the future
macrumors 68040
No, it is not. It is orders of magnitude bigger. And so will the consequences be if this bubble bursts.
Last edited:
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple Will 'Watch Everything Burn' When AI Bubble Bursts - Ed Zitron
- Thread starter MacRumors
- Start date
- Sort by reaction score
No, it is not. It is orders of magnitude bigger. And so will the consequences be if this bubble bursts.
OpenAI most likely will go belly up - that’s a no brainer, but LLMs are here forever in one form or another
theapplehead
macrumors 6502a
Unfortunately Ed's explanation and understanding of this seems to be a bit naive and not based in the reality of what the market is actually doing. That's not to say that there won't be a correction, but I would not say that "AI is a bubble".
First, nearly all demand for the parts that primarily drive these hyperscaler's systems are accounted for in 2027 so at the very least, this won't even be a thing until 2028 at the soonest.
Second, while tokens are expensive to use today from Anthropic and OpenAI, they will be forced to lower their rates as newer and more affordable options like Moonshot Kimi k3 take the stage at a fraction of the price, making AI access more affordable for corporations and even individuals that can use these models but just don't want to break the bank. Not only are these models more affordable, but they're also built with fewer high cost parts lowering the strain on the market as a whole. And sure, there's a chance that there is an import/export ban on these more affordable Chinese models, but even still, developers will find ways to imitate the cheaper models they created and ultimately mimic them in the US. This in turn will help ease the strain for all companies over the next couple years as Anthropic and OpenAI are forced to actually compete by offering more affordable token costs as individuals migrate to options such as Kimi.
Lastly, like it or not, I don't see AI going anywhere. There are trillions of practical applications for it in everyday life, many of them uses that we don't even realize yet and will subsequently generate many future jobs/opportunities for individuals as people realize how it can be used for problem solving new and unique issues. Sure, AI in it's current state is unsustainable with the lack of profit and massive cost of investment/high cost of entry, but that's the cool thing about the world. People innovate, find cheaper ways to do things, and then force the competition to do the same or leave the market.
So in my opinion, it is not so much as a bubble as it is a process (like anything) where more refinements will be made as ineffeciences are rooted out and improved upon.
First, nearly all demand for the parts that primarily drive these hyperscaler's systems are accounted for in 2027 so at the very least, this won't even be a thing until 2028 at the soonest.
Second, while tokens are expensive to use today from Anthropic and OpenAI, they will be forced to lower their rates as newer and more affordable options like Moonshot Kimi k3 take the stage at a fraction of the price, making AI access more affordable for corporations and even individuals that can use these models but just don't want to break the bank. Not only are these models more affordable, but they're also built with fewer high cost parts lowering the strain on the market as a whole. And sure, there's a chance that there is an import/export ban on these more affordable Chinese models, but even still, developers will find ways to imitate the cheaper models they created and ultimately mimic them in the US. This in turn will help ease the strain for all companies over the next couple years as Anthropic and OpenAI are forced to actually compete by offering more affordable token costs as individuals migrate to options such as Kimi.
Lastly, like it or not, I don't see AI going anywhere. There are trillions of practical applications for it in everyday life, many of them uses that we don't even realize yet and will subsequently generate many future jobs/opportunities for individuals as people realize how it can be used for problem solving new and unique issues. Sure, AI in it's current state is unsustainable with the lack of profit and massive cost of investment/high cost of entry, but that's the cool thing about the world. People innovate, find cheaper ways to do things, and then force the competition to do the same or leave the market.
So in my opinion, it is not so much as a bubble as it is a process (like anything) where more refinements will be made as ineffeciences are rooted out and improved upon.
That just helps to blow a hole in OpenAI’s bottom line. They don’t get paid for on-device. They will likely continue to tell folks that they REALLLY need to use their big honking servers.
I think a big part of what’s been written here is saying that OpenAI aren’t profitable and there’s not really an upside where they eventually become profitable due to things like on-device AI and others.
