Hi @Strawman, can’t open the page. Pls check the link? Cheers!

It was a "Mailto" link @SayWhatAgain (behind the text) - this one works: https://habla.news/u/[email protected]/1755292995242

If that still doesn't work - just copy and paste this: https://habla.news/u/[email protected]/1755292995242

I'm getting about a 25% success rate - as in 75% of the time the page doesn't fully load then on the fourth attempt (approx.) it does.

Aha! Thank you, @Bear77 got it!

Oh right, apologies for that @SayWhatAgain. Thanks @Bear77

Sobering reflection in the dc frenzy. Thanks for sharing @Strawman! I agree that technological revolutions don’t guarantee sustainable profitability, and as he well reminds us, overbuilding without profitable demand can leads to painful market corrections! Good read :)

I agree with this take. But just to take the other side, what if you can build a self-improving AI by just chucking more compute at it (https://ai-2027.com/)? Because labour is the biggest part of the world economy you get the biggest transition in economic growth since the industrial revolution. Everything else, the internet, cars, etc, just a footnote, growth wise. All of that is to say, it's rational for them to invest so much, while other bubbles weren't! Even if the chance of success is small it's literally the largest payoff in human history and probably the last growth mode transition. Now with GPT5 not showing as most improvement as previous iteration it seems like the runaway growth model is less likely than before so it makes sense to discount this scenario more. I expect more bubble takes to come, not sure how much to hedge for this though. I'm sure a chance at taking over the world is worth 400b a year, and it's risky for us pleb to have no exposure - it's the only mistake in your life (assuming property rights still exist) you can't afford to make.

@Strawman - thanks for that. Absolute cracker of an article. Ought to be compulsory reading for anyone investing in the AI space, if only to be forced to face these arguments and take them on board.

That's a great point @twee. For me the challenge is in identifying who the big winners will be, because they'll be a small minority of those that enter the fray, and probably even in unexpected ways.

Eg, I saw someone mention energy as the best play, because we'll likely need a lot more of it if compute really amps up.

I also think sometimes with paradigm shifts it's best to hold off a little while the eventual winner is unclear, but go in hard once a dominant lead is established. You miss out on a lot of the early upside, but there are still great gains to be had from that point, and on much better risk adjusted terms (eg like when Buffett backed Apple).

I've mentioned it a lot before, but that's the gist of The Gorilla Game book, which talks of tech investing.

In the hard takeoff scenario, it doesn't matter as much who you pick as long you have some exposure to the effects, like.an ETF, because of the way the maths works. If the economy is doubling in the space.of days, waiting is the wrong move. But yeah for all other normal, more likely,.scenarios that makes sensre.

Maybe talking past ya, but I guess the way I think about it to have a small bucket of broaf market exposure, gazillion ETFs that.do this, that gets exposure to any of these low probability, high impact events Then you do your picking winners in another bucket but just focus on quality companies and don't care about the crazy scenarios.

It is a really interesting article. Thanks for sharing.

A very unique perspective on the earnings vs costs. However, I feel his assumption that the way they are currently generating income will remain the same moving forward may be a little narrow. What about data? AI will be recording record amounts of data. How about advertising? Will there be an option in the future to pay for AI to push customers to your business? Will subscriptions become more necessary. For example, Microsoft includes Copilot in its 365 subscription. As businesses use it more and become more reliant on it, will they introduce different tiers with different functionality? A cut down version is included for free but if you want to keep using it the way you have it is another $5 a month?

I don't know the answers to these questions, but I am sure the smart people have a long term view and have already planned out different revenue streams. Take Apple CarPlay for example. It is included in most new vehicles and people that don't have it in their car want to try and get it. Surely Apple are looking past the small licensing small and will be planning to use the data and functionality to advertise in the future. Big fan of Guzman y Gomez? Up pops an ad on your screen in advance of you driving past one. Don't want ads? Pay a ad-free subscription.

