believe they are going to do it.

2:03 But this time does feel a little bit different.

2:05 Recent AI advances have been mocked for not quite living

2:08 up to the bold claims of their tech bro founders.

2:12 But even in their current imperfect form, LLM's,

2:15 general use robots, and generative models are already replacing jobs.

2:19 And they are getting better every day.

2:21 So that's bad for workers.

2:23 But if you're a senior corporate executive or company owner,

2:26 maybe you should be asking yourself, if we automate everybody's job,

2:30 who is going to buy all of your[ __] I have

2:35 some good news and some bad news for your theoretical company.

2:38 The good news is that labor reduction systems

2:40 of all varieties have already cut out millions

2:43 of man-hour in America alone and made the workers

2:46 who are left more efficient at their jobs.

2:49 Artificial intelligence is just another tool that your company can use

2:52 to get more work out of fewer staff or replace teams entirely.

2:56 Even here at Little Old Works Media Group,

2:58 we used to have someone working part-time whose job it was

3:01 to cut out images on Photoshop to use in our goofy little animations.

3:04 Now, Adobe Suite has inbuilt AI features which can automatically remove

3:08 backgrounds from any image with absolutely no human time or skill involved.

3:12 Now, we love our editors, so we gave him more work editing our history videos,

3:17 but there are a lot of companies that now don't need

3:20 to or don't want to employ people for these simple jobs.

3:23 According to a survey of 697 companies conducted by Metr,

3:27 a research and strategic advisory firm, whatever that means,

3:30 companies who did not use AI in calendar 2023,

3:33 hired 89% more agents than those who did use AI in their contact centers.

3:39 For existing employees, it was even worse.

3:41 When AI was added to a contact center,

3:44 36.8% of companies laid off an average of 26.1% of their employees.

3:49 These alarming numbers were within just one year.

3:52 and just within call centers which have already been hit by outsourcing,

3:57 understaffing, and automation harder than

3:59 most other professional services for now.

4:02 So, congratulations.

4:03 Your business now has reduced headcount and significantly reduced expenses.

4:07 But here comes the bad news.

4:10 In 1914, Henry Ford doubled the salary of his factory workers to $5 per day.

4:14 The story goes he did this because he wanted all of his employees

4:18 to be able to afford a Ford Model T of their own, boosting company revenue.

4:21 Now, this is basically complete nonsense.

4:24 In reality, Ford just wanted to reduce staff turnover and denied talent to upand

4:29 cominging car companies like Dodge to maintain

4:31 his monopoly over the growing automobile industry.

4:33 But as with every good fairy tale, there is a nugget of truth in there.

4:37 If the employment rate suddenly drops as hard as it did

4:40 in the call centers from the study in every company around the world,

4:43 then companies will undeniably struggle with reduced revenue as people

4:46 without jobs don't have the buying power they once did.

4:49 And even people with jobs will cut back out of fear that they might be next.

4:53 But there is a business plan that you as a hypothetical titan

4:56 of industry might want to consider so that this isn't a problem.

5:00 And that's just making stuff exclusively for other wealthy business owners

5:04 while diverting resources away from people who have nothing to offer

5:07 you that can't be done more efficiently by a machine

5:09 or a cheaper worker on the other side of the world.

5:12 Now, if you think that sounds a bit crazy,

5:14 you should know that it's already kind of happening.

5:17 And the best place to see it is in video games.

5:20 Yes, those occasionally fun distractions from real life are the perfect

5:24 example of the shift happening in the real world right now.

5:27 So, let me cook.

5:29 According to Grand View Research,

5:30 the video game industry is now bigger than the movie,

5:34 music, and television industries combined,

5:35 and it has honed in its most profitable strategy.

5:38 Free-to-play games are ironically some of the most profitable games

5:42 on the market because they have

5:44 perfected a business strategy called premium pricing.

5:46 Most people will play their games for free, but a select few,

5:50 affectionately known in the industry as whales, will spend thousands,

5:53 sometimes even millions of dollars on a single game to unlock in-game

5:57 perks that would be almost impossible to obtain as a free user.

6:01 It's important that these games still have free users,

6:04 though, because without them, the wheels would have nobody to show off

6:07 to or pone with their paid for advantages.

6:10 In a world where people are looking for budget friendly forms of entertainment,

6:13 the game industry has realized that the best way

6:16 to turn a profit is to cater to those who have

6:18 the financial means or at the very least the credit

6:20 limit to spend ridiculous amounts of money on a video game.

6:23 The people who have the most financial means don't work for their money.

6:27 They own incomeroucing assets.

6:28 And if automation does replace jobs on an even more widespread scale,

6:32 that will make your ability to work

6:34 less valuable and simultaneously incomeroucing assets more valuable.

6:37 The rate of change in AI and automation makes the future impossible to predict.

6:43 But if the people investing in these technologies

6:45 want to see a financial return on that investment,

6:47 they are going to learn what the game industry already has.

6:50 There is more money to be made by catering to other

6:54 rich people and everybody else can be kept around as entertainment.

6:57 Now, if you think that sounds a bit depressing, well, welcome to this channel.

7:02 But also, I should tell you that market trends already say this is happening.

7:06 AI isn't going to change your world.

7:08 It's just going to continue a trend that's already been happening for years now.

7:12 So, it's time to learn how money works to find out how

7:15 companies are adapting to a world where nobody can afford anything anymore.

7:19 Some of the most valuable assets

7:21 in the modern economy are things you cannot sell.

7:24 Your files, documents, and spreadsheets generate real value, just not for you.

7:29 That's why this video is sponsored by Proton Drive.

7:31 Proton is the team behind Proton Mail

7:33 trusted by over 100 million people worldwide.

7:36 From newsrooms and activists to security experts,

7:39 they are built based in Europe with a privacy first

7:42 mission and strong protection shaped by the European regulatory context.

7:46 Proton Drive is endtoend encrypted cloud storage.

7:48 So your files, photos, and even metadata stay private.

7:52 Nothing is scanned, nothing is profiled,

7:54 and your data is never used to train AI models.

7:57 This matters a lot for things like work documents,

7:59 budgets, and financial spreadsheets.

8:01 Proton Drive also includes Proton Docs and Proton Sheets,

8:04 so you can write, calculate,

8:06 and collaborate with the same encryption guarantees,

8:08 meaning only you and the people you choose can access that content.

8:12 I use Proton Drive for both work files

8:14 and personal projects I do not want exposed.

8:17 Everything stays synced across my devices,

8:19 and I like knowing my cloud storage is private by default.

8:22 And when I need to send something,

8:24 I can share files with password protected or expiring links.

8:26 If you want cloud storage that is private by default and free to get started,

8:30 you can try Proton Drive using the link in the description and pin comment.

8:34 Thanks to Proton Drive for sponsoring this video.

8:38 In the 1992 movie Glengar, Glennon Ross,

8:40 the iconic always be closing scene has Alec Baldwin's

8:43 character bragging to another salesman about driving an $80,000 BMW.

8:47 Adjusted for $192.

8:48 According to the Bureau of Labor Statistics,

8:51 that would be the equivalent of just under $182,000 today.

8:56 That is a very expensive automobile to be sure.

8:59 But these days, if you drive around a major city,

9:02 you are likely to see dozens of cars that cost that much or more,

9:06 and they probably wouldn't even stand out that much.

9:09 According to company sales data and historic records,

9:11 Lamborghini has sold more cars in the last 10 years alone

9:14 than it did in all the rest of the company's history combined.

9:17 because there is simply a bigger market for $300,000 cars than ever before.

9:21 The same general trend is true for other extreme luxury brands like Bentley,

9:26 Ferrari, Paddock Philippe, luxury groceries at Arowan,

9:29 gym memberships that cost more than rent,

9:31 hotels that make Hilton look like a holiday in and to top it off,

9:35 private jets are all now way more popular than they ever have been.

9:39 One thing that often gets lost in discussions of wealth inequality is that yes,

9:44 billionaires are getting richer,

9:46 but there are also far more rich people around than ever before.

9:49 And it's because the value of an hour

9:51 of human labor has stayed more or less stagnant,

9:53 while the value of an asset and the availability of credit has skyrocketed.

9:56 People who held on to an even modest portfolio

9:59 over the last four years now have enormous fortunes.

10:02 Certain individuals got incredibly lucky with good investments,

10:05 and other people accessed easy credit to get absolutely jacked to the tits

10:09 on leverage and ride the longest asset bull run in history.

10:12 Very few people have made modern fortunes through work alone.

10:15 And you might have even noticed how investing

10:16 went from something that you should do to secure

10:19 your retirement to something you need to do to have any quality of life at all.

10:23 If robots really do take over,

10:25 then it will just continue to push the value of an hour

10:28 of human work down because it could be done by a machine instead.

10:31 and it will rally the value of assets because business

10:33 owners and investors will benefit from the free labor substitute.

10:36 This is where people point to policy fixes like a universal basic income.

10:40 If you haven't heard about this idea by now, that's impressive.

