Apple Will 'Watch Everything Burn' When AI Bubble Bursts (macrumors.com) 136
MacRumors interviewed AI critic Ed Zitron, author of the Where's Your Ed At newsletter and host of the Better Offline podcast, about what could happen to Apple if the costly AI infrastructure boom collapses. Zitron argued that Apple is relatively well insulated because it has spent far less on data centers than rivals such as Microsoft, Google, and Amazon. He said the company could use the downturn to make selective acquisitions, though it might simply continue operating largely as before. "I think they will sit on the sidelines and watch everything burn," Zitron said. Here's an excerpt from the report: If the bubble deflates the way you expect (write-downs, too much compute supply, possibly OpenAI cratering), can you walk us through what that would actually look like for Apple? Does anything break for iPhone users, or does Apple mostly watch it happen from the sidelines? Could it even benefit Apple?
I think things would look much the same for Apple. I think they will sit on the sidelines and watch everything burn. I could see them doing some choice acquisitions as things begin to collapse, but I could also see them do nothing.
I think Apple is in a very weird place at the moment. The Vision Pro was a dud, but it was also the most interesting and future-forward thing I've seen anyone put out in a while. If they were smart, they'd tread water and sink as much money into making that as small as humanely possible -- no matter how long it takes -- because the entire AI bubble is a result of everybody running out of hypergrowth ideas, mostly because we're flat out of new interfaces.
I want to be clear that what they want to do with the Vision Pro requires it to basically be weightless and invisible and never need adjustments. When it works, it's genuinely awesome. But that's a load-bearing when. One slight movement means the whole thing goes out of focus. I can't even use mine anymore because it needs an update that requires you to wear the thing the whole time. So much promise, released too early, shoved out the door by a CEO on his way out.
I think things would look much the same for Apple. I think they will sit on the sidelines and watch everything burn. I could see them doing some choice acquisitions as things begin to collapse, but I could also see them do nothing.
I think Apple is in a very weird place at the moment. The Vision Pro was a dud, but it was also the most interesting and future-forward thing I've seen anyone put out in a while. If they were smart, they'd tread water and sink as much money into making that as small as humanely possible -- no matter how long it takes -- because the entire AI bubble is a result of everybody running out of hypergrowth ideas, mostly because we're flat out of new interfaces.
I want to be clear that what they want to do with the Vision Pro requires it to basically be weightless and invisible and never need adjustments. When it works, it's genuinely awesome. But that's a load-bearing when. One slight movement means the whole thing goes out of focus. I can't even use mine anymore because it needs an update that requires you to wear the thing the whole time. So much promise, released too early, shoved out the door by a CEO on his way out.
AI break down (Score:5, Insightful)
I can easily see the AI bubble crashing.
But I do not expect the 'excess compute supply' to be a problem. That is very likely to be the main asset that people buy up. The price will drop for it as compared to now, but not by much.
Memory prices however I expect to crater. Once that compute supply is shifted away from making AI slop to more productive uses, we will have a glut of memory chips combined with excess production.
The main fall will be the idiots that 'invested' in AI companies. The AI founders will still come away in the 1%, but most of that money will have been striped away from the investors.
Re:AI break down (Score:5, Interesting)
But I do not expect the 'excess compute supply' to be a problem. That is very likely to be the main asset that people buy up.
I not so sure about that. There's not a lot of value in a depreciating asset. At least the dark fiber left over from the dot com boom had a life that could be measured in decades. Server farms are only good for 3 to 5 years, and even then become unreasonably expensive to operate vs. new equipment that's more economical to operate.
Re: AI break down (Score:2)
The assets not just the chips, in fact thatâ(TM)s a smaller value than the overall infrastructure. Once the massive Dc is built, they can keep swapping out the chips to keep it fresh.
Re: (Score:3)
People are currently still running - and buying - 2018-era GPUs for inference locally. In fact, most cards cost markedly more (and continue to cost more) than they did new. They're also more capable at running inference than they were even a short while ago: you can do things on a 12GB 3060 today which weren't even possible on the 24GB 3090 6 months ago ago. For the time being, these cards are still increasing in value.
Short of electrical failure, the forefront GPUs will be capable for both inference and tr
Re: (Score:3)
But I do not expect the 'excess compute supply' to be a problem. That is very likely to be the main asset that people buy up.
I not so sure about that. There's not a lot of value in a depreciating asset. At least the dark fiber left over from the dot com boom had a life that could be measured in decades. Server farms are only good for 3 to 5 years, and even then become unreasonably expensive to operate vs. new equipment that's more economical to operate.