But where to invest? Do you go a chip maker like NVIDIA? Maybe earlier in the chain to miners or raw material? Or later with energy like @Strawman mentioned. Maybe find a mix in something ... like silver with its industrial application and a store of wealth (next to gold)?

All in all, AI is a very interesting space and I really enjoyed the "old school" assessment in the article.

Funny i had no exposure to cannabis, shale or the tech bubble, didn't believe in any of them, but I'm a believer in the AI revolution. are the numbers big? yes very big. Will there be losers, stranded assets? most likely, hard to identify the losers at this stage. I wrote a piece a little while ago, MSFT looks best placed-but maybe OpenAI dependent? i don't like that. GOOGL has a full stack, Deeepmind/model/infrastructure/distn, AMZN is the leader in cloud but probably not a technologically AI advanced as GOOG or having the embedded connections of MSFT. the others i think it will be hard for them to find a place. Meta giving it a go. the smaller guys maybe they find a niche. Too early to identify the winner. on all the earnings calls the hyperscalers (all of them) are consistently saying demand outstrips supply, do you believe them? at some point, they probably go beyond demand for a period. the reality is that corporations see not embedding AI as leaving them vulnerable to competitors, as long as that holds the hyperscalers are on a winner.

im a huge sceptic, but i can see a business case here but not for all comers. we shall see, will be plenty of opportunities to reset on this one, it is a long journey.

Agree with your commentary. Co-Pilot is getting plenty of use because of the way it is bundled into existing Microsoft packages. My feeling is that the boffins will go for powerful uses like they did Linux, but the average dude in the street (me) will stay within the training wheel lanes and be satisfied with what they get. Afterall we will never know what is possible because we lack the depth of knowledge to experiment on optionality.

A different way of playing this game short term is to ‘pick and shovel’ the development of data centres. SXE is a big holding for me though, of late, SKS has emerged with a number of contract wins. This might explain the unusual market fall of SXE after an excellent FY25 result. Their order book isn’t increasing amd maybe SKS are pinching their bids.

Expanding on @Solvetheriddle's post. I don't think it is a matter of if we are in an AI bubble, but when.

As far as I know, most major technological advancements (e.g, railroads, semiconductors (in the 70s and 80s), the internet) have resulted in over-exuberant investment and capex spending. Which, while it created a boom and bust cycle, it set up for the future to take advantage.

The point is, inevitably, they will make too many AI data centres and investors will lose as they won't make the money back.

However, I think AI will enable a massive level of productivity increase - but again, it will be a matter of not if, but when. I think there is a good chance our understanding of the capability of AI will take some time to establish - There is a good chance that maybe we will see the limit of what, the current conception of, AI will get to. Maybe we won't get to AGI, or super intelligence, that can become a self-improving flywheel of intelligence; it'll get to a point and stop getting better. However, even then, even with what we have now, there is massive capability waiting to be realised.

But, I won't be surprised if this takes longer than people think. It will take time for business and tech people to set-up systems and databases that enable the best use of AI. And, I think, while it is improving so rapidly it is hard to know when to start - why build to the capability of AI today, when it will be surpassed in a few months. We almost need to speed of it's advancement to slow down some (which, as noted about GPT5, may already be happening).

What does this mean for investing for me? Pretty much what everyone has been saying. Look for the picks and shovels, look for the clear winners and buy them, buy an index (productivity increase means a likely out performance across the board) - but expect it to be rocky, there will be over-exuberance and there will be a crash/correction/pullback.

Interesting view of the economics of AI, providing context of cost to revenue opportunities is a great reality check. However, I will provide a flip side perspective on some of the figures, not as an AI follower, just trying to get an idea of where it’s heading and the opportunities.

Let’s take the articles need for $480b in revenues. I would question 25%GM, it’s software – 75%+ GM% is more likely, but it matters little per below.

If AI is replacing or augmenting workers, then the context is the value of the work or workers replaced and unlike Netflix, this is an enterprise business where uses have very big price tags if they can justify the cost saving or increase in gross margin $.