10:44 But it's basically a government payment made to everybody regardless of personal

10:48 financial conditions that should be enough to cover basic living expenses.

10:51 It's been proposed as a solution to keep the whole system going if people

10:54 can't earn an income for themselves because

10:56 their labor has been outsourced or automated.

10:59 But that is still not a desirable outcome for anybody.

11:03 If comprehensive automation of human labor gets as far as needing

11:06 this, then you will fall into one of two defined classes of people.

11:09 You will either be getting by based on a government

11:12 lifeline or you will be in the class

11:14 of people that own the businesses complaining about all

11:16 the taxes they need to pay to fund that lifeline.

11:18 I find it a little bit easier to sympathize with one of those groups.

11:21 But there are still no winners.

11:23 Just last week, the results of one

11:25 of the most comprehensive universal basic income experiments was published.

11:29 The research gave 1,000 people living in Texas

11:32 and Illinois $1,000 a month with no conditions attached.

11:35 To qualify for this study, their 2019 household income had to be less than 300%

11:41 of the federal poverty line or $37,470 for an individual.

11:46 There was also an unlucky control group of 2,000

11:48 people who were only given $50 a month.

11:50 This is because when your personal finances are being studied,

11:53 you are more likely to be careful about your spending

11:55 decisions and the researchers wanted to control for these changes.

11:59 The study itself was funded by an organization called Open Research,

12:02 which is not technically affiliated with OpenAI,

12:05 but they are both managed by Sam Alman, who has been very vocal about how

12:08 this technology could displace billions of jobs.

12:10 Now, if you ever find yourself running an AI

12:13 startup that burns millions of dollars a day,

12:15 then one of the best ways to ensure that you keep that investor money coming

12:18 is to pretend to be really scared about

12:20 how your technology is going to replace workers.

12:22 Because, well, replacing workers is really exciting to investors.

12:26 So, realistically, this study was probably little more than a marketing

12:29 stunt conducted by one of the best funded organizations in the space.

12:33 But that doesn't mean that the results aren't still interesting.

12:36 The key results related to employment was that the group that received

12:40 the $1,000 payment ended up earning significantly

12:42 less for themselves than the control group.

12:44 After the payment was accounted for, they still earned more.

12:47 But $12,000 alone is not enough to live

12:49 anywhere in America without other forms of support.

12:52 All households ended up earning more,

12:54 but that was largely because this experiment started right at the beginning

12:58 of CO when employment particularly

13:00 in low-wage professions was hurt by lockdowns.

13:02 The reason the researchers gave as to why households receiving the subsidy

13:06 ended up earning less money for themselves was because many people took

13:09 it as an opportunity to cut down on how much they were

13:12 working and commit more time to leisure and looking after their family.

13:14 This fits the optimistic idea of an AI future where we can have

13:18 machines do everything for us while we

13:20 just relax and collect universal basic income.

13:22 But the findings still told a story of people who were financially struggling.

13:26 Mental health improved because people were less stressed,

13:28 but those improvements faded by the end of the first year.

13:32 Food security also increased, but again only in the first year

13:35 as even people with additional income fell victim

13:37 to rising living expenses and that was

13:39 when they could work to earn additional money.

13:42 When compared to being plunged into poverty, it's an okay compromise.

13:46 But this isn't going to happen overnight.

13:49 Outsourcing turned entire cities in America into ghost towns.

13:51 But it happened so slowly that big changes like this never got addressed.

13:55 The endgame for people investing into automation

13:58 is to build that automated utopian future.

14:00 The endgame for everyone else is to make sure that they aren't excluded from it.

14:05 The thing is though that the very senior executives trying to use AI

14:09 to reduce their staff costs were not

14:11 expecting that they too might be replaceable.

14:15 Senior executives, business owners,

14:17 and management consultants all have a strong incentive

14:19 to cut down costs in their businesses wherever they can.

14:22 It's been less than 3 years since Chat GPT first launched, but since then,

14:26 all of these people have clearly been planning a way

14:29 to replace expensive human intelligence work

14:31 with far cheaper artificial intelligence work.

14:33 You could almost argue that as business

14:35 leaders with a fiduciary duty to their shareholders,

14:38 they would be irresponsible not to pursue this potential cost cutting,

14:41 and the market would probably agree with you.

14:44 Already, several high-profile business leaders and entire teams

14:47 have been fired for not going allin on AI.

14:50 But for everybody else who did, there have been a few problems along the way.

14:54 These managers are realizing that most jobs are more

14:56 nuanced and variable than the reports could truly capture.

14:59 And at the same time,

15:00 AI is turning out to be a lot more expensive than they expected.

15:03 Now, for a lot of businesses, this hasn't really slowed them down.

15:07 Some of you watching have so perfectly put it that you

15:10 are not worried that artificial intelligence can do your job.

15:13 You are worried that some manager will think it can do your job.

15:16 A lot of businesses are now in the find out stage

15:19 of this cost cutting exercise and have

15:20 started begging their workers to come back.

15:22 However, there is something else that has come out of this.

15:25 It turns out that the most expensive and most analytical work

15:28 in a business is done by senior executives and management consultants.

15:32 The boys got really excited about layoffs without stopping

15:35 to think who might be the first to go.

15:38 The thinking jobs are the ones

15:40 that are most at risk from artificial intelligence.

15:43 The AI experiment here for McDonald's doesn't seem to have gone that great.

15:48 It's been revealed Deote used artificial intelligence to prepare a report

15:52 to the federal government's Department of Employment and Workplace Relations.

15:56 Big mistake really to underestimate how quickly these AI models could improve.

16:01 So McKenzie's second centuries is going

16:04 to present some very profound challenges for it.

16:07 Management consulting firms like Bay& Company, McKenzie,

16:09 and BCG bring in billions of dollars in revenue every year from a long list

16:13 of corporate clients that go to them for advice on how to run their company.

16:17 That advice can be on anything from how to expand

16:19 into a new country down to who to hire as their next CEO.

16:23 Since this is something that most companies do not do on a regular basis,

16:26 they hire these firms because they do handle these cases all the time,

16:29 so they can give workable advice based on the experience

16:31 they gained from other companies they have helped in the past.

16:35 That's the theory anyway.

16:36 In reality, one of the biggest problems with management

16:38 consulting firms is that a bulk of the actual

16:40 work gets delegated down to junior analysts who

16:43 are more often than not fresh out of college.

16:46 Since it's typically a bit awkward to have someone who is still learning to put

16:49 on their tie tell an executive with decades

16:51 of experience how to run their business,

16:53 these guys are typically delegated to roles behind the scenes.

16:55 They will collect and collate huge amounts

16:57 of data from the company's own records,

16:59 pair that with information they gather about the broader industry,

17:01 and then pass those insights along to a partner on their team.

17:04 The partner will then fact check their work,

17:06 grill them over all the dumb mistakes they have made,

17:08 and then once that is all done,

17:10 they will use the gathered information to make recommendations to the business.

17:13 The actual end product that gets handed over

17:16 to the company is almost always an extensive

17:18 PowerPoint presentation and report that the most

17:20 senior consultant on the team presents to management.

17:22 Now, I promised myself I wouldn't let this whole

17:24 video devolve into a rant about management consultants,

17:27 but you can probably already start to see

17:29 the problem with this particular business model.

17:31 Collecting, digesting, and regurgitating huge amounts of data

17:34 into something somewhat actionable with a healthy

17:36 amount of fact-checking is exactly what current AI models excel at.

17:40 Good consulting team leaders are going

17:42 to doublech checkck the work of their juniors.

17:44 Anyway, additionally, when they hand back their findings and recommendations

17:47 for them to turn into a nicel lookinging presentation,

17:49 modern AI programs can do that, too.

17:51 A report by Harvard Business Review found that groups within

17:54 these firms were changing from a pyramid structure with lodge of juniors

17:57 at the bottom and then tapering headcount towards senior manager partners

18:01 to today where the corporate structure is representing more of an obelisk.

18:04 Fewer juniors at the bottom doing leg work all assisted by AI programs

18:08 to maintain the rain makers at the top actually bringing in the contracts.

18:12 Of course, since these consulting teams are working with internal

18:15 reports from major companies that are often listed on public markets,

18:18 data security is a massive consideration.

18:21 So they shouldn't just mindlessly upload sensitive files to chat GPT.

18:25 Although they frequently have been caught doing exactly that.

18:28 In just the most beautiful twist of unintentional irony in a headline ever,

18:32 Deote was caught using AI to produce a report delivered to the government

18:36 of Australia on a contract reportedly worth over a quarter of a million dollars.

18:39 They were discovered because they admitted the fact-checking part of this entire

18:43 operation and just ran with the numbers that AI spat out.

18:46 The ironic part was that this quarter of a million dollar report was

18:49 about the most effective ways to crack

18:52 down on overly generous government welfare.

18:54 But anyway, the issue here was not that Deote was using AI at all.

18:58 In a press release about this particular incident,

19:00 they proudly announced that they were actually investing $3 billion into AI.