The consumer market is already gone but Meta, OpenAI and xAI won't admit it. The new Siri and Gemini on Android will effectively wipe out the consumer AI market as it gets into more hands. xAI, Meta and OpenAI don't have context about our daily lives like the phones of people in the android and apple ecosystems. Without it, a chatbot is just a chatbot. Thats what Meta and OpenAI will always be, regardless of whatever slop Jonny Ive and Alan Dye can crap out. If people want to use AI in their daily lives they will always default to the path of least resistance. That's a free (currently) model on your device that can give you contextual answers.
how about self-driving cars, robots, breakthrough in medicines? I think they are genuinely important as population in first world countries age.
IIGS User
macrumors 68000
AI is here to stay. No doubt about that. This whole thing has the same smell as the Dot.com bust and the Telcom bust of the late 90's early 2000s though. The tech is useful. The product has a market. The resulting infrastructure is tangible and will have utility after a majority of the people who put money into it lost their collective shirts.
But it's never going to generate the required profit to make ROI justify it. Lots of people either individually or via institutional investments are going to take a mega hit when it comes crashing down. There's no way it comes to a screeching halt without a recession. Which to be honest we are overdue for.
I think Apple either dropped the ball or fell asleep at the switch on AI. It doesn't seem to me like someone at Apple sat there and said "AI is going to be a bust so we won't put any effort into it and let it crash and burn around us while we have billions in cash on hand to scoop up the remains at pennies on the dollar".
But it does seem like that's going to be the end result. Sure, Apple is paying Google for the Gemini model but given how much Google pays Apple on a yearly basis it's kinda like renting it for free. Apple is just "keeping up with the Joneses" until the whole thing crashes and burns.
Of all the tech companies wrapped up in AI either directly or adjacently, Apple is one of the best positioned to weather the gathering storm.
But it's never going to generate the required profit to make ROI justify it. Lots of people either individually or via institutional investments are going to take a mega hit when it comes crashing down. There's no way it comes to a screeching halt without a recession. Which to be honest we are overdue for.
I think Apple either dropped the ball or fell asleep at the switch on AI. It doesn't seem to me like someone at Apple sat there and said "AI is going to be a bust so we won't put any effort into it and let it crash and burn around us while we have billions in cash on hand to scoop up the remains at pennies on the dollar".
But it does seem like that's going to be the end result. Sure, Apple is paying Google for the Gemini model but given how much Google pays Apple on a yearly basis it's kinda like renting it for free. Apple is just "keeping up with the Joneses" until the whole thing crashes and burns.
Of all the tech companies wrapped up in AI either directly or adjacently, Apple is one of the best positioned to weather the gathering storm.
needsomecoffee
macrumors 6502a
Bumped my recent MBP order up to 32 gig (which I do not need) from 16 because I figure at some point I will be running a local AI model. If local models eventually can handle 90% of inquiries(casual use) with the other 10% going to data centers (too hard for local), and we are paying $20 to rent a local model, then it makes a lot of sense. Having a couple billion users sending 20+ casual AI prompts per hour to a datacenter... yeah then Ed's gonna be right.
Zitron's schtick is to be hyperbolic. I wish he wasn't like that because I think a lot of what he points out is important to understand. I think AI is here to stay and will continue to improve, but I also think the business of AI in its current form won't last. At some point the Capex will collapse under its own weight, and new companies will emerge from the ashes.
It's also hard to know where the value of AI will be captured. It could end up like airlines, capital-intensive businesses that facilitate an enormous amount of GDP, but only capture a small amount of that value. The push toward open-weight models running on prem may be a glimpse into this future.
Plot twist, The interview was conducted with an LLM, not with the real Ed Zitron.
Just kidding 🫠. I liked the content. LLMs are real and have very good use cases. But as others have mentioned, the math of the current build out of data center just does not add up. It's going to be painful.