... when in the past twenty years has replacing data center equipment been more economical? I don't get where you're coming from. We write down deprecation on IT equipment, that doesn't mean its value changes. Servers got bigger and bigger while workloads stayed about the same. It's what allowed VMware to grow into the behemoth it became. It also made replacing physical servers less and less necessary. You can run them well past their support life and replace only when truly dead or unreliable, or give up a
Re: (Score:2)
Short term server farm depreciation is why they need to be built in space.
Re:AI break down (Score:5, Insightful)
The problem is that this is not general-purpose "compute". Hence nobody actually has a real use for it that would justify the power consumption. This hardware may actually be negative worth after the cash. I also do not see why RAM should "crater". Most of the RAM used in AI datacenters is also not general purpose and has no use outside of LLM accelerators. Yes, RAM can often be customized for different usage scenarios, but that is a final manufacturing step. Once it is finished, changing that becomes hard or impossible.
What could have interesting and unpredictable effects is if Nvidia dies. That is not assured but a definite possibility. A lot depends on the details of the crash.
Re: (Score:2)
Let's look back at this comment in 5 years?
Re: (Score:3)
Re: (Score:3)
AI is better and faster than humans: in the areas where it is better and fast.
Also: it opens paths, we yet right now can not do without AI.
No idea why you AI haters do not at least start reading the basics about it.
Here are companies that craft "individual software". Every customer they serve is unique. Except for framework code, they share no common code.
The companies I work with, can do jobs, a non AI company can not do. The condensate a man year or two, into 1 or 2 14 days sprints.
That means, software th
Re: (Score:2)
AI is better and faster than humans: in the areas where it is better and fast.
Also: it opens paths, we yet right now can not do without AI.
AI is not a single entity. It will not completely work or not work. Some things will work and others not. Some things will make money and others not.
The biggest takeaway from the original comment about Apple watching AI burn is that Apple has not stepped firmly into AI. So, it will not suffer for those AI things that don't work, and it won't profit from those AI things that do work.
We often use AI LLM chatbots as the face of AI. However, that's a misleading broad generalization. As one example of AI t
Re: (Score:2)
No idea hat Apple or Android is doing with AI in Photos.
Point is: most mainstream AI stuff as in "generative AI" EVERYONE can try for free with ChatGPT or what ever.
Find video making or photo making services, and add a prompt like: """
I like two town houses in typical Thai sub urb environment, side by side.
Left side a fruit and vegetable shop, with a sign "Nonglak's organic fruits - fruit of the moon"
Right side a solar panel and infrastructure shop, with the sign: "Enkis Solar Center"
"""
And look what happen
Re: (Score:2)
"Haters" , poor language. How about sceptics?
There is a lot to be sceptical about AI. Aside from software dev and established machine learning applications, where are the big AI success stories apart from in its supply chain? What business has been radically transformed and achieved huge profits by rolling out co-pilot? Surely it should be happening by now its fairly ubiqutous.
For us coding boilerplate most of the time, sure it's a wow moment. For the ordinary punter, they've got a search and summary tool t
Re: (Score:2)
The success stories are all over the place.
But it is not my business to spread news, I just work with companies that utilize AI successfully.
I used to train AI systems for Scale.ai ...
So I might be biased, but I know hundreds of systems that work just fine.
Re: (Score:2)
The success stories are all over the place.
We have anecdotes of minor wins for AI. The problem has never been AI is not useful. The problem is companies selling AI as replacing humans entirely has been a farce with todays level of technology.
I used to train AI systems for Scale.ai ...
And how did your training help any company with cost savings, productivity, etc so that companies can just use your AI to replace humans?
Re: (Score:2)
What baffles me is why aren't many companies using open weight models. Now a days they seem to be as good as the commercial ones. Of course you will have to invest upfront in hardware or go to openrouter.
Re: (Score:2)
Same reason, most companies rent data centers, instead of running their own.
They lack the expertise and are afraid about the risks (as investing to much, having the project to implement it fail, and so on).
Side note: to run your open weight models, you actually need a kind of small/not so small data center.
Re: (Score:2)
No idea why you AI haters do not at least start reading the basics about it.
Me: AI has not turned out to be better or cheaper than humans right now
You: Hater!
Companies like Ford and Meta have had to rehire the people they laid off for AI because AI was not suitable to replace them. Those are only the most recent ones I know about.
Re: (Score:2)
[adds discussion to SlashDot 5 year time capsule]
Re: (Score:2)
Re: (Score:2)
That is what I expect as well. RAM prices will just go back to normal. And, incidentally, I did not claim anything else in my previous statement.