19:04 The problem was that they hadn't bothered to fact check it at all.

19:08 They submitted their homework with references to non-existent

19:10 academic papers and completely fabricated federal court judgments.

19:14 Either way, these consulting firms are still going full

19:17 steam ahead with AI implementations in their own workplace,

19:19 and they have reasons beyond just cutting down their own staffing costs.

19:24 Revenue for most of the major consulting

19:26 firms have been skyrocketing in recent years.

19:28 Deoid, for example, has doubled the revenue over the last decade,

19:32 and high-end boutiques like BCG have tripled theirs.

19:34 A major driver for this revenue growth has been consulting

19:37 other companies on how to implement AI into their own workflows.

19:41 By going all in on AI, they can better present their firm as subject matter

19:45 experts to this new group of highly lucrative clients.

19:47 Even if the results have so far been questionable.

19:50 However, while this has been a boon to business in the short term,

19:55 it could end up being a major risk to the industry in its entirety.

19:58 As part of a study, which we will actually dive into soon,

20:01 a group of researchers from Cambridge noted that most of the hard

20:04 work of these consulting firms can be done by data digesting AI systems.

20:08 So, why would companies pay firms like McKenzie hundreds of thousands

20:11 of dollars when they could get

20:13 the same vaguely reliable recommendations internally?

20:15 It's no secret anymore that more often than not,

20:18 when hiring outside consulting firms,

20:20 a lot of company managers have already decided what they wanted to do already,

20:24 and they just want a firm like McKenzie to officially agree with them.

20:27 Depending on the industry, it could be just as impressive to a clueless

20:30 board to say that key business decisions were driven

20:33 by cuttingedge generative models as it would be to say

20:36 our decisions were thoroughly vetted by top industry consultants.

20:39 Translated from corpo, they both mean the same thing.

20:42 Anyway, there is however one group of jobs that have shown great promise

20:46 when it comes to let's use corporate consulting speak to say refining headcount.

20:50 The only problem is they are the same

20:53 people that hire the consultants in the first place.

20:55 So there is a general assumption that when

20:57 and if new technologies come for our jobs,

21:00 the process of elimination will start from the bottom and that the people

21:03 running the companies will just end up lording over a big army of robots.

21:06 However, early research has found that there

21:08 might be three big advantages from starting at the top first and equally three

21:13 challenges that stop it from actually happening.

21:15 First and foremost, senior executives just cost a lot more than average workers.

21:19 You have probably heard that stat before,

21:22 but according to the Economic Policy Institute,

21:24 the average CEO at the top 350 firms in America

21:27 now makes 280 times more than their average employee.

21:30 And that number is actually down from the big

21:33 spikes after 1999, 2008, and CO 19.

21:35 At the same time, a report by MIT found that over

21:39 95% of AI pilot studies within measured businesses were failing,

21:43 especially amongst businesses that had developed their own bespoke tools.

21:47 Building an AI system, hiring a team to manage that system,

21:50 and investing in the accompanying machinery required

21:52 to do entry-level jobs is a big

21:54 task when a human could do the same thing for a modest salary.

21:58 But if the role of a senior executive could be replaced,

22:00 that's a massive cost-saving that could easily pay the salaries of a full team

22:04 of engineers there to babysit the system

22:06 and make sure it doesn't go off the rails.

22:08 Now, if you do what the attention-grabbing headlines

22:10 didn't want to do and actually read this report,

22:13 they go on to say that by far the best implementation of AI is

22:17 in small tools that augment human work rather than try to replace it entirely.

22:20 Now, this is not a great sign for the swelling investments

22:23 made by AI companies into radically restructuring the nature of work itself,

22:27 but it's also not a great sign for people at the top of the corporate ladder.

22:31 Another study conducted by professors at Cambridge University

22:34 in collaboration with industry experts and published in the Harvard

22:37 Business Review ran a test pitting 344 business students

22:40 and business executives up against an off-the-shelf AI model.

22:44 They simulated a company in the auto

22:45 industry and instructed the participants to make

22:47 key business decisions indicative of those made by real CEOs in real industries.

22:53 Additionally, just like real CEOs,

22:54 their work was also judged on two key metrics.

22:57 The first was the overall performance

22:59 of the business based on variables like market share,

23:02 profit, and market capitalization.

23:03 The second was not getting fired by the board of directors.

23:07 Overall, the researchers found that the LLM consistently

23:09 outperformed the top human participants in the trial,

23:12 including the actual executives, on almost every business metric,

23:15 and their conclusion as to why kind of makes sense.

23:19 Business leaders should be able to make cold, calculating,

23:22 and often unpopular decisions about how to run their company without letting

23:25 human emotions like greed or fear sway those choices once they are made.

23:29 Of course, LLMs are prone to hallucinations or in plain English,

23:33 just confidently making[ __] up.

23:35 But the researchers actually addressed this and found

23:37 that even when it did happen,

23:39 the benefits outweighed the risks in these broad roles,

23:42 especially since human CEOs were also prone to confidently

23:44 make up from time to time as well.

23:47 Now, this all did come with a big caveat that we will get to soon.

23:51 But the final advantage of a more

23:52 mechanized management hierarchy is simply availability.

23:54 Getting a meeting with a CEO or other senior leaders in a large company can

23:59 often take a very long time because there

24:01 are a lot of functions that they oversee.

24:02 And depending on how they run the company,

24:04 big decisions often need to be run by them before being implemented,

24:08 creating an operational bottleneck.

24:10 Theoretically, a well-run piece of software could

24:12 be available to approve or deny requests 24/7, eliminating that bottleneck.

24:16 or more realistically increasing the bandwidth of a smaller management team.

24:20 Now before you get too much shout in Freuda from watching

24:23 the layoff happy management team score an AI own goal,

24:26 there are a few problems but also an incredibly unlikely advocate.

24:30 The first and most immediate limitation is that to comply with Sarbain

24:33 Oxley public companies are legally required

24:35 to have listed a principal executive officer,

24:38 a principal financial officer and a principal accounting officer.

24:41 Normally these people are just given the title CEO,

24:44 CFO and controller and the last two

24:46 roles are normally managed by a single person.

24:49 These roles do need to be filled by actual

24:51 human beings though and this is largely so

24:54 that the SEC has someone to hold accountable

24:55 if the company operates in a willfully illegal manner.

24:58 Even the Cambridge study simulating the role of a CEO

25:01 said this was going to be one of the biggest challenges.

25:04 You can't send an AI to prison.

25:06 And unless they really are sentient beings with a conscious,

25:08 they probably don't really care about being turned off either.

25:11 However, that doesn't mean that this idea is completely dead in the water.

25:15 There are still plenty of other leadership roles

25:18 that can be replaced outside of these official designations.

25:20 And a more pragmatic solution is just to make smaller teams of execs

25:23 to do the work of many with the help of these programs.

25:26 A study conducted by the Harvard School of Business on more than 50,000

25:30 participants found that this is basically

25:32 what was happening in the included teams.

25:34 management structures were flattened and the leadership

25:36 that was left could simultaneously fill multiple roles while being more

25:40 available to their clients and subordinates.

25:42 Sounds great, right?

25:43 Well, there are three problems.

25:44 The first is that this study was only done on people in tech roles who

25:49 are naturally going to be more adept

25:51 to using these programs and understanding their limitations.

25:53 The second problem is that companies are likely

25:55 to see research like this, not actually understand it,

25:58 and use it as an excuse to give a person previously working one job the workload

26:02 of three people and make it all better by tossing them a chat GPT subscription.

26:07 Beyond that, the bigger problem for average

26:08 people is that if this trend continues,

26:10 it's likely to be rolled out on middle managers first,

26:13 which could seriously hamper regular career progression.

26:15 If someone above you leaves the company, the business could hire you.

26:19 or they could give an existing manager a small

26:21 pay bump to absorb that role into their own.

26:24 Now to be clear, businesses absolutely already do this all the time.

26:28 The only difference is that now they can say they are doing it

26:31 under the guise of promoting a flat

26:33 corporate structure by implementing emerging technologies.

26:36 Now, as for the CEOs themselves,

26:37 the second hurdle was actually outlined in the same

26:40 study that spoke about how effective an AI alternative was.

26:43 It turns out that over the course of the experiment,

26:46 AI excelled at maximizing business performance,

26:48 beating out all the other participants,

26:50 but it really sucked at not getting fired by the simulated board of directors.

26:54 Now, that's not because they were anti-clanker.

26:57 It's because the AI was really good

26:59 at hyperoptimizing variables during regular business operations,

27:01 but it was really bad at accounting for black

27:05 swan events like a simulated recreation of CO 19 disruptions.

27:08 The solution put forward by the researchers was pretty obvious.

27:11 AI systems can almost act like a modern

27:14 autopilot for a business instead of a plane.

27:16 It can handle regular operations during normal times

27:18 and monitor far more data than any single human could.

27:21 However, in times of crisis, it's still good to have a real human at the helm.