Perplexity tells me total wages in the US in 2023 were $11.08T, so $480b is 4.3% of US payroll. It also says average wages are $80k but median is around $60k so let’s take that and so $480b would need to replace 8m jobs full or FTE.

That’s just the US and just payroll savings – not savings on other information systems that AI would replace.

Added to this, their will be a few winners and a lot of losers, so the ROCI for those winner will be massive as they eat the lunches of the losers. Not playing this game is not an option for the big tech companies – they can’t afford to lose.

I will leave it to those who know a lot more about the industry to suggest possible winners or ways to pick it – it’s outside my circle of competence, so I am not playing, but it looks like it’s worth playing if it is your thing. 

Great article, especially since every week or two I feel like I read about a tech CEO’s new record prediction for data centre builds or energy needs.

Does the picks and shovels method make the most sense then I wonder? Something like ASX:AINF.

Profit off the buy and build process, then sit back and relax if there is indeed a realisation that all of those data centres aren’t actually necessary/profitable. Although in that situation, the global economy will get a little shake up regardless.

Are a quick look AINF could be a good option for a short term play @Stumpy as there are lots of names in there that benefit from the current demand for data centres, chips and power.

Probably not for me though as the assets under management/trading volume are a bit too low for my risk profile at this stage. 

Quote from the acritical: I don’t pretend to understand technology. However, I’m a guy who understands cash flow, and there is none. I don’t see how there can ever be any return on investment given the current math. Instead, I just see endless losses, and we’re far enough along in this S-Curve, to think that we can at least start to model the returns—except they’re horribly negative. If the management teams at these megacap tech companies do not pull the plug on this adventure, eventually the shareholders will. I shudder to think about how nasty that could get for equity markets.

 With all due respect to the author, this one paragraph, and more specifically the section I’ve highlighted in bold, says everything to me. Through that lens, I can’t agree with most of what they’ve written or the conclusions they’ve drawn. Put simply, AI is going to cause more disruption than most people currently expect, and it will arrive far quicker than most anticipate. We are still very early on the S-Curve, but we’ll rush through it in just a handful of years, and to compare what’s happening in AI to the Dot-Com bubble or similar events is to fundamentally underestimate both the speed and scale of what’s unfolding.

Why is that? Three reasons: AI’s impact on software, hardware, and the new economy.

Most commentary I see focuses on the software side. I agree there’s a strong argument this will become a race to the bottom, as the ‘free’ versions of AI agents improve and unlock an explosion of apps that’s hard to fully grasp right now. The impact of literally anyone being able to create their own software or apps is going to be immense, especially for software development companies and developers themselves. We already see early signs of this, for example, Meta employing a relatively small number of developers on million-dollar salaries. Further, jobs that are focused on taking a dataset and applying a set of rules/thoughts/logic to reach an outcome (think Lawyers, x-ray analysis and the list goes on and on), not to mention call centre staff (I say that acknowledging the discussion on the Motley Fool yesterday, I just see it differently), and a number of other professions is already impacting hiring decisions.

But let’s move past software for a moment. Robotics has to be factored into this conversation. Self-driving cars are already here and will scale into the tens, if not hundreds, of thousands over the next 3-5 years at the most. Just think of what that means for jobs, as well as the massive upside for companies focused on AI, data, and power storage. And what about robots equipped with AI that performs at a PhD+ level across factory, office, and household settings? The potential use cases, and impact on jobs, not to mention profits, are almost endless.

Lastly, the combination of AI and stablecoins, not to mention Bitcoin, will have a profound impact on banking/the financial system. While I can’t do this topic justice here, it’s enough to look at what PayPal, Visa, BlackRock, and even JP Morgan are already doing in the space. This shift will be huge, generate significant returns, and reshape traditional banking and financial systems.