Re: (Score:2)
I am not sure what will cause nvidia to "die". They're still the best at what they do and AI/LLMs aren't going to evaporate, it's far too useful. I can definitely see them having lay offs though.
Re: (Score:2)
The problem is the extreme investments they made and make. Those may kill them if the timing of the crash is wrong.
Re: (Score:2)
It only took a minute to look it up and find out that they have $84B in cash and short term investments ($259B total assets) and only $64B in total liabilties. ($43B current liabilties)
If all their revenue dropped to 0 for a couple years they would not die, and that's not what a crash would look like for them anyway. It would only be a downturn for them.
Re: (Score:2)
They are an enterprise that produces goods. Your argument is invalid for those. If they were, say, an investment outfit, yes. But they have running costs and they need to keep their tech current.
Re: (Score:2)
There is no part of this which even aligns partially with reality.
Let's assume there's a catastrophic blow up with the AI market. Anthropic and OpenAI are gone, or crippled. Resold for parts. NVIDIA closes up shop or goes back to making video game cards after a restructure.
What then?
There's still incredible demand for inference. Every business is scrambling to get more, cheaper, better inference. Almost all of them - whether they're successful with it or just getting in the ropes. That demand isn't just goi
Re: (Score:3)
The cratering of RAM prices has nothing to do with people thinking they can remove a chip from an NVIDIA V100 card and put it in their motherboard, and everything to do with the fact that supply contracts are locked in for years in advance which will almost certainly be cancelled leaving market oversupply in the production sector.
Re: (Score:2)
What could have interesting and unpredictable effects is if Nvidia dies.
NVidia has $160 billion in the bank [yahoo.com], enough to keep running at a loss for while if there's a massive crash. They aren't going anywhere.
Re: (Score:3)
Why would NVIDIA die? If anyone is making money out of this AI bull run, its NVIDIA. The rest are burning money, is what I thought...
Re: (Score:2)
Nvidia seems to be the makers of shovels.
Re: (Score:2)
That is very likely to be the main asset that people buy up.
Who will buy this up? The CPU hardware for AI was specialized; they do not make good webservers using AI hardware.
Memory prices however I expect to crater. Once that compute supply is shifted away from making AI slop to more productive uses, we will have a glut of memory chips combined with excess production.
The RAM was specialized for AI called HBM; they cannot be used in place of DDR5 in consumer or servers. Those RAM chips can't even be melted down to be recycled as silicon as they have too many impurities. HBM might be used by HPC like the top 500 supercomputers.
The main fall will be the idiots that 'invested' in AI companies. The AI founders will still come away in the 1%, but most of that money will have been striped away from the investors.
The main fall will be any one that indirectly invested in AI. Retirement accounts that had stock in Intel, Nvidia, Oracle etc. Basical
Re: (Score:2)
I can see some research institutes snapping up some cheap just-out-of-warranty Blackwell 200s and running them until they die. I've got proximity to that, so it's visible to me... I can't see a lot of other use cases. I imagine they're going to be similarly niche.
Most HPC systems use commodity hardware, just a lot of it in cluster. The most recent nodes we have on the system I help to manage are equipped with either ECC DDR5 5600 or 6000, depending on which batch they came in.
Zitron's analysis definitely ag
Re: (Score:2)
"The main fall will be the idiots that 'invested' in AI companies."
That's so far from the truth. This has the possibility of completely cratering the IT industry as a whole, because these companies are not going to give up on AI. They'll cut other things first.
The blast radius is huge: OpenAI/Anthropic are heavily invested in Google, Amazon, Oracle, and NVIDIA. Those companies in turn, are invested in myriad AI research efforts (of which OpenAI/Anthropic are in no small part).
Anthropic is likely the lynch p
Re: (Score:2)
I've watched this guy on several videos. He isn't saying anything that tech folks don't know. We know this is an asset bubble, the AI hyperscalers are going to collapse, the companies left holding the bag will be Oracle, who is just rent-seeking, and all the others building data centers are rent-seeking.
Like the fact that some of these places are just burning natural gas like there is no tomorrow shows a fundamental flaw in their plans. All it takes is one backhoe 10 miles from the data center to "oops" the
Re: AI break down (Score:4, Insightful)
Re: (Score:2)
A lot of this RAM is HBM or server ECC chips that can't be used in consumer hardware. Most people probably don't want a server system at home either. They tend to be loud and power hungry.
Fortunately if you look at what Chinese recyclers are already doing, it seems likely that they will take those parts and remove the memory chips, and put them on consumer RAM modules. I'm not sure what they can do with HBM, maybe take some of the server GPUs and move them to new PCBs?