27:26 This seemed obvious to the corporate executives addressing the study,

27:29 but that may have been because it was their own jobs on the line.

27:33 The exact same thing could be said about almost any other

27:36 modern job being augmented by AI or any technology for that matter.

27:40 But for some reason,

27:41 the decision makers aren't as receptive to the same conclusion.

27:45 Now, obviously, I am being facicious because of course

27:47 the biggest hurdle that's really going to stop

27:50 AI from replacing senior executives is that they

27:52 are the ones who decide who gets replaced.

27:54 And it may not surprise you to learn that they

27:57 statistically consider their own jobs far harder to replicate.

27:59 In a survey conducted by IBM of all things and Oxford

28:03 Economics surveyed 3,000 global seauite executives

28:05 across 20 industries and 28 countries,

28:08 77% of respondents said entry-level positions were already seeing the effects

28:12 of generative AI and that will intensify in the next few years.

28:16 However, only 22% of them said

28:18 the same for executive or senior management roles.

28:21 Now the disconnect in all of this is

28:23 that when they were questioned separately about specific business functions,

28:26 97% said they think employees in procurement and 93% of those in risk

28:31 compliance and finance were already being replaced

28:33 or augmented by AI in coming years.

28:36 But only 77% of customer service

28:38 roles were likely to be significantly disrupted.

28:40 Even if AI doesn't end up replacing everyone,

28:43 we could still lose our jobs anyway simply because

28:46 so many companies are jumping the gun on AI disruption.

28:49 If AI fails, the entire economy collapses with it and we lose our jobs.

28:54 But if AI succeeds, it will only be because it has taken all of our jobs.

28:59 Whether we like it or not, our immediate economic future is riding

29:02 on the success or failure of this technology.

29:04 Just the top companies are on track to spend

29:07 $400 billion on data center buildouts this year alone.

29:10 And the entire stock market has clearly been buoied by investor hype

29:14 around what this technology could do at some undetermined point in the future.

29:19 Now, that has done two things in the short term.

29:22 Trillions of dollars worth of data centers may or may

29:24 not turn out to be a spectacular waste of money, but at least for now,

29:29 they are providing tens of thousands of high-paying jobs to engineers,

29:32 tradesmen, and technicians all across the country.

29:35 The same goes for the stock market.

29:37 These impressive gains may or may not be backed by sustainable fundamentals.

29:41 But at least for now,

29:43 people who have seen their portfolios double in value over the last 3

29:46 years are feeling pretty good about treating

29:48 themselves to a bit of conspicuous consumption,

29:50 helping to keep the rest of the economy afloat.

29:53 Now, I know the problem of our entire economy being kept alive by unsustainable

29:57 capital expenditure and questionable financial gains is

30:00 not exactly shocking news to anybody anymore.

30:02 In fact, if you're watching this video,

30:04 there's a good chance your entire subscription feed

30:07 hasn't shut up about the AI bubble all year.

30:09 And in our defense, if this bubble pops,

30:12 it will probably take the rest of the already shaky economy down with it.

30:16 This also makes the fact that people are treating

30:18 it like a foregone conclusion all the more dangerous.

30:20 But perhaps we have all been missing the more important question,

30:24 which is what does a good outcome actually look like at this point?

30:28 Is there any way that we could still quietly back out of this unscathed?

30:32 Google spending tens of billions of dollars

30:35 to build three massive AI data centers.

30:37 Economists are saying electricity prices are climbing

30:40 more than twice as fast as inflation is.

30:42 Some projections show that spending may be fueling

30:45 nearly half of this year's estimated GDP growth.

30:48 Tech companies just are cutting tens of thousands of jobs

30:50 and some of them some of them are blaming AI.

30:53 Now they are pivoting capital away from labor and towards AI infrastructure.

30:58 Okay.

30:59 So to really hedge our bets,

31:00 we need to look at the three ways this technology could pan out

31:03 and what each of them would do to the economy that regular people live in.

31:08 The uh optimistic outcome where

31:10 artificial intelligence becomes exponentially more

31:12 capable and comprehensively substitutes human

31:15 workers in almost every profession.

31:17 The pessimistic outcome where technology plateaus

31:19 and it becomes clear that trillions

31:21 of dollars worth of data centers were a huge waste of resources.

31:24 and perhaps the scariest outcome of all,

31:27 which is a scenario where this technology finds some useful applications,

31:30 but overall just continues to be a bit meh.

31:34 So, the optimistic outcome is what most of the big

31:37 AI companies and personalities are pitching to the world.

31:39 That is with the exception of Peter Teal.

31:42 I am not really sure what he wants.

31:44 You would prefer the human race to endure, right?

31:48 Uh, you're hesitating.

31:49 model.

31:49 Anyway, the idea of this scenario is that AI models

31:52 and the tools they enable get so good that it renders most human

31:56 work obsolete and whatever human work is left becomes thousands of times

32:00 more productive thanks to tools that do the non-creative work for us.

32:04 Skilled trades are done by highly capable humanoid robots.

32:07 Logistics is fully automated and even health

32:10 and elderly services leverage mechanical muscles and mechanical minds

32:13 to care for an aging population with a smaller

32:16 group of actual doctors watching over everything.

32:19 Eventually, the only jobs left will be those that need human creativity,

32:22 assuming we can't automate that, too.

32:24 If you have always wanted to design your own video game,

32:27 you won't need a full development team.

32:29 You could just vibe code your way through production with AI tools,

32:32 creating something that used to take thousands of real manh hours.

32:34 Yes, I know a lot of you actual developers

32:37 or skilled artisans are rolling your eyes right now.

32:40 But remember, this is the dream scenario the AI companies are selling.

32:43 And to play devil's advocate, it's kind of already happening.

32:46 Even if you completely ignore the current focus on generative AI and robotics,

32:50 there are hundreds of technologies that have already done the same thing.

32:54 To use the example of making your own game again,

32:56 there are modern development platforms and tools that have made it far

32:59 easier for a single person to put together a pretty good game,

33:03 even working out of their bedroom.

33:04 The same thing has happened across almost every industry in our economy.

33:08 It's the reason why our labor productivity

33:10 gets better and better almost every year,

33:12 meaning for every hour that we work, we are producing more value.

33:16 The optimistic view of new AI technology is that it will accelerate

33:19 this trend so much that just a few hours of human labor,

33:22 mostly to oversee the work of the machines, will produce enough value to give us

33:26 a lifestyle we couldn't even dream of today.

33:28 So yeah, even if we lose our 9 to 5 jobs,

33:31 it won't really matter because there will be

33:33 so much stuff getting made that only a minuscule

33:35 amount of hours will need to be traded

33:37 in return for everything you could ever need.

33:39 In fact, for a lot of people, it could be so minimal that it's not really

33:43 worth having any kind of formal employment at all.

33:45 Everybody from Sam Alman to Jerome Powell has

33:48 said the same thing about AI improving productivity,

33:50 although admittedly with differing degrees of optimism.

33:53 And the thing is, the idea itself isn't inherently wrong.

33:57 But there is a small problem.

34:00 Worker productivity has already increased by a lot for a number of reasons,

34:04 including technology.

34:04 But for at least 50 years, almost none of that increase has

34:08 resulted in higher compensation for regular workers.

34:10 So then why would this time be any different?

34:13 Well, the big AI players actually do have some answers to this obvious question,

34:18 but these talking points need to be considered alongside some logical

34:22 ironies that would almost be funny if they weren't so important.

34:26 So, the widening productivity compensation gap is

34:28 one of the most hotly debated issues

34:30 in all of economics because everybody has something

34:33 different they would like to blame it on.

34:35 Some people will say that migration has flooded the labor market

34:38 with people who are willing to accept lower wages for the same job.

34:41 Some people point to women entering the workforce,

34:43 which also increased the total supply of workers.

34:45 Some people point to outsourcing high labor

34:47 industries like manufacturing or technologies like computer

34:50 systems that made the same amount of clerical

34:52 work possible with a far smaller team.

34:54 Then of course other people will say it's because

34:56 labor unions were crippled in the early 1980s which

34:58 made it harder for workers to negotiate collectively

35:01 and turned the jobs market into every man for themselves.

35:03 Now the reality is that it was probably a little

35:06 bit of everything because they all effectively did the same thing.

35:09 Businesses got access to a greater supply of labor than they

35:13 demanded and without any negotiating power that pushed labor prices down.

35:17 AI is really just another technology

35:18 that will mean businesses demand fewer employees

35:21 to perform the same operations and something

35:23 that reduces negotiating power as well.

35:25 A report published by the executive outplacement firm Challenger Gray

35:29 in Christmas tracks layoff announcements and categorizes them by type.

35:32 This year outside of federal worker layoffs from Doge,

35:36 the number one reason cited for layoffs was a cost cutting shift towards AI.

35:40 In total, since the firm started tracking this category of layoffs in 2023,

35:44 more than 150,000 job losses have been directly attributed to this technology.

35:48 Now, I know what you're about to say.

