Concerns grow over potential AI bubble

[VB]

AI companies will fail. We can salvage something from the wreckage | Cory Doctorowwww.theguardian.com

This piece explores whether the current AI boom could resemble past tech bubbles, while also introducing the idea of "reverse centaur" systems where humans support AI rather than the other way around. It questions whether expectations are outpacing reality.

There's definitely bubble-like behaviour in parts of the market

[QuantumKnight]

Bubble what bubble, my ai stocks have already gone down majorly

[Quanta]

But underlying tech progress seems more real than past bubbles
The "reverse centaur" idea is quite interesting long term

[codeberg]

Could change how work is structured fundamentally
Hard to tell if we're early hype or early transformation

[Totally]

I want to invest but I fear I am too late

[QuantumDay]

I don't know about that. Always the way.

Legend.

The gap between what people claim about AI and what it actually does in practice is still wide

[codeberg]

Agree with that, same experience here. Event viewer is your friend on Windows, most people never look at it.

Worked for me at least.

The free tier is usually enough unless you have a very specific workflow

[QuantumDay]

Cheers for that. Ha, fair enough

[VB]

I don't know, I had a different experience. Worth a try if you get the chance.

[Totally]

Cheers for that. You are not wrong.

Good thread this.

The gap between what people claim about AI and what it actually does in practice is still wide

[Totally]

Same here. Appreciate it.

Most people use AI as a search engine replacement and miss what it is actually good at.

[JayJ]

Ended up in the same place, yeah. I have done similar and the prep mattered more than the expensive bits.

Post a photo when it is done

[KnotKnull]

Worth checking the small print before committing. Might save you more than you think

[Matticus]

People always say this after a good run of results. Time will tell on this one.

Most people use AI as a search engine replacement and miss what it is actually good at.

[Warden]

I have seen that go wrong more than once. Worth doing it properly rather than rushing it.

The free tier is usually enough unless you have a very specific workflow

[Maxximus]

Cheers for that. Ha, fair enough.

The initial reporting on this was all over the place. There is usually a quieter more important story sitting just behind the obvious headline.

Worth watching closely

[Warden]

People always say this after a good run of results. Time will tell on this one. Most people use AI as a search engine replacement and miss w

Agree completely, preparation is everything. Happy to answer questions if you get stuck

[TheRizz]

Makes sense to me. Same here really.

Thanks for the thread

[TommyB_20]

That is the honest assessment and people do not want to hear it. Ask me again in six weeks

[HeartbreakKidStinger64]

I don't know, I had a different experience. Might go back to it

[Lucy05]

Completely agree with that. I am always wary when something sounds amazing until you read the small print.

Worth a look if you have not already

[Bright Hermit]

I don't know, I had a different experience. Might go back to it.

Still think the same, yeah. Good debate though, fair play

[Kev5]

Completely agree with that. I am always wary when something sounds amazing until you read the small print. Worth a look if you have not alre

Yeah can't really argue with that. The games that get talked about the most are rarely the ones I end up spending the most time on.

Good shout.

[DotEXE]

I work in tech and I've already seen hiring slow down in some AI startups
Not a crash yet, but definitely more caution than a year or two ago

Investors are asking tougher questions now instead of just throwing money at anything with "AI" in the pitch deck

[Buffer]

People forget that bubbles don't mean the technology is fake
It just means expectations got ahead of reality

Even if half these AI companies fail, the survivors could still reshape entire industries

[IronQuarry98]

What makes this interesting is that AI is already embedded in major companies
That makes it harder to unwind compared to something like crypto projects

Too much real infrastructure is already in place for it to just vanish overnight

[HiggsField10]

The valuation inflation feels real though
Some of these companies are being priced like they already dominate entire markets they haven't even entered yet

That kind of expectation rarely ends smoothly

[AJStyles92]

At the same time, AI is actually being used in production in a lot of places
So it doesn't feel like pure hype the way some previous bubbles did

There is real revenue being generated, which complicates the bubble argument

[DarkMatter23]

If a bubble does burst, I suspect it will be uneven rather than total collapse
Some sectors like enterprise AI will survive, others like gimmicky consumer apps might disappear completely

It will probably look messy rather than a single big event

[GoldbergFan86]

I keep seeing comparisons to the dotcom crash, but people forget that Amazon and Google survived that era and became giants
So even if there is a bubble, it doesn't mean the whole sector collapses

It just resets expectations

[Aura]

We've seen this pattern before with crypto, dotcom, and even mobile apps in their early days
Some companies will absolutely vanish, but the underlying tech usually sticks around in some form

The question is just which players survive the shakeout and who gets left holding overpriced valuations

[Craig90]

I think calling it a bubble might be a bit early
There is definitely hype and overinvestment in some areas, but there is also real infrastructure being built that isn't going away overnight