Re: (Score:2)
AI Bubble Burst Economic Impact (Score:2)
Re:AI Bubble Burst Economic Impact (Score:4)
We will see. At the moment the only national economy massively propped up by the LLM craze is the US one. The rest of the world may be essentially ok.
Re: (Score:3)
Taiwan and Korea come to mind as economies having a boom due to the LLM craze (just look at Samsung's profits!)... there are probably a bunch of others as well.
Re: (Score:2)
Samsung is a vastly diversified enterprise. They will be ok.
Re: (Score:2)
But why did the KOSPI (stock market) crash?
Patrick Boyle is a finance guy. He has a youtube video called "How to be Right and Lose Everything" which explains what's going on with the kospi. TLDR the kospi is largely made up of hardware companies with high exposure to AI. Most of the trading activity is driven by retail investors. The retail investors actually buy leveraged, auto-balancing derivatives. This makes the whole thing a vastly unstable boom.
Re: (Score:2)
hrrrm.... Softbank are up to their eyeballs and represent a fair chunk of Japanese capital
It will primarily hit the US, that's true... but it wont be confined there.
Based on how my own indexed fund is set up (what we call 'superannuation' in Australia), I expect that I'll also take a hit... I'll just be insulated by my other investments (hopefully).
Re: (Score:2)
I don't think so; I think it will be a couple weeks a choppy trading followed by cyclical bear market.
Look at this way the AI infrastructure that physically exists today is subscribed. It has paying users, it has paying users that are likely to keep paying even the higher prices after the VC money stops subsidizing.
Maybe people will get bored with having AI redraft their e-mails with include more '-' characters but they are not going back to buying stock photos, hollywood is going to want to continue to use
Re: (Score:2)
Maybe people will get bored with having AI redraft their e-mails with include more '-' characters but they are not going back to buying stock photos, hollywood is going to want to continue to use it for effects, people like better text to speech, it is effective and being a smarter grep/search. It can generate a lot scaffolding and code quickly in the right hands. The list of use cases goes on and they are real.
That's realistic. And the use cases are all low frequency and niche. It's a bit like a swiss army knife, lots of applications I don't use that often, but someone skilled could build a wonky house with it. Eventually I get bored carrying it in my pocket.
Not enough to justify this level of investment.
AI's greatest success has been marketing, over inflating its value and utility. Crypto operates on the same premise.
I can see LLM sticking around too, this time next year, we wont be talking about it. I guess r
Economics (Score:5, Interesting)
For instance, self driving cars will be a huge market. Several promising startups failed or reorganized because funding dried up, as investors shift their money into LLM based AI tech, which doesn't help self-driving cars at all.
Think about it this way. If AI goes away, will nVidia fail? Even before the huge demand from AI, they were the most successful graphics company. They use their GPGPUs in supercomputers. They are the go-to for gaming.
The other factor is AI isn't going away. The AI bubble burst will be like the
Re: (Score:2)
If AI goes away, will nVidia fail? Even before the huge demand from AI, they were the most successful graphics company. They use their GPGPUs in supercomputers. They are the go-to for gaming.,
They might if they produced a lot products they cannot sell. In favor of AI, Nvidia stopped making consumer GPUs. That means they used their allocations from TSMC to make data center computers. If the AI companies renege or do not pay NVidia for all of those chips, it is not like NVidia can sell them to consumers.
A lot of startups will go under. Most established companies will take a licking, but will be fine (Oracle is in a LOT of verticals) AI will still be around.
Oracle's credit rating was downgraded to BBB-/A-3 because they borrowed a lot of money to fund their AI expansion. It does not matter if Oracle is "in a lot of verticals" if the company as a whole
Re: (Score:2)
They might if they produced a lot products they cannot sell. In favor of AI, Nvidia stopped making consumer GPUs. That means they used their allocations from TSMC to make data center computers. If the AI companies renege or do not pay NVidia for all of those chips, it is not like NVidia can sell them to consumers.
80% of the current TOP500 supercomputer systems used nVIdia GPGPUs. Worst case scenario, nVidia sells cheap upgrades to existing supercomputer systems. There is demand for these things in a lot of sectors besides AI. Video/3D rendering acceleration, finite element analysis, fluid dynamics, etc... There's actually pent-up demand for these things because they are being diverted for AI.
Re: (Score:2)
80% of the current TOP500 supercomputer systems used nVIdia GPGPUs. Worst case scenario, nVidia sells cheap upgrades to existing supercomputer systems.
