mazz0
macrumors 68040
And hopefully cheaper RAM, just in time for Steam Deck 2
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
If you don't know, that tells me you don't work in any industry directly involved with building AI(s) or taking advantage of AI(s) in their products and services.
AI existed long before LLMs in the form of machine learning/learning engines and will only continue to evolve. My company has been heavily invested in and utilizing said technology for nearly a decade, always expanding its use as the tech advances, ensuring our products and services are cutting edge, industry leading offerings trusted by every Fortune 500 company, various governments, and countless enterprise-level players around the globe.
Last edited:
klasma
macrumors G4
Open-weight SOTA models also don't help if you still need a datacenter to run them.
99% of people think AI is just a chatbot that answers questions. Which is extremely idiotic, there is no bubble. It will continue to evolve and spill into every aspect of this country. The people who think it’s a “bubble” are scared their job won’t exist. And it probably won’t. But none of this is a bubble. This is common sense commentary from someone who’s a computer engineer and uses it daily. The smartest companies in the world aren’t spending 300+ million on a bubble. Believe me or not. But your life is going to change drastically over the next 2-3 years. Adapt and learn. Don’t be stupid and stuck.
There might be a small correction as investment slows to allow the capital to catch up. However, the big push by late 2027 and 2028 will be efficiency, something xAI seems to be prioritizing over directly competing with the flagships, which will alleviate a lot of the very real issues this guy mentions.
Everyone who is calling for or expecting an AI bubble is missing one very important point. It's already too engrained in the average person. Students are learning this in their classrooms. People are using it for everything from research projects to asking relationship advice. It's become a friend, confidant, doctor, assistant, for too many people for it to just go away.
AI will kill Google before it would go away.
The market may change, it may end up being one or two players, it may end up in lots of people running local models, but AI isn't going anywhere.
AI will kill Google before it would go away.
The market may change, it may end up being one or two players, it may end up in lots of people running local models, but AI isn't going anywhere.
Ed is a rare voice of sanity in this AI madness. More like this!
Arislan
macrumors member
One of my biggest fears in this is when we start to see the non-profitability of this, companies will start making cuts to things like guard rails and behavior monitoring. Suddenly, to save cost, the AI is able to do a lot more nefarious things and we all end up paying the price.
I heard many people echo much the same sentiment in 1999 and 2000. Then that bubble burst.
Look, AI has been growing for decades. It has its uses, yes, but to defend the absolute market insanity that's occuring right now as "common sense"? No. This is the same "irrational exuberance" we saw in the late 1990s. What's useful will survive, but have no doubt—this is a bubble.
AssassuN
macrumors regular
A bubble doesn't mean the underlying technology isn't valuable. The internet didn't disappear with the dotcom bubble. It just corrected and shed the dead weight.
obviouslogic
macrumors 6502
It’s beta. Comparing a beta model to well established models that have been developed for years, isn’t going to give any result that any reasonable person could make a judgement call over.
Apple just made the deal with Google 7 months ago. It could very well take a couple of years before they’re able to tweak, refine and distill their foundation models, which again, is why it is still in beta.
I used the old version of Siri all the time, and it was great for what I needed it for, especially while driving; directions, playing music, reading/responding to texts. I’ve been using Siri AI for a couple of weeks now, and aside from a few connection issues here and there, I haven’t really had any problems using it for the same things. And it 1000x better than the old Siri when it comes to personal context - kind of blows my mind the stuff it can dig up.
Yr Blues
macrumors 68040
How much do you have invested in AI? 😆
Not discussed in the interview but what keeps the data centers from being valuable capex is the amount of energy required to run them. When/if the AI bubble bursts nobody will buy the data centers for pennies on the dollar because the cost of running them won’t come down but likely will increase. There is a total mismatch in how long it takes to add electric infrastructure vs. data center infrastructure.
When the dotcom bubble burst, the dark fiber was still valuable at some point. Some of the AI infrastructure will probably be reused after a collapse but it is anybody’s guess how much.
Yr Blues
macrumors 68040
Um, it costs way too much to run. That's the entire point. Not to mention hallucinations.
He is correct that as of 2026, the cost to provide the service is higher than what any customers would pay.
This is because LLMs are a really horrible technology. They predict only one word and then you run them again to get the next word and so on and all you ever get out of them is an imitation of the training data.
But what happens if some advance happens? Maybe Yann LeCun’s world-model ideas pay off or we find some way for "one shot" learning? What I mean is that there is a good chance of finding a way to make the AI both dramatically smarter and 1,000 times cheaper by getting rid of our transformer-based LLMs.
I agree the current AI tech will peak just like steam locomotive technology peaked, but no big deal; we all went to using airplanes. The transportation industry did not burn up and die.