35:51 Yes, a lot of companies are just claiming

35:53 that they are doing layoffs because of AI

35:55 when really they just want an investorfriendly

35:57 way of saying that they are cutting costs.

35:59 But that's actually exactly the point.

36:01 Even without the ultra optimistic tools

36:03 that could be made possible by this technology,

36:06 it's already becoming a great negotiating tool against workers

36:09 pushing the gap further apart rather than closing it.

36:11 So this is where we get onto the narrative of living in a world

36:14 of such material abundance that the idea of money becomes completely irrelevant.

36:19 It's a nice idea, but there are a few problems.

36:22 As I was putting this video together,

36:24 a group of researchers from OpenAI very publicly resigned because they said

36:28 that the company's economic studies were

36:30 slowly being pushed into outright AI advocacy.

36:32 If they found something that would present a future of AI in a negative light,

36:36 they would just bury it and broaden the scope

36:39 of their research until the results looked good.

36:41 One of them was quoted saying,

36:43 "The economic research team was veering away from doing real

36:46 research and instead acting like its employer's propaganda arm." Now,

36:49 this is just the word of one man,

36:51 but he probably gave up a pretty big payday to get the message out,

36:54 and it does align with other reports coming out of the organization.

36:58 The technically separate organization, Open Research,

37:01 conducted one of the biggest universal

37:03 basic income experiments ever on lowincome households.

37:05 The results were mixed at best,

37:07 but the headline press releases were overly positive about how

37:11 great an AI powered UBI future will be for everybody.

37:14 Now, maybe it shouldn't be shocking that research coming out of a company

37:17 with a very clear motivation to push mass

37:19 AI adoption would be a little bit biased,

37:22 but the ironies do go deeper than that.

37:24 Remember when OpenAI was a nonprofit and the founders

37:27 didn't care about making money off it?

37:29 Well, according to reports from Reuters,

37:30 it is now gearing up to go public in a move

37:33 that could net the CEO as much as $10 billion.

37:36 Now, if only he still had any economists left on his team,

37:39 they could tell him that all of that money

37:41 will be meaningless in the coming AI utopia.

37:43 All right, joking aside though,

37:45 it's pretty clear that even if this technology meets

37:47 the expectations of even the most ambitious techno optimists,

37:50 there are very few systems in place to ensure

37:53 that the increased productivity from AI actually benefits regular people.

37:56 It's a lot of technoeconomic jargon to try

37:59 and gloss over what you probably already know,

38:01 which is that if the people that own these tools can

38:04 make a lot of money off not needing to pay workers anymore,

38:07 they are going to try their very best to do it.

38:10 But that was the optimistic outcome.

38:12 The most immediate fear that a lot of people have right now is that the hype

38:15 around this technology will die off and bring down the entire economy with it.

38:19 Again, as I was putting this video together,

38:21 the jobs report for November was published after

38:23 October was skipped due to the government shutdown.

38:25 Overall, unemployment rose to 4.6% with an additional 4.1% of people

38:30 in part-time or gig work who would prefer to be working full-time.

38:34 This is the highest it has been since 2021, and it's clearly trending upwards.

38:38 This is also before revisions which have consistently

38:41 been downwards for every report so far this year.

38:43 It's not great, but the numbers have

38:45 been helped a lot by three standout categories.

38:48 The first two were healthcare and social services

38:51 primarily to look after an aging and ailing population.

38:54 The third category was construction,

38:56 specifically specialized non-residential construction or in plain

38:59 English all of those data centers.

39:01 This was something that the BLS explicitly highlighted

39:04 in its March report and has only become more

39:06 relevant as the rate of data center buildouts

39:09 accelerates and other government infrastructure projects are scaled back.

39:11 Now eventually this will be a problem because

39:14 theoretically we are going to slow down on building

39:16 these data centers at some point before we

39:18 turn our entire solar system into a matrioska brain.

39:21 So there's still like a lot of things like I would love to go

39:25 build the Dyson sphere around the solar

39:26 system and like you know make the world's

39:28 gigantic data center with the entire energy output of the sun but obviously we

39:32 can't do that right now so I have to like wait a couple decades.

39:35 Okay, my mistake.

39:37 Well, at least for the next few decades I guess.

39:41 Anyway, the total employment up and down the supply

39:43 chain for these buildouts is very hard to precisely calculate,

39:46 but it is in the hundreds of thousands according to most industry estimates.

39:50 A lot of this construction is also taking place in remote rural areas where

39:54 these companies have been able to lobby

39:56 for favorable local tax and regulatory treatments.

39:58 Now, that whole game is a separate problem that I know

40:01 that the old team over at Micro is actually making a video on.

40:04 So, I am not going to steal their thunder

40:06 on exactly how messed up this can all get.

40:08 But the point is, as it relates to employment,

40:10 a lot of on-site workers doing this construction

40:13 are only in these locations while buildouts are happening.

40:15 That's very stimulating to the local economy for the two

40:19 years it takes to get these facilities online.

40:21 But if it slows down nationwide, it could have broader employment impacts beyond

40:24 just the workers directly in the supply chain.

40:27 Now, in the short term,

40:28 even if this technology never becomes commercially viable,

40:30 it could be written off as tech companies using the piles of cash

40:34 they have accumulated to fund their own jobs program for a couple of years.

40:37 It would be like if the wealthy old man from Up the Road paid

40:40 you and your friends a million dollars to dig a giant hole in his backyard.

40:43 Would there have been a better way to use those resources?

40:46 probably, but it wouldn't cause any other ongoing problems, right?

40:49 Well, it might actually, and we will get to that.

40:51 But right now, the concern is about the existing problems it is covering up.

40:55 $400 billion in capital expenditure this year

40:58 alone is a pretty big economic stimulus.

41:00 Nationwide, according to data from Bloomberg and Renaissance Macro,

41:04 company spending on AI is now contributing

41:06 more to GDP growth than consumer spending.

41:09 That money is ultimately flowing from the big bank accounts of big

41:11 tech firms and investment funds into the hands of thousands of businesses,

41:15 suppliers, and workers.

41:16 In fact, if we just look at the numbers alone,

41:19 then in terms of money being pumped into an otherwise unhealthy economy,

41:21 this is almost exactly the same as the combined stimulus and troubled asset

41:26 relief program spending from 2008 to 2009

41:28 to stabilize the economy after the GFC.

41:31 Now, of course, these payments are not as direct.

41:33 And for now, the good news is that most of this money is still coming

41:36 from cash reserves these companies already had

41:38 with borrowing mostly being used for cash flow purposes.

41:41 But it should still give you a sense of scale as to how

41:45 much economic activity is being generated

41:47 on the contingency that AI hype continues.

41:49 If that gets taken away very suddenly,

41:52 it could quickly reveal how shaky everything else was under the surface.

41:55 Then of course, there is the wealth effect

41:58 of people watching the green line go up.

42:00 A lot of retirees and independently wealthy households have seen

42:03 their investment portfolios almost double over just the last 5 years,

42:06 which means that they have been a lot happier to spend conspicuously.

42:10 This spending may be very uneven, but in theory at least,

42:13 someone renovating their vacation home or buying a fifth sports car is

42:16 at least giving money to contractors or a commission to a car salesman.

42:19 A lot of people living off their investments actually have a set

42:22 draw down rate of about 3 to 4% for their everyday spending.

42:25 You might have heard about the 4% rule for early retirement.

42:28 And this rule prescribes that you don't spend more than 4% of the value

42:31 of your invested assets every year so

42:33 that you don't eat into your principal wealth.

42:35 Now, if your portfolio doubles in value over 5 years,

42:38 that means you can do twice as much spending without changing your 4% rule.

42:42 Now, a lot of wealthier households are not this mathematically rigid.

42:45 But the wealth effect happens subconsciously as well.

42:48 My good friend Ben Felix has done a series of great videos on this subject,

42:51 as well as a video he made last week

42:54 analyzing the current risks in our very concentrated investment market.

42:57 So, if you're interested in a detailed breakdown of these ideas,

42:59 I will leave a link to them below.

43:01 If people see the value of their homes double,

43:04 even though it's difficult to actually access those funds,

43:06 they are still going to be more likely

43:08 to get themselves to do something like a big remodel.

43:10 Now, so far, this has actually helped to keep things ticking along.

43:13 But it's also no secret that a lot

43:16 of these companies might be a teensy bit overvalued.

43:18 The only market worth disrupting after burning

43:21 this much capital is the labor market.

43:22 Even a small slice out of the $12 trillion

43:25 worth of annual payroll could recoup these costs pretty quickly.

43:28 But without that, investors are only going to have so much patience.

43:32 Even then, it's not exactly this straightforward.

43:34 One of the biggest ironies in this push to automate jobs is that the tech

43:38 bros are trying to automate jobs that barely make any money in the first place.

43:43 Uber drivers subsidize the wear, tear, and depreciation of their cars,

43:46 and the taxpayer frequently subsidizes warehouse workers with food stamps.