Unlikely. 1) Certainly GPUs made for AI might be able to be used by the top 500. However, NVidia integrated GPUs with CPUs for their AI chips so the two cannot be separated. The top 500 have little use for the AI CPUs. 2) The top 500 represents a very small portion of the market and most of them require years of government planning and funding. Even if the Nvidia had lots of spare AI chips the Top 500 could use, it would take years of negotiation to get them off their hands.
There is demand for these things in a lot of sectors besides AI. Video/3D rendering acceleration, finite element analysis, fluid dynamics, etc
Not when Nvidia combined GPUs and
Sure but Siri is a dumpster fire (Score:4, Insightful)
Re: (Score:2)
Siri's Decline (Score:2)
It's transcription is hit-or-miss with me. It either nails it or gets one thing bafflingly wrong.
The thing I use it for the most, playing music through CarPlay, is almost worthless. It used to get everything right, even with weird track names in French or German. Now it barely gets English bands or song titles correct. 40% of the time it will play the wrong thing, 40% it will play any random song, and 20% of the time it will get it right.
Re: Siri's Decline (Score:2)
My problem with telling Siri to play music in the car is that it just stops working. I say "hey Siri" and nothing happens. To fix it I have to stop the car and unlock and lock the phone. Sometimes I even have to reboot the phone. I guess the listening daemon crashes and nothing restarts it. This is really annoying when I have twenty more minutes of driving before my destination.
Re: (Score:2)
They changes from a simple system that could do few things but good to a more general purpose system that has some parts that suck more while providing in theory more functions than before. Years ago we didn't have multimodal LLM, so we used dedicated speech-to-text systems. They have limitations and some multimodal LLM are clearly an improvement, but others are not. And if you have one that sucks on recognizing the band name while the older system got it right (or got the phonemes right), than you're out o
I only want a simple, effective siri (Score:3)
An assistant
Re: (Score:2)
What voice assistants rely on is enunciation. You making an announcement doesn't help Siri.
Re: (Score:2)
Yep, Siri has been a dud. iOS27 Siri is not perfect but is now much, much better.
One weird thing about transcribing that I discovered years ago...Siri would constantly get my transcriptions wrong. But if I said "Hash tag" and then the message, it would almost always get it correct, but written like a stupid Twitter hashtag.
Re: (Score:2)
Siri's poor comprehension and general utility used to be a funny joke but it's gotten to the point where I'm questioning my longtime iPhone use
Worst reason to buy an iPhone in the first place. Siri? Who gives a shit?
Apple Plays It Smart (Score:5, Interesting)
From the start, the current AI iteration led by the likes of OpenAI has been a dumpster fire for investor money. No profit in sight, and a continuous burn rate that dwarfs every other financial bubble in history. A few firms in the hardware business have cashed in on the shovel trade, while outright fraudsters like the DRAM cartel have raked in a windfall, but neither is sustainable in the long term. Apple, unlike just about every other tech firm out there, has contented themselves with sitting out the AI bubble, and it very much looks like they're going to benefit. At the end of the day Apple will have pile of cash on hand to cherry pick the remains, while the likes of Microsoft and Meta will be left holding so much debt it could well sink them. Oracle will be toast.
Re:Apple Plays It Smart (Score:4, Funny)
I would be entirely fine with an outcome where Meta and Oracle die. And MS? When they meet their richly deserved fate, I will bring out the good stuff to toast that event.
Re: (Score:2)
Can I come to your party? I'll bring my good balloons :-). We can wear Facebook privacy-invaders and roast Clippy for entertainment.
Re: Apple Plays It Smart (Score:2)
Facebook knows too much about too many people. They have enough kompromat to be able to survive as a mass blackmailing business.
Re:Apple Plays It Smart (Score:4, Funny)
Generally speaking, Apple doesn't follow trends, it sets them.
Perhaps in some cases, but they were not first to market with MP3 players or phones. They just watched the early attempts and did them better. I wouldn't be surprised if Apple watches the AI market, picks a few winning approaches and makes appropriate acquisitions out of the rubble.
Apple might buy Microsoft for that cool solitaire game and scrap the rest.
Re: (Score:2)
I wouldn't be surprised if Apple watches the AI market
Apple has already bought 4 AI focused companies for many billions of dollars. The most recent of which was Q.AI, the most ignored of which was Curious AI which resulted in Core ML framework in iOS that the industry collectively yawned at, and the most embarrassing of which was Voysis, possibly one of the first and most promising AI voice assistant companies, only for 6 years later Siri to be the most useless assistant to the point where investors are suing Apple for promising they would deliver on AI.