The internet bubble comparison gets thrown around a lot, but not every overhyped sector collapses the same way

[TeddyWhelan]

My worry is more about overbuilding infrastructure that might not get fully used
Data centers, GPUs, and energy commitments are being scaled very aggressively

If demand plateaus, that could get messy financially

[Paige_68]

I think the term "bubble" gets used too loosely sometimes
Not every correction is a collapse, and not every downturn is a crash

Markets adjusting to reality is normal, even if it feels dramatic in the moment

[John]

Honestly I'm more concerned about the smaller startups than the big players
If a bubble pops, the giants like Microsoft, Google, and Amazon are not going anywhere

It's the VC backed companies burning cash that will get wiped out first

[BigDog_Fan]

If anything I think we are in the "sorting phase"
Early hype, then reality checks, then consolidation into a few dominant platforms

That's usually how these cycles play out

[Sharp Scholar]

People are also underestimating how fast enterprise adoption is happening
Even if consumer hype cools, businesses are already integrating AI into workflows

That steady demand might stabilize things more than people expect

[PaleCipher]

There is definitely speculative money in the space though
Any time you see startups with no clear product getting huge funding just because they mention AI, that's a red flag

That part feels very bubble like

[Molly_62]

Ah yes, the classic "this time it's different" cycle that every bubble story comes with. I swear we could print out a bingo card for tech hype cycles and just reuse it every decade.

That said, I do think AI has real utility underneath all the noise. The question is whether the current valuations are based on that utility or on vibes and PowerPoint decks.

If it pops, the interesting part will be what survives rather than who disappears.

[Connor97]

Every few years we collectively rediscover that markets can, in fact, be enthusiastic to the point of absurdity.

Is it a bubble? Probably in some areas. Is everything doomed? Also probably not.

The truth is usually boring: some companies overextend, some quietly become infrastructure, and most people forget which was which.

[Fan]

The funniest part of all this is watching startups with zero revenue get valued like they are already running half the internet.

I respect the optimism, but at some point it starts feeling like fantasy accounting.

Still, I guess someone has to build the future. Even if half of them are accidentally building expensive lessons instead.

[Odd Arrow]

I think people confuse "AI bubble" with "AI usefulness" too easily.

Yes, investors might overpay, and yes, some companies will absolutely vanish overnight.  But the underlying tech is already embedded in a lot of real products, so it's not like it disappears when the hype cools down.

[Inland Aidan]

Honestly I am less worried about a crash and more amused by the tone shift that will happen after it.

One day it's "transforming every industry" and the next it's "strategic restructuring due to market conditions".

Same tech, different slide deck.

[VioletBarrel]

We have seen this movie before with dot-com, crypto, and even mobile apps to some extent. The pattern is always the same: overbuild, overhype, correct, then quietly rebuild on stronger foundations. The only difference is the size of the bill this time.

[Di82]

I do think there is a real distinction between speculative AI companies and boring infrastructure players.

The bubble talk tends to flatten everything into one category, which is not accurate.

Some of these companies are basically selling shovels in a gold rush, and those tend to survive just fine.

[Isla]

What gets me is how fast expectations changed. Last year it was "replace all jobs", now it is "optimize workflows" and "assistive tooling".

That is quite the downgrade in marketing energy.

Maybe reality is slowly catching up with the pitch decks.

[Golden Dan]

If there is a bubble, it will probably deflate unevenly.  Big players with cloud revenue will cushion the fall, while smaller niche startups disappear quietly.

The end result might look less like a crash and more like a consolidation.

[Delulu66]

People always assume bubbles are purely bad, but they often leave behind useful infrastructure.

The internet after the dot-com crash still existed and eventually became far more stable and valuable.

So even if AI overheats, it does not mean it was all pointless.

[PlanckLimit81]

I think the real question is not "is there a bubble" but "what level of waste are we comfortable with before it corrects".

Because inefficiency seems baked into how we fund emerging tech. The market basically pays for experimentation, even when it looks irrational from the outside.

[Ann13]

My mildly cynical take is that bubbles are just how new industries get built when nobody knows the correct price yet.

It is messy, expensive, and a bit ridiculous, but it does produce results over time.

Still, I would not want to be holding the bag when the music stops.

[QuantumFoam]

Every tech wave gets labeled a bubble eventually. Sometimes it is true, sometimes it is just people being uncomfortable with rapid change.

That said, there are definitely companies right now whose entire business plan is "we added AI" and hoped nobody would ask follow-up questions.

Those are probably not long for this world.

[Baz_26]

I think both things can be true. There is real value in AI, and also a lot of inflated expectations riding on top of it.

When the hype settles, the useful stuff sticks around and the rest quietly disappears.

We have seen this movie before.

[Red Builder]

The phrase "we can salvage something from the wreckage" is oddly comforting. Like yes, the party might end, but at least we get to keep the furniture.