43:50 The financial reality behind taking these kinds

43:52 of jobs with expensive technology is questionable at best.

43:55 And if too many people start asking those questions, that could be a problem.

44:00 If these values are reconsidered in a major way,

44:03 then both infrastructure spending and investment

44:06 fueled household spending would fall simultaneously,

44:08 seriously hurting employment.

44:10 We have always wanted Green Line to go up,

44:12 but never before has so much real employment depended on it.

44:15 If this technology does turn out to be everything

44:18 investors are hoping for, it's going to take our jobs.

44:20 And if it becomes clear it can't do that, it will very likely crash

44:24 a very frothy and concentrated market and we are going to lose our jobs.

44:28 It sounds pretty bad, right?

44:30 Well, there is actually a third option.

44:32 It's almost become the assumption that this will all end spectacularly,

44:35 but there isn't actually any guarantee of that.

44:38 For now, the companies involved have more than

44:40 enough cash between them to cover outside debt.

44:42 And even if less liquid players like OpenAI can't

44:45 meet the record- setting spending obligations with companies like Oracle,

44:48 the recourse to actually collect on those payments isn't very strong.

44:51 Nobody wants to sue the next person along from them in the financial circle.

44:55 I'm so happy that my lawyers have something to do.

45:00 I'm so happy that I get to sue how

45:03 money works and show them how the legal system works.

45:06 Well, almost nobody, I suppose.

45:08 Anyway, to keep their own market stable, the most likely thing to happen is

45:12 that these companies just take equity in exchange for compute

45:15 time or quietly renegotiate commitments behind the scenes

45:18 and just stretch them out over longer periods of time.

45:21 So, even if advancements continue to slow down,

45:23 this could plot along for a lot longer than you might expect.

45:26 But in the long term,

45:28 this kind of scenario could actually be an equally damaging outcome.

45:31 Consistent investment into AI above anything else is going to starve

45:34 other hopeful technologies of getting the funding they need for development.

45:37 The joke about putting AI in the name of the company to get investment is true,

45:41 but a lot of companies do it because

45:43 they need that investment to develop their product.

45:45 An anecdotal example I saw while putting

45:47 this video together was the startup Boom Supersonic.

45:49 It has been raising money for the past

45:51 decade to develop a modern supersonic passenger jet aircraft.

45:54 Now, maybe you think faster air travel is a worthy technology to pursue.

45:59 Maybe you don't.

45:59 But that doesn't really matter.

46:01 The point is that this month they came out

46:03 with a new product which was basically taking one of their jet

46:06 engines and attaching it to a generator which they were

46:09 pitching as a new and innovative way to power data centers.

46:12 Now, this of course is just one example,

46:14 but startups trying to develop new technologies are all being forced

46:17 to play some version of the same game to keep funding going.

46:20 On a larger scale, there has been a big push towards reshoring manufacturing.

46:23 This may very well be a noble goal with strategic value,

46:27 but it's also made much harder by rising energy costs and electrical

46:30 grids that need to support factories as well as data centers.

46:33 Right now, the data centers can just afford to pay more,

46:36 which means factories and regular households

46:38 have to suffer the financial consequences.

46:40 The same kind of financial crowding out

46:42 is happening in a lot of markets already.

46:44 The reason you are going to need to sell a kidney to afford 2 gigs

46:48 of RAM is because the entire production base

46:50 is being redirected towards serving this one industry.

46:52 Going allin on one hand may or may not pay off,

46:56 but it also means we don't have any chips left

46:58 to take better bets if and when they come along.

47:01 And the thing is, we've kind of already bet

47:04 our entire economy on this one technology working out.

47:06 The value of the entire stock market has nearly doubled over the last 5 years.

47:11 This has been driven almost entirely by just 10 companies whose

47:15 growth was in turn driven almost entirely by the promise of AI.

47:19 Most people understand that this isn't sustainable,

47:21 but they also understand that technology like this has changed the world before.

47:26 Improvements are slowing down as new

47:28 models are showing disappointing improvements.

47:29 But every day, we are discovering new things we can do with these products.

47:34 For every argument, there is a counterargument.

47:36 You don't need to watch another video speculating about the future

47:39 of AI because the truth is nobody has a damn clue.

47:43 But what we can do is look back over these last 3 years and find

47:47 out how much this whole game has cost us because the numbers are not good.

47:51 It's almost like investors are saying 100 million isn't enough to make it too.

47:56 And we're told that Zuckerberg at that time really said okay I'm

47:59 thrilled we get to do this in the United States of America.

48:01 I think this will be the most important project of this era.

48:04 The dormant nuclear reactors at 3M island because decades after

48:08 a partial reactor meltdown shut down part of the facility.

48:12 Okay.

48:12 So, everybody has been paying a lot of attention

48:15 to the astronomical valuations that have been assigned to everything,

48:18 even partially AI related.

48:19 And they have also been paying attention to billions of dollars

48:22 in value vanishing into thin air whenever Jensen Huang sneezes the wrong way.

48:26 Now, to be fair, we kind of have to pay attention to it

48:29 because that's all of our retirement savings tied up in this fun little game.

48:33 But all of this value probably means a lot less than you think.

48:38 Market capitalization, which is what most AI headlines focus

48:40 on, is simply a calculation of how many shares

48:42 are in circulation multiplied by what number

48:44 the last speculator in line made a trade at.

48:47 If Nvidia has in value, that doesn't mean that $2 trillion has been destroyed.

48:51 It just means we have put a new price tag on what this business does.

48:56 Just like if Walmart marks down a banana from $10 to $5, it hasn't destroyed $5.

49:01 The actual numbers we should be paying attention to are the actual resources

49:05 that we have put into this system versus what we have gotten back out.

49:09 And by resources, you can't just think of cash.

49:12 Cash is really just the receipt for resources.

49:14 So, while dollars are a great unit to tally everything up,

49:18 what we should instead be keeping track of is how many cold,

49:21 hard, tangible Age of Empire style resources we

49:24 have used to fuel this one particular tech tree.

49:26 Because only then can we find out if

49:28 this is really making the world a richer place.

49:31 All right.

49:32 So, how do we do that?

49:33 Well, we can start from a simple analysis of the direct business

49:36 expenses taken on by companies desperate to stay ahead of their competitors.

49:40 And after that, we can work down to the less tangible stuff,

49:43 like the opportunity cost of sanctioning China over AI chips or dedicating

49:47 some of the world's best researchers to producing whatever the hell this is.

49:51 The first major major resource being used to fuel

49:55 this machine is the most obvious one, all of those chips.

49:58 In our Amazon video released just a month ago,

50:01 the tally for AI capital expenditure amongst

50:03 just four of the top tech companies had

50:06 reached $200 billion for the year and $60

50:08 billion for the most recent reported quarter alone.

50:11 But since that video went live, new filings have come out.

50:14 And now, according to company reports collated by Bloomberg,

50:17 spending for this year is already on track to hit $344 billion.

50:22 That is more than 1% of America's entire GDP being

50:27 used building compute infrastructure to develop and operate new models.

50:30 Now, if this sounds just a little bit dumb to you, don't worry.

50:34 It's actually worse than you think.

50:36 For one thing, this number only includes the top four tracked public firms,

50:40 Meta, Google, Microsoft, and the biggest spender of all, Amazon.

50:44 Smaller businesses or private companies like OpenAI are not included

50:47 in these numbers because they don't have to release financial information.

50:51 But from the outside, we can still make some educated guesses.

50:54 The scale and frequency of fundraising taking place at OpenAI,

50:57 the makers of Chat GPT, is growing exponentially every year,

51:01 even as they have started squeezing out revenue.

51:03 In March this year, the company received $40 billion

51:07 from investors in the largest private tech deal in history.

51:10 And then just 4 months later, they raised another $8.3 billion.

51:13 Every time they do this, their existing investors are

51:16 giving up some of their stake in the company.

51:18 So the only reason they would go ahead

51:20 with it is if they really needed the money,

51:23 which they do, mostly to keep on building new compute capacity.

51:26 The second reason this spending is probably even worse than it looks,

51:30 has a lot to do with accounting.

51:32 Capital expenditure just means businesses spending

51:34 money on fixed assets like land,

51:36 vehicles, and in this instance, gigantic data centers.

51:39 For the accountants,

51:40 this distinction is important because if a company makes a million dollars

51:44 in gross profit and then spends a million dollars buying a building,

51:47 it doesn't get to subtract the total purchase from its revenue upfront like

51:50 it would if it spent a million dollars on a regular expense like salaries.

51:53 Instead, they have to use a depreciation schedule,

51:56 which accounts for how much value an item loses

51:58 with every year of use in line with industry standards.

52:00 For server equipment, that is 5 years.

52:03 Which means if a company spent a million

52:05 dollars on a fancy new rack of Nvidia GPUs,

52:08 they would only be able to claim $200,000

52:10 a year as actual expenses against their taxes.

52:13 Normally, companies want to maximize

52:16 this appreciation to minimize their tax liability,

52:18 artificially making things look worse in the business than

52:21 they actually are as far as the IRS is concerned.