Apple
Re: (Score:3)
Apple tried to.follow the AI trend. They failed.
Re: (Score:2)
They don't set them. They claim them for themselves. How many people think Apple invented the smartphone? A lot!
Re: (Score:2)
Depending what you want to call a smartphone: they did invent it.
If you think the BlackBerry was a smartphone, most people would disagree.
Perhaps the PalmPDA/Phone combinations would count.
Re: (Score:2)
BlackBerry was even a full business solution. But BlackBerries are not the only smartphones that came before the iPhone.
https://www.textline.com/blog/... [textline.com]
Re: (Score:2)
The point is: a BlackBerry is not considered a smart phone. The first iPhone is considered one.
Hence the nitpicking about: Apple did not invent the smart phone, makes no sense.
Blackberrys had keyboards. Not even a touch screen. No real apps. How can one think that is the same as an iPhone or better or a predecessor - is beyond me.
Re: (Score:2)
You're playing the true scotsman game. Many smartphones had keyboards and it's a shame there are none that do anymore. The timeline explicitly lists when the term smartphone was coined.
Re: (Score:2)
Generally speaking, Apple doesn't follow trends, it sets them.
That is bullshit. They've had a few trend setting successes, but they are also very much a follower for most of the rest of the market.
Not following the AI "trend" is in their DNA, and very much to their benefit.
You're attempting to re-write history. Apple didn't "not follow" the AI trend. They *failed* at it. They promised it. They promised it to users, they promised it to investors. There are multiple court cases underway right now about them failing to deliver AI enabled hardware. They had a whole lot in the AI pipeline, from AI Siri, to AI wearable pins, to even AI glasses. Very
Ed Zitron is a hypocrite (Score:2)
His podcast Better Offline is squarely targeted at people who are very much online. Indeed his entire gig is the epitome of online living.
If Ed Zitron really felt better offline, he wouldn't be making a living off of a podcast and constant AI bashing.
Not that I necessarily disagree with him though... It's just that the messenger doesn't really befit the message.
Re: (Score:2)
Re: (Score:2)
Perhaps you should read the article? And try to comprehend it? ...
I see no connection between it and you parents post
That word doesn't mean what you think it means. (Score:3)
So, no painful implants?
The future is becoming... (Score:2)
...increasingly unpredictable
But there seems to be money to be made by predicting doom
Evidently some people enjoy thinking about doom
The best AI strategy for this hype cycle (Score:2)
"No."
Not everything will burn (Score:2)
But from the heat map we can probably map concentrations of stupid.
the Long-game for Apple Vision Pro (Score:2)
AI skeptic, but... (Score:4, Informative)
So as I understand it, current AI models need about two orders of magnitude efficiency improvement to be sustainable as a business. You burst the bubble if usage growth doesn't exceed roughly infrastructure growth divided by efficiency improvement-- flat usage means a 10x growth in infrastructure requires a 10x efficiency improvement. There is some fudge factor to cover sunk costs, but I'm not sure how meaningful that is as you talk about orders of magnitude.
Most of what AI is useful for today could be done more efficiently with small models, and the idea of AGI is a long long time away despite what true believers hold as gospel. That, from my limited understanding is where most of the token spend is today-- the money could easily be starved in that context.
Re: (Score:2)
Now wonder about why they want to ban open models.
Re: AI skeptic, but... (Score:2)
Have you tried small models ? I have. They hallucinate way too much. I had codex run a whole bunch of tests over the weekend to analyze a relative small code base. It automatically downloaded many models from huggingface, with various quants, and tried them all. Few completed within a timeframe that a human would be willing to wait for. We are talking 1 hour timeout. But for the ones that did, the report was mostly incorrect, with the local models hallucinating 5 different flaws that didn't actually exist
Re: (Score:2)
Actually, what happened was that Codex saw you trying out the competition, and intentionally made the others fail!
Re: AI skeptic, but... (Score:2)
Lol. I wish. It matches with my own findings, unfortunately.
Local LLM is good enough as Home assistant conversation agent, though.
Re: (Score:2)
Qwen 3.6 35B-A3B and Qwen 3.6 27B Coder were the largest 2 I tried to run. Neither really "runs" on my 3060 Ti / 8 GB or GTX 1660 Super. Crawl is more like it. But even when waiting a long time for jobs to complete, the result on the codebase analysis was worse than useless - only hallucinated findings.
For other simpler tasks, I have seen Qwen write some useful code. But it often rewrites unwanted code as well, and breaks that one in the process.