I'm talking not about incremental changes to current AI but something as revolutionary as "multi-headed self-attention" was. Like replacing a train with a jet airplane. It will happen
This is because LLMs are a really horrible technology. They predict only one word and then you run them again to get the next word and so on and all you ever get out of them is an imitation of the training data.
But what happens if some advance happens? Maybe Yann LeCun’s world-model ideas pay off or we find some way for "one shot" learning? What I mean is that there is a good chance of finding a way to make the AI both dramatically smarter and 1,000 times cheaper by getting rid of our transformer-based LLMs.
I agree the current AI tech will peak just like steam locomotive technology peaked, but no big deal; we all went to using airplanes. The transportation industry did not burn up and die.
I'm talking not about incremental changes to current AI but something as revolutionary as "multi-headed self-attention" was. Like replacing a train with a jet airplane. It will happen
AssassuN
macrumors regular
Great point. Yeah can't imagine a low barrier to entry is enticing when the cost to run is still absurd.
This is definitely a bubble predicated on the optimism of investors that AI will scale faster than costs rise.
What an excellent interview.. I have always said Ai is fake, it is based on nothing but hype and paper promises and he just states the same thing.. companies invest in it then realise it is worst then having humans do the work and wonder what was the point? Precisely.. also just WHY are we allowing huge damage to the environment by these Ai companies building countless Au data centres huh?
The bubble WILL burst and it seems like the everyday worker.. as USUAL.. will end up paying the bill whilst those elites at the top will be sitting pretty will be a few more million plus richer.
The entire Ai and LLM concept is flawed.. using tokens as the method and forcing companies to pay per token used! Utterly ludicrous and no doubt a charging system born out of greed and it has dramatically backfired. Good!
The bubble WILL burst and it seems like the everyday worker.. as USUAL.. will end up paying the bill whilst those elites at the top will be sitting pretty will be a few more million plus richer.
The entire Ai and LLM concept is flawed.. using tokens as the method and forcing companies to pay per token used! Utterly ludicrous and no doubt a charging system born out of greed and it has dramatically backfired. Good!
What you should have said is "It currently costs too much to run". Costs will come down. I think there is room for a factor of 1000 cost reduction.
Given enough time, we might have AI based on quantum computing that needs very little energy and we might see cost fall by a million times. But likely not in this century
IJ Reilly
macrumors P6
The scenario he paints reminds me of all the doom-and-gloom comments about the Internet, circa 2000. Masses of investment capital were being shoveled into startups willy-nilly. Most were never going to turn a profit; only nobody knew at the time which ones were DOA or going to thrive. It was mostly guesswork. Do we need to hear examples of both? Hope not.
When the Internet bubble burst 25 years ago, it validated the chicken littles who said "I told you so!"
But what they failed to recognize then is that it's in the very nature of major disruptions that we won't be able to predict winners and losers. It took time to separate them then -- just as it will now.
Then he commits himself to a major contradiction: AI is a sinkhole of cash that will never turn a profit for investors. And yet, because of AI, Oracle is dead meat. So he undercuts his entire argument by essentially admitting that he doesn't know how this disruption will play out. Some will win. A lot of others will lose. But after will be very different than before.
Hey, I could have told you that. Without any numbers.
When the Internet bubble burst 25 years ago, it validated the chicken littles who said "I told you so!"
But what they failed to recognize then is that it's in the very nature of major disruptions that we won't be able to predict winners and losers. It took time to separate them then -- just as it will now.
Then he commits himself to a major contradiction: AI is a sinkhole of cash that will never turn a profit for investors. And yet, because of AI, Oracle is dead meat. So he undercuts his entire argument by essentially admitting that he doesn't know how this disruption will play out. Some will win. A lot of others will lose. But after will be very different than before.
Hey, I could have told you that. Without any numbers.
Not much to this that AI couldn't have written with a basic prompt. But what VisionOS update is he talking about? I'm on the dev betas and have never had to wear the headset for it to update. That's one way to discredit everything you just said.
beermode
macrumors 6502
AI is so fake I already wrote scripts and apps making my job easier. Wow. Tell us more.
So much of that fiber stayed dark for so long. Looks like there isn't a consensus on how much is still dark, but I found a reference to 95,000 "route miles" (miles of bundled fiber cables) that are dark. Who knows? Eventually it may all get used.
I worry about when the data centers are no longer needed—either the bubble bursts, companies go under, and the data centers stop; or the tech simply becomes vastly more efficient, and the tech scales smaller. Then these vast data centers become dead space. I grew up in Detroit—I've seen this happen (look up the Packard plant—it went under in the late 1950s, and demolition only finally finished in 2024). In a decade or two, will we need these data centers?