52:25 But right now, they are kind of doing the opposite.

52:28 Chips that were state-of-the-art in 2022 when

52:31 Chad GBT first launched are almost completely useless

52:34 in today's modern AI data centers and are

52:36 largely not even worth the energy to run.

52:38 It's highly unlikely that the billions of dollars in chips that they

52:41 are buying right now are going to take 5 years to fully depreciate,

52:44 even though that is what it's going

52:46 to say on their official financial statements.

52:48 This treadmill of technology has created a higher

52:50 level of spending than any other project in history.

52:53 And what is easy to forget with all of these insane numbers being

52:56 thrown around is that these are resources that could have been used elsewhere.

53:00 One way that you could interpret these numbers

53:03 is that the rapidly escalating AI arms race has

53:06 become an incredibly efficient machine that turns investor

53:09 dollars into e-waste with a byproduct of brain rot.

53:11 But maybe we are getting ahead of ourselves.

53:14 So far, total spending on just this technical hardware alone has

53:17 amounted to around a trillion dollars by just these four companies.

53:21 According to data from Stanford, Chinese state-owned enterprises have invested

53:25 an additional $500 billion into AI infrastructure.

53:28 And then every other organization in the world that isn't Amazon, Meta, Google,

53:32 or Microsoft has dropped in another half trillion working

53:35 backwards off the sales data from Nvidia and AMD.

53:38 But that's just the first item on the budget sheet.

53:41 And these companies are finding out the hardware is the cheap part.

53:44 The costs of getting AI to where it is now have already been staggering.

53:48 But when you do a costbenefit analysis,

53:51 it's also important to look at, well, the benefits.

53:54 The only problem is those benefits have been surprisingly hard to find.

53:58 A report by MIT tracked 300 companies implementing

54:01 generative AI into their workflows on an enterprise level.

54:05 Despite an additional 30 to40 billion of investment,

54:07 which we can add to the tally,

54:09 the study found that 95% of the organizations generated zero returns from AI.

54:15 Now, for what it's worth,

54:16 I would like to say that I don't want to completely ride off AI,

54:20 even though I know it's sometimes really easy

54:23 to hate on how it's being sold to the world.

54:26 AI, like most new technologies, is coming along with a bit too much hype,

54:30 but underneath that, there are some cool things you can do with it.

54:33 I know I have used this example before,

54:35 but even here at Little Works Media Group,

54:37 we used to spend hours every week cutting

54:40 out images to use as assets in our videos.

54:42 Now, most of that can be done with AI tools

54:45 letting our editors actually work on something creative or, you know,

54:47 just getting some time to go out and touch grass.

54:50 These are real practical tools that have

54:52 helped make someone's working day easier, more productive, and less monotonous.

54:56 The problem with this, though, is that when companies, governments,

55:00 and uh nonprofit organizations have poured

55:02 trillions of dollars into developing this technology,

55:04 useful practical tools are not good enough.

55:07 Companies want to be able to replace their workforce,

55:09 not make their work slightly easier.

55:11 Governments want a super intelligence, not something that can trim out JPEGs.

55:16 And honestly, I don't even want to know what someone like Peter Teal wants.

55:20 You would prefer the human race to endure, right?

55:23 Uh, you're hesitating.

55:24 Well, I Yes.

55:26 I don't know.

55:27 I I would uh Yes.

55:30 Okay.

55:31 But but whatever these grand ambitions end up being, the bill is adding up.

55:35 And we haven't even got to the big expense yet, energy.

55:38 Earlier this month, a group of AI researchers started making

55:42 headlines after they returned from a tour of facilities in China.

55:45 Their claim was that America had probably already lost the AI arms race.

55:49 Not because China had better technology or researchers,

55:52 but because they had an actual functional electric grid.

55:55 According to the Department of Energy in a report,

55:58 the foundation of America's electrical grids were built in the 1960s and '7s,

56:02 and a lot of the infrastructure is still being used today,

56:05 well past the end of its intended life cycle.

56:07 China has industrialized much more recently than we have here in America.

56:12 So, while our grid is barely holding on, they have had

56:15 the advantage of practically building theirs

56:16 from scratch with more modern technology.

56:18 Now, this still hasn't been cheap for them.

56:21 But since the 1990s,

56:22 while America's economy has been primarily fueled by consumer spending,

56:26 China's has grown on exports and construction.

56:28 According to their national statistics

56:30 and corroborated by Bloomberg Intelligence,

56:32 the country spent nearly $2.3 trillion on infrastructure in 2022 alone,

56:37 compared to $1.1 trillion worth of spending in the US, spread out over 5 years.

56:42 This is not to say what China is doing is a good idea either.

56:45 This kind of spending has put them in a potentially ruinous amount of debt.

56:49 But the reason this is important for the AI discussion is

56:53 because Chinese grids frequently run with energy reserve margins of 100%.

56:56 Which means it has twice the energy capacity it needs.

56:59 Whereas in America, we operate with margins of just

57:02 15% according to the researchers who made this report.

57:05 That means that AI data centers in China are

57:07 actually a useful way to soak up extra energy capacity.

57:10 While the same facilities here in America

57:12 are a strain on already crumbling infrastructure.

57:14 The International Energy Agency estimates that data

57:17 centers accounted for 1.5% of the world's electricity

57:20 consumption in 2024 and is projected to double

57:23 by 2030 if current trends continue with AI.

57:26 If you were to get that out of your outlet at home,

57:29 450 terowatt hours spread out over the last 3

57:32 years only would have cost you around $18 billion,

57:35 assuming you were able to negotiate wholesale

57:37 rates of about $40 per megawatt hour.

57:40 That's actually not bad, but it's far from the whole story.

57:43 That's just the cost that the AI operators will pay, not everybody else.

57:47 You have probably noticed that your electricity

57:50 has become more expensive over recent months.

57:52 Or maybe you haven't because let's be honest,

57:55 everything has become more expensive over recent months.

57:58 But anyway, I want to play a little game.

58:01 This is average energy prices across American cities according to the BLS.

58:04 Can you see anything that stands out?

58:07 Let me give you a hint.

58:08 Major AI development really kicked off right here.

58:12 Now, the war in Ukraine, more EVs,

58:15 and overall inflation played a part in this, too.

58:18 But here in America, we never really relied on Russian energy.

58:22 Before the recent surge in AI,

58:24 electricity prices were actually declining compared to inflation.

58:27 But now, they have almost doubled within the last 3 years alone.

58:30 This is costing us all in more ways than just

58:32 the big beautiful energy bills we have to deal with every month.

58:36 Other industries like manufacturing also use a lot of energy.

58:38 And if they have to pay twice as much for this crucial input,

58:42 they are going to pass those costs along to all

58:44 of us and be less competitive in global markets.

58:47 According to the Energy Information Administration,

58:50 use electricity expenditures totaled $419 billion in 2021 before this jump.

58:54 They have not published data on 2024 yet,

58:57 but if we were using more electricity and paying almost twice as much for it,

59:01 the actual cost fell across the economy is easily an additional $400 billion.

59:05 Again, this is just in America.

59:07 Now, some of the largest AI players have responded

59:10 to these market conditions by just commissioning their own power plants.

59:14 But other businesses are pursuing a more tried and tested approach,

59:17 begging the government to fix their problems for them.

59:20 Big tech companies have been lobbying hard

59:22 to get money allocated to upgrade energy infrastructure.

59:25 They are using the threat of moving

59:27 their operations offshore if their needs aren't met,

59:29 which is an effective message since the government

59:31 sees this as such a strategic industry.

59:33 Now, if we end up getting a better

59:35 electrical grid because some big tech companies wanted it,

59:38 maybe that might actually be an example

59:40 of political lobbying accidentally doing some good.

59:42 But according to the University of Texas Energy Institute report,

59:45 this could take trillions of dollars,

59:47 and nobody is arguing that the grid doesn't need an upgrade.

59:50 The problem is that nobody else is getting a say

59:52 as to whether there are better places to be spending that money first.

59:55 The real question is though,

59:57 are there really better places to be investing right now?

1:00:00 If everybody has agreed that this is a bubble,

1:00:03 then why are informed investors still piling

1:00:05 billions of dollars into it every month?

1:00:07 Are they dumber than us or do they see something that regular people don't?

1:00:11 Now, I want to say that for the record,

1:00:13 I think this whole thing is absolutely cooked.

1:00:16 However, the best way to really understand something is

1:00:19 to not seek out information that confirms your beliefs,

1:00:22 but instead those that challenge them.

1:00:24 So, to play devil's advocate,

1:00:26 how is it possible that this whole thing is not just speculative mania?

1:00:29 Now, the thing about bubbles is that usually they

1:00:32 take some kind of outside force to pop them.

1:00:34 The dot collapse was kicked off by a combination of factors,

1:00:37 but three really stand out.