Zitron (Score:2)
I have a like/not-like opinion of this guy. His seemingly adderall-induced patter is fun to listen to, and I sort of agree with him on many points, but it is basically an auto-repeat loop now. Not much changes. I suppose that’s true of many slow-motion train wrecks.
Likely-erroneous assumptions (Score:5, Interesting)
This whole article is based on the suppositions that (a) the AI bubble is going to burst and (b) it's going to take down Apple's competitors. I don't think that's a likely sequence of events.
I do think we're in something of an AI bubble, and that it will burst... but it won't burst in the sense of "All this AI stuff will go away and no one will need data centers", it will burst in the same sense that the dotcom bubble burst, or the railroad bubble burst. What happened in those cases wasn't anything remotely like "and the hyped technology became unimportant".
I think the railroad analogy is the most interesting because while a lot of railroads went under and a lot of railroad investors lost their shirts, it wasn't because railroads -- especially trans-continental railroads -- were just as huge a game-changer as everyone thought they were, it was just because the capital investment required was massive and the payback didn't come as fast as the financial structures built to fund the construction required. What happened was that those with the money snapped up the failing railroad assets and made a lot of money as the massive economic growth unleashed by the railroads was realized, over the course of a few decades.
Similarly, the dotcom bust famously left a lot of "dark fiber" laying around... but all of that fiber got lit up within a handful of years, plus we've added a ton more.
AI is looking like it will follow a similar path. It absolutely is a revolutionary technology, it will upend the structure of our economy -- even more than railroads did -- and there is a lot of money to be made. And even if AI gets much more efficient and doesn't need all of those data centers, they're still going to get used. The problem is just that some of the early investors pushed so hard trying to out-grow the competition that they've built some unsustainable financial structures, and there's a good chance that AI won't generate payback fast enough to keep those cards from tumbling down.
This probably won't be because AI doesn't advance fast enough, but because it takes time to figure out how to integrate any new technology, to make the business and even social adaptations to make use of it. I suspect those things will happen far faster then they ever have for such a large shift, because AI will actually help to accelerate them, but they still probably won't happen fast enough.
OpenAI has the most exposure. Anthropic has done a better job of figuring out how to monetize AI and is a little less exposed, though still quite exposed. I don't think Google, Amazon or Microsoft have significant exposure, mostly because they all have very large non-AI revenue streams and large piles of cash, both of which can help them survive heavy investment. For example, although Google did post its first quarterly net loss in forever, thanks to heavy data center investment, it was only a $6B loss, and for a company that would otherwise be generating $80B per quarter in net profits and has a $130B cash warchest, that's nothing.
As for Apple? Well, Apple might be able to sit cagily on the sidelines, waiting for the crash so it can use its pile of cash to snap up the assets and buy in. But it's also very possible that they're going to miss the boat. Not completely, of course, but somewhat like Microsoft missed the boat on the Internet. I do agree that really making something like the Vision Pro work really well would be huge.
Re: (Score:2)
>the AI bubble is going to burst and
Kimi K3 docs and tooling, and weights, dropped this morning. Kimi is selling K3 tokens at $3/million tokens and if you're running it using GB300 gpu racks @ $6 million each... the math mostly checks out. Anthropic is selling what some people call roughly equivalent Fable tokens at $50/million each. I don't think AI is going anywhere, but valuation of companies like Anthropic, OpenAI in the $500b-$1T range isn't sustainable when "open" models sell tokens at a pr
Re: (Score:2)
And why does the "AI bubble" burst, when Moonshot is making good money with their model? Maybe Anthropic bursts ... but that doesn't mean that AI itself will have a problem. It looks like these labs don't have much to fear even when their own model is available for free. Giving the model to everyone who wants to provide a competing service is a clear sign that you're convinced your service is fair and the best offer. That's way more convincing than saying "You don't have an alternative anyway" like the two
Re: (Score:2)
Catching up is great but there will still be a premium for higher and highest quality outputs for a long time because the bar where we all agree "this is smart enough, I do not need more intelligence" is really fucking high and will take us a very long time to get there.
Another threshold is when you can run an acceptable level of output from a local system that doesn't cost an arm and a leg. But then why would you not back that up with the highest quality remote model, pay the subscriptions but not rely 100
Re:Likely-erroneous assumptions (Score:4, Interesting)
But it's also very possible that they're going to miss the boat. Not completely, of course, but somewhat like Microsoft missed the boat on the Internet.
So...Microsoft certainly didn't get the massive market share AOL had in the 90's, and they weren't too successful at getting the "walled garden" they were probably hoping for around that time...but they did just fine. Azure isn't *quite* AWS, but they're doing better than Google in IaaS revenue, and they spent most of the 00's and 2010's running the business world with desktop Windows and Windows Server - alongside Linux and not at super massive scale, granted, but they did just fine.