Just kidding 🫠. I liked the content. LLMs are real and have very good use cases. But as others have mentioned, the math of the current build out of data center just does not add up. It's going to be painful.
nikhsub1
macrumors 68030
This article seems to just be a bearish article on AI. Not even a very well-substantiated one at that. Token cost is not nearly as high as the author implies. For one, the companies using huge numbers of tokens are using them for highly complex software engineering tasks that can run for hours at a time or in some cases with dozens of sub-agents. That's a very different amount of computation than "help me plan my summer vacation" or some more typical AI task. Another, is that cost per-token continues to go down over time. Each new hardware architecture is driving 10x or more improvement in token cost, and there are a lot of additional improvements coming in the software stack, etc. The usefulness of AI continues to expand as there is better integration with more services.
The whole notion of "an LLM is an LLM is an LLM" is ridiculous too. Maybe some day that will be true for the bulk of tasks, e.g. even when the "worst" LLM is capable of handling day-to-day queries. Right now however there is significant benefit in being at the frontier, especially for complex tasks.
The whole notion of "an LLM is an LLM is an LLM" is ridiculous too. Maybe some day that will be true for the bulk of tasks, e.g. even when the "worst" LLM is capable of handling day-to-day queries. Right now however there is significant benefit in being at the frontier, especially for complex tasks.
kirbyrun
macrumors 6502
I *think* what he’s saying is that his Vision Pro needs an update, but in order to apply the update, he has to wear the headset for the length of the update process…which he’s not willing to do. Therefore, his VP isn’t usable to him right now.’’
In other words, “I want to start the update and then come back to it later when it’s done, like I can with my other gadgets.”
But I don’t own a VP, so no idea if that’s how the update process works. That’s how his statement reads to me, though. 🤷♂️
Agreed! Reminds me a bit of when Microsoft and Intel watched all the companies "wasting" money on mobile!
Nobody knows. It's ironic that some people are worried about the west/US losing some perceived lead in "AI" while the current policy has been to pay companies to halt energy production projects like off-coast wind turbines. Yet, other major world powers like China are investing heavily in reducing their dependence on non-renewable and non-domestic energy in tandem with whatever AI investments they're making.
Meanwhile, we're literally paying companies our tax dollars to not build something that would generate cheap electricity for us and supplement our strained, aging electric grid; all while simultaneously constructing some of the most power-hungry data centers en mass across the continent to support anticipated AI workloads that nobody is actually willing to pay for.
The foreign psyop isn't getting us to abandon wasting our energy and money on AI. It's getting us to self-sabotage by abandoning cheap, renewable, domestic energy solutions and ignoring practical advancements in technology like autonomous, electric vehicles and batteries while increasing our dependence on oil and foreign energy sources, and building a huge AI boondoggle to swallow up whatever electricity we manage to generate so that grandma can make an AI-generated cartoon birthday cake.
MyDogPia
macrumors newbie
Well, he's not 100% correct! I am now able to generate massive amount of Matlab code via Claude that I wouldn't be able to do otherwise. I am now able to build models I wouldn't have been able to do before. I suck at coding. Now an unintentional consequence is I have two other people who spent money to buy Matlab toolbox licenses so they also can create code in Matlab to build models. So, what I am saying, money is flowing to other places besides just Claude / ChatGPT/Gemini.
BarredOwl
macrumors 6502
Yes, this was the best article I’ve seen on this site in ages. Thank you!
Boy, you took it personally. Sorry you made those investment choices.
xAI will only persist because of Musk's own weird view of his self-worth. Eventually he will run out of ways to play shell games, the podcasters will quit calling, all the debt (both financial and goodwill) will come due, and someone (probably plural; I'd guess a Saudi-led consortium right now) will purchase SpaceX and shut down Twitter and xAI.
Timpetus
macrumors 6502a
Sure, there's no difference between "AI" and Bitcoin. Just ignore the fact that one takes energy and turns it into useless garbage, while the other turns it into durable value and a useful hard money without the historical disadvantages of such.
spartan1967
macrumors 6502a
Hurry up already so we can get graphics cards pricing back to where it should be and put them back in the hands of gamers.