1:00:39 Microsoft being sued for violation of antitrust laws,

1:00:41 Micro Strategy doing a massive revising of their financial results,

1:00:45 causing their stock to fall 60% in a single

1:00:47 day and getting them into hot water with the SEC,

1:00:50 and a single article that pointed out the unviable

1:00:52 business model that a lot of companies were running on.

1:00:55 A few years later, the housing bubble

1:00:57 was popped by the subprime mortgage crisis, which itself was kicked off by rate

1:01:01 resets and loan originators filing for bankruptcy.

1:01:03 So, if the AI industry was in a bubble,

1:01:05 it's had a lot of things that have come a long way, which could have popped it.

1:01:09 Market instability around tariffs, legislative controls over AI chips,

1:01:12 rising interest rates,

1:01:14 legal challenges over training data, organizational shocks,

1:01:16 infrastructure problems, concerning studies over business use cases,

1:01:19 and what looks like increasingly desperate

1:01:21 attempts to generate any kind of revenue.

1:01:23 That's not to mention that if one article

1:01:26 back in 2000 could start unraveling the dotcom bubble,

1:01:29 surely the daily articles coming out about the problems

1:01:31 in this industry should do the same, right?

1:01:33 Well, so far at least, the market has pretty much just shrugged all of this off.

1:01:37 And it's been able to do this for three very important reasons.

1:01:40 The first is where all of this money is actually coming from.

1:01:43 Over the last decade and a half,

1:01:45 big US tech companies have slowly built up an enormous pile of cash.

1:01:49 Outside of insurance or financial firms,

1:01:51 which are legally mandated to have cash on hand for compliance reasons,

1:01:55 tech companies like Microsoft, Meta,

1:01:56 Apple, and Alphabet have more cash than any other companies on the planet

1:02:00 after saving it away for the better part of two decades.

1:02:02 They had been doing this for two reasons.

1:02:05 The first was that before 2017, tax laws heavily incentivized shifting cash

1:02:09 into offshore accounts primarily held in Ireland.

1:02:11 The exact structures that they used to erode

1:02:14 these profits into a haven like this were very complicated.

1:02:17 But once the cash was there,

1:02:19 they couldn't really touch it unless they wanted to pay tax on it.

1:02:22 This means that they just slowly accumulated

1:02:24 cash waiting for something to use it on.

1:02:26 However, in 2017, these companies were offered a one-time deal to bring

1:02:30 back these offshore savings into America for a small tax concession,

1:02:33 which gave them all a lot of dry powder to make some big local investments.

1:02:37 Now they have used a lot of that money to do share buybacks and they have kept

1:02:41 a lot of it abroad to fund their international

1:02:43 operations but they also earmarked billions for future capital expenditures.

1:02:48 Now the second reason they built such huge cash reserves was because

1:02:51 for a while these companies were actually

1:02:53 struggling to find projects worth investing in.

1:02:55 They had become so dominant in the respective

1:02:57 markets that spending a lot of money on development was seen as an unnecessary

1:03:02 expense that wasn't really worth the risk.

1:03:04 Today that has obviously changed and the big companies

1:03:07 that are driving most of the expenditure on data centers

1:03:09 are almost making up for a lost decade where

1:03:11 they were arguably not investing enough money into new projects.

1:03:14 This also applies to all of the money they are introducing

1:03:17 into the system to support less established firms like Open AAI.

1:03:21 Sure, the money is getting passed around a lot once it's in the system,

1:03:24 but the initial source of these funds is largely coming

1:03:27 from piles of cash these companies had sitting on the sidelines.

1:03:30 This means compared to something like the housing bubble which

1:03:32 was propped up on a lot of debt and rigid derivatives,

1:03:34 these companies are at least building a top a solid fiscal foundation.

1:03:38 Now, of course, having a strong foundation does not guarantee

1:03:40 that the house you build on top of it will also be good.

1:03:43 Similarly, just because these companies happen to be

1:03:46 holding on to trillions of dollars doesn't necessarily

1:03:48 mean that setting it all on fire to chase one single bet is a good idea.

1:03:51 It also doesn't mean that regular investors won't get burned

1:03:54 if this money suddenly gets yked back off the table.

1:03:57 One of the most concerning trends that has developed in the space is

1:04:00 the circular dealing between all of the companies

1:04:02 involved in different areas of this industry.

1:04:05 We first covered this about 2 months ago when Oracle stock

1:04:08 price spiked after reporting a huge data center rental commitment from OpenAI,

1:04:11 who had raised billions of dollars from Nvidia,

1:04:14 who made that money in the first place

1:04:16 by selling graphics cards to companies like Oracle.

1:04:19 Since then, the deals have only gotten bigger,

1:04:21 and reporters have done a really good job piecing together

1:04:24 just how far and wide this web of financial[ __] goes.

1:04:27 Now, most commentary has rightfully called this out

1:04:30 as companies pulling themselves up by their own bootstraps.

1:04:33 They're making their revenue look better than they

1:04:35 really are by investing in their own customers.

1:04:38 Not exactly a sustainable business model.

1:04:40 However, to play devil's advocate again, there is something to be said about

1:04:44 this strategy from a risk mitigation perspective.

1:04:46 When people look back with the benefit of hindsight at the dotcom bubble,

1:04:50 they all say the same thing.

1:04:52 Yeah, the market was dumb.

1:04:54 But after a major correction, there were still some big winners that emerged.

1:04:58 Some of them being the same tech companies involved in the AI market right now.

1:05:03 Nvidia investing into a company like OpenAI

1:05:05 right now looks a little bit suspicious.

1:05:07 But had a company like AOL invested in Amazon back in 1999,

1:05:11 we would probably be a little less critical.

1:05:13 By investing up and down the supply chain,

1:05:16 if you could call it that, these companies are in theory maximizing

1:05:19 the chance that they will capture the value eventually generated through AI.

1:05:22 There is also one other really important detail that a lot

1:05:26 of people gloss over when exposing this financial circle.

1:05:28 It's easy to look at this and conclude that this whole

1:05:31 market is just a Ponzi scheme popped up by investor hype.

1:05:35 The only problem with that deduction, though,

1:05:37 is that they aren't really taking investors money.

1:05:40 Netted out, the major players in this industry are paying out far more money

1:05:44 through dividends and stock buybacks than they

1:05:46 are taking in through stock issuance or borrowing.

1:05:48 When compared again to the dot bubble, the story was very different.

1:05:52 Hyped companies were dependent on bringing

1:05:54 in a continuous stream of investor money to keep

1:05:56 the lights on in businesses that made no

1:05:59 profit and often didn't even make any revenue.

1:06:01 A company like OpenAI is also in this position

1:06:04 where if they don't keep on bringing in new investors,

1:06:07 they won't be able to continue operating,

1:06:09 but they are not raising money from regular investors.

1:06:12 They are primarily getting it off companies that have plenty of cash to invest.

1:06:16 Now, I am not exactly going to say I feel bad for big tech companies,

1:06:21 but nothing they do with their money right now is going to be popular.

1:06:25 If they make capital investments into data centers,

1:06:27 people will say they are blowing their money

1:06:29 on chips that will be obsolete in two years time.

1:06:31 If they buy their own shares,

1:06:33 people will call them out for driving up demand on an already overvalued stock.

1:06:36 And if they buy shares in other companies,

1:06:38 people will call them out for circular dealing.

1:06:40 Yeah, I know.

1:06:41 I am sure they are truly devastated.

1:06:44 Now, with all of that said, that doesn't mean that these companies

1:06:48 and the wider economy are completely safe.

1:06:50 Share prices are clearly elevated

1:06:51 on the expectation that AI products and services

1:06:54 will eventually bring in trillions of dollars

1:06:56 to major participants in this industry.

1:06:58 If this doesn't pan out, then those prices could be reconsidered very quickly.

1:07:02 The major incumbent players aren't at immediate risk

1:07:05 of collapse because they still have far more

1:07:07 cash than debt and they still have functional

1:07:10 parts of the business to fall back on.

1:07:12 Their biggest risk is that if this does go tits up,

1:07:14 then they will have to explain to their investors

1:07:16 why they thought it was better to spend

1:07:18 hundreds of billions of dollars on redundant data centers

1:07:20 instead of just paying out that money to them.

1:07:23 Now, nobody can truly predict what the future of AI will look like.

1:07:26 And even the CEOs themselves have admitted

1:07:28 that, but they are framing it like this.

1:07:31 They are betting $500 billion on a dice roll.

1:07:33 If it comes up to six, they will make $10 trillion.

1:07:37 It's a risky bet, but that doesn't necessarily mean it's a bad bet.

1:07:40 Oh, and it takes the sting off knowing that the government will probably

1:07:44 be there to comp them some chips if they just keep the game going.

1:07:47 But we don't say that part out loud.

1:07:50 Now, if you are still not convinced, don't worry.

1:07:52 I'm not convinced either.

1:07:53 But go and watch this extended cut next to see why

1:07:56 we might just not be able to trust the numbers anyway.

1:07:59 And don't forget to like and subscribe to keep on learning how money works.