Where I think Apple's opportunity comes in is to replay the iPod playbook. There were a bunch of MP3 players before the iPod, and most of them had trouble once Napster fell. Apple ran the 2000's on the iPod not just because of the hardware, but because of iTunes and the iTunes Music Store, making it possible to use both existing MP3 libraries and to create new ones, easily, legally, and virus-free.
I think that playbook can work here as well. I've got two computers with GPUs, one an RTX3060 and one an AMD 9060XT. Not flagships by any means, but enough to run local models well-enough. My Macbook issued to me by work has an M5 CPU, and it generates tokens within 20% as quickly as those cards, on battery. If Apple can conjure up some sort of AI App Store that serves up AI models and purpose-built apps for them, that run on Apple hardware with no subscription tied to them, and make the process smooth, seamless, and generally-useful, Apple can probably stand on OpenAI's shoulders and make a local AI laptop that's worth owning...and if they can do that, they can profit at the endpoint level without having to invest billions at the datacenter level like OpenAI and Anthropic are doing.
The Vision Pro, I think, would work well if they sold it as a business tool - diagram overlays, status windows, that sort of thing...but fashion accessories that obscure the face are VERY difficult to make work; people wear contact lenses because they prefer those to face-obscuring glasses that correct their vision. As a video game console or a work utility that's put on for a function and then turned off, there's a niche, but it's not going to be the sort of thing people just walk around wearing.
Re: (Score:2)
But it's also very possible that they're going to miss the boat. Not completely, of course, but somewhat like Microsoft missed the boat on the Internet.
So...Microsoft certainly didn't get the massive market share AOL had in the 90's, and they weren't too successful at getting the "walled garden" they were probably hoping for around that time...but they did just fine.
No, they didn't. You're judging based on a successful recovery they made in the subsequent 30 years. They really missed the boat, and lost the lock they once had on computing in general. Those who didn't live through it don't realize how Windows was synonymous with "computing". MS had the only real consumer computing platform (Apple existed but it was noise, relegated to a particular niche and lots of people with Macs had to have a PC, also). The web fundamentally changed that.
Re: (Score:2)
" it was just because the capital investment required was massive and the payback didn't come as fast as the financial structures built to fund the construction required."
The real problem was that much of the financial structure was bogus, propped up by government shenanigans around the financial instruments rather than addressing the industry that the instruments supposedly represented. The required payback was never going to be possible.
The *useful* assets were snapped up, but a lot of it was garbage.
Will bots buy Apple products? (Score:3)
Unwarranted skepticism (Score:2, Interesting)
That guy is a doom monger who makes his living bashing AI and I don't believe a word he says.
We're still in early days with these LLM's and AI in general. There's huge potential and very large demand for it, that's the reason for the infrastructure boom. The companies that could benefit the most from AI haven't had time to properly integrate it, so naturally there are growing pains. Anthropic and OpenAI are themselves just fledgling companies trying to grapple with success.
Meanwhile, token prices have falle
GPUs for everyone!!! (Score:2)
Get ready, we can finally afford RAM and GPUs again!
LLM results getting much worse (Score:2)
News that doesn't matter (Score:2)
I'm trying to understand why I should care about this guy's opinion about a hypothetical future that may or may not happen and what Apple's role might be in it. So far I've been unsuccessful.
Re: (Score:2)
All the overused AI phrases are just "good english" -- including the em-dashes. It only "is AI" because AI uses it waaaay too often.
Re:Microsoft, Google, & Amazon are "too big to (Score:5, Interesting)
TARP was a net positive for the US. The banks that got bailed out paid back their loans with interest, and the stocks the US government took in those banks paid dividends. Eventually the US government took equity in hundreds of banks around the country through direct stock purchases and warrants (think options) of around $245B, and sold them for $275B for a $30B profit. There were several social programs to help individuals that ended up being a net loss that weighed down the overall TARP program, but the government did get paid back and in general made some money on it.
But that was when the US government was spending around 10% of net revenue on debt services and had a debt-to-GDP of around 35%. Today it's around 100% and it spends more on debt servicing than the entire US Defense budget. So a TARP-style protection program could be much more difficult.
But aside from all of this, the government won't want tech to fail. I doubt they really need Anthropic or OpenAI, but they will need Microsoft, Oracle, NVidia, and others for both DoD purposes and export purposes. So most likely there will be some kind of bailout when this all shakes out.