So, how do you invest in this space? This is where I agree it’s still far too early to pick winners purely on the software side. That said, we can make educated guesses about where robotics will have an initial, outsized impact, while also positioning ourselves for what’s coming in financial services. As has been noted before, it helps to think of the gold rush days: we know there’s gold in the hills, but we can’t be sure which claim holds the most. What we do know is that shovels, picks, food, and shelter will all be needed. Applying that lens, it’s safe to assume demand for semiconductors will continue for a period of time yet (and not just NVIDIA, Elon Musk is buying chips from Samsung), alongside power generation, the rails for financial infrastructure, and autonomous systems ranging from cars to highly capable robots to mention but a few areas for consideration.

Personally, I’m taking a basket approach across these areas, primarily through funds and ETFs. I’m not smart enough to pick the individual winners, but being broadly right is likely to do well. Or to borrow a quote I’ve shared before that sums this space up perfectly: Wayne Gretzky was one of the greatest ice hockey players because he would “skate to where the puck is going to be, not where it has been.”

Thanks for sharing this @Noddy74 i hadn’t seen it. But it really does reinforce how we’re fast moving into a new paradigm.

I know people will reasonably point out where this stuff still falls short, and how humans still lead in other important ways. But the magnitude and pace of progress is just wild. It doesn’t feel like we’re anywhere near the limits yet. And even if we were, what wwe already know is possible is enough to shift a lot of things. Like, the Zero to one moment has happened. From here, brute force optimisation and refinement can take us a long way. Probably far enough to shake things up quite a bit.

Lots won’t change (humans gonna human), but ive been thinking lately that some interesting things that might change are:

Cross-domain talent -- the most valuable humans might be the ones who can move comfortably across domains. Systems-level expertise will be the easiest to commoditise. You want people who can orchestrate and align different resources.

Explosion of micro-enterprise-- anyone can now spin up a business (especially online) at a fraction of the time and cost. A lot will fail, sure, but the overall system becomes way more dynamic and specialised. Ideally that drives more value creation.

Operational flex --businesses could get a lot more done with a lot fewer people. That's already been the case for the last 50 years, but revenue per employee could be truly gigantic for those that achieve scale (Tether has recently achieved an eye watering $93m/employee in 2024, which is not an example of AI, but what happens when tech scales rapidly to a global customer base)

Fascinating stuff @Noddy74

@Strawman , sorry I had to take the bait !

im officially out of a job! Mostly because I’m retiring but it would definitely give me cause for thought if I was just emerging into medical practice. Certainly teaching medicine is going to get even more outdated very quickly. That’s probably generalisable to most professions

I would raise a couple of points:

• the data set is basically a bunch of “House” cases. Interesting rare and difficult to diagnose. In many cases clinicians would never have heard of the condition let alone seen one. A simple google search of plugging in a list of symptoms for this subsection of pathology has also been proven to be more accurate than a clinician.
• It’s a work backwards study not a work forward one. By this I mean, it compares how accurately, quickly and cheaply AI performs in situations where there is a true but rare pathology that is difficult to diagnose vs an experienced physician.
• Imagine one were to put the same collection of initial symptoms into the algorithm that these rare cases presented with for every patient that presented with them, would one end up with the same diagnostic accuracy?
• The google symptom prompt study and this analysis would suggest yes. I would counter that the vast majority of those patients would have an entirely different pathology. And top of the list would be an anxiety related somatisation issue: the physical symptoms of a primarily psychological disorder.
• we tend to organise symptoms into systems Respiratory, gastrointestinal, neurological etc. When you get symptoms in more than 3 systems the chances of it being a psychological problem increase exponentially.
• so much of the diagnostic process involves a weighting of probabilities based on data that is not captured: eye contact, non verbal communication, their relatives behaviour, previous attendances of a similar nature, seeking reassurance and on and on. As yet I don’t think AI algorithms are anywhere near incorporating this.
• approximately 20-30% of patients in any of my given shifts will have at least some physical symptoms related to an underlying psychological distress that if incorporated into a diagnostic algorithm as ACTUAL symptoms would be disastrous. (low specificity) eg shortness of breath and chest pain = terrible badness vs anxiety attack.
• of course the flip side is sometimes symptoms are attributed to psychological distress when they shouldn’t. (Better call Saul)
• I have no doubt they will be able to outperform in real life clinical scenarios, probably earlier than any of us think!

I am not immune to my own bias in this situation. I strongly believe that the use case for AI in medical diagnostics exists already in many areas (certainly in many types of imaging), will rapidly expand and may well end up being better than humans in most areas. I’m not convinced this study is that representative of generalised diagnostic superiority, though.

In the medical sphere i am particularly hopeful for the use case of AI in drug design and development. Alpha fold and similar projects will dramatically shrink the time and cost of both the discovery and development of new drugs. Bring it on, pharma needs a revolution as the current system is broken.

I have been trying out Gemini lately after they released the 3 Pro model recently (although I think I'm mainly using 2.5 as I'm not going paid yet). Is it weird that I felt the need to close my ChatGPT browser for fear it might disapprove? On my phone I use Claude and only Claude, but I won't use it on my PC. Anything else feels like I might not be doing the right thing by Claude or the other models! It's like AI Anthropomorphism. Gemini says it's also called the ELIZA effect (but don't tell ChatGPT where I sourced that!).

@Noddy74 I'm afraid to break it to you, but after 30 days (or whatever it is) the SM paywall/embargo comes down, and everything we write here becomes LLM fodder.

I've even had the work of illustrious StrawPeople served back at me by AI in referenced deep dives! Many times.

Sorry, but you are busted.

I've also been using Gemini 3 quite a bit lately (I'm using the paid "Thinking with Pro 3" model)

It does seem to be surprisingly good... the only thing I hate is that when it's dealing with large outputs (to Chat or Canvas) it's basically guaranteed to eventually freak out (it either just randomly truncates bits out of inside your expected responses (you're expecting a couple hundred KB of code or text or whatever and you get tens of KB), or worse it spends ages (15 mins?) "Thinking" (and progressively building and outputting the response) before it just randomly logs you out and when you reconnect it's forgotten your last request and it's response and you often try again hoping it will get it done this time, only to be disappointed again in an almost endless loop – it's very frustrating and unreliable

Looking at comments online there's many reporting the same thing and it appears it's actually been this way for ages (months to years) with no actual fixes from Google. I was hoping going from Pro 2.5 to the new V3 release would resolve the issue but alas ....

The only way I can get it to even work (partially) is to not enable and use Canvas mode for chats that will involve large outputs (anything over about 150KB in a response) and get it to output what I want to the Chat instead (e.g. in a codeblock) - sometimes I also have to get it to split and chunk the output and "hope" it gets the splits right and the pieces actually join together and without truncation or missing pieces (which often they don't)

I have had some success however in getting it to modularise the thing I'm working on (split a larger codebase into multiple 10-50K chunks) and work that way and get it to only output each of the modules which have been updated one at a time with it waiting for me to respond inbetween. And eventually when things start going pear shaped (which it invariably does when the chats get very long) I just start a completely new chat and load the last working iteration of all the modules at the start of the new chat and continue working from there

This is not the way I want it to operate, it makes it much slower and more tedious than it needs to be, and not being able to use Canvas is disappointing (it doesn't work with modularised code anyway so there's that too)

Love this @Noddy74.

I've been going down the AI rabbit hole too ... just in time for Easter!

It's been nice to anchor to something less ephemeral than the moods of our new global overlord...

This Alice in Wonderland reorg of international trade makes the AI assault on humanity seem like the logical successor to the old mutually assured destruction thesis that kept the weapons of war in the holsters of their owners back in the 80's.

I like the Mogan Housel angle that the first 50 years of the 20th century were the biggest step change in humanity - we entered 1900 on horseback and by 1950 were in the nuclear age. Anyone turning 20 at the turn of the last century would have not recognised the world they were living in by age 70 from the one they grew up in.

Hard to argue that we won't go through a similar transformation over the next 10, 20, 30+ years.