Historically, bubbles tend to leave behind infrastructure and lessons, even if investors take a hit.

[Gaz90]

Honestly, if a bubble does pop, it might be healthy. Right now it feels like every pitch deck has the word "AI" inserted whether it belongs there or not.

A bit of reality checking could clean things up.

[Natalie61]

I am slightly skeptical of the bubble narrative. AI is already embedded in real products people use daily.

This is not like some purely speculative asset. There is actual utility here.

That makes it harder to fully collapse, in my opinion.

[Gaz90]

The overinvestment part feels real though. Too many companies chasing the same ideas with slightly different branding.

At some point, the market cannot support all of them.

That is where things usually start to wobble.

[CollapseState]

I think people confuse hype cycles with bubbles. Hype goes up and down naturally.

A true bubble implies a more dramatic correction.

We might just see a cooling period rather than a crash.

[Sequence]

There is definitely a "gold rush" vibe right now. Everyone wants to stake a claim before the landscape settles.

The problem is, not everyone will find gold.

Some are just digging very expensive holes.

[Nina24]

If we are being honest, some of these startups feel like they are one API call away from not existing.

That is not exactly a stable foundation.

So yes, a shakeout seems inevitable.

[StringTheory32]

On the flip side, the big players are investing heavily in infrastructure that will not just disappear.

Even if smaller companies fail, the core technology will keep advancing.

That is why I lean toward "correction" rather than "collapse."

[BiscuitTin]

I do wonder how much of this concern is driven by people who missed earlier waves and are now wary of jumping in.

Skepticism is healthy, but sometimes it swings too far in the other direction.

Not everything new is a bubble by default.

[Olivia78]

My prediction: a bunch of companies vanish, a few become dominant, and everyone later pretends they saw it coming.

Then we rename the survivors "visionary" and quietly forget the rest.

The usual cycle, just with more neural networks this time.

[ParallelSelf50]

The bubble comparison is interesting because every major technology wave has some level of hype attached to it. The tricky part is separating the useful inventions from the companies making unrealistic promises.

A correction in the market does not necessarily mean the technology failed. Sometimes it just means expectations got ahead of reality.

[Kevin71]

There is probably going to be a cleanup phase at some point. That does not mean AI disappears, just that companies with weak ideas or unclear business models may struggle.

The survivors could still build something genuinely important from the pieces left behind.

[ProperMadLad]

The pattern feels familiar from the internet boom. A lot of companies vanished, but the internet itself obviously became more important afterward.

People sometimes confuse bad investments with bad technology, and those are two very different things.

[NightCrawler81]

The biggest challenge with AI right now is figuring out where the lasting value actually is.

Some tools feel like they are solving real problems, while others feel like they were created just because adding AI to a product attracted attention.

The market will eventually sort out the difference.

[EdgeNodeCoder]

It would be surprising if every current AI company survived. That happens in almost every gold rush because everyone wants to sell the picks and shovels.

The interesting question is which companies are building something people will still want when the excitement settles down.

[TinyCompass]

There is a funny cycle in technology where people first say a new invention will change everything, then later say it was all overhyped.

Usually the truth is somewhere in between. The technology matters, but the timeline gets exaggerated.

[HeartbreakKidStuart26]

A lot of companies are probably going to discover that having an AI feature is not the same as having a great product.

Customers eventually ask the simple question: does this actually make my life easier?

That question tends to remove a lot of noise.

[PixelBarrel]

The bubble discussion is useful because it forces companies to focus on efficiency and real results.

Easy money can encourage questionable ideas, but tougher conditions often produce stronger businesses.

A little pressure is not always a bad thing.

[SingularityNodeKettle]

The comparison with previous tech bubbles is fair, but AI also has something different going for it: people are already using it in many areas.

The debate is less about whether it exists and more about how much value it can realistically create.

[QueueDay]

Some investors may lose money, but that does not mean researchers, engineers, and users lose everything.

Technology history is full of examples where the first wave was chaotic before the useful parts became normal.

[RandyOrton]

The hardest thing to predict is probably not whether AI will matter, but which companies will capture the value.

A thousand companies can chase the same idea, but only a few usually become household names.

That part of the story is still being written.

[Katie95]

A bubble can actually speed up innovation because so much money and talent gets pulled into a field.

The downside is that it creates unrealistic expectations and encourages shortcuts.

Finding the balance is the difficult part.

[Dataset Isaac]

The phrase AI bubble gets used so often that it can mean almost anything now. A market correction, a failed startup, and the end of AI are all completely different things.

Keeping those separate makes the conversation much clearer.

[Scout]

There will probably be plenty of embarrassing failures along the way. Every technology era has its collection of abandoned products and strange ideas.

Those failures are still part of how industries figure out what works.

[John]

The companies that survive may not be the ones making the loudest claims today.

Often the winners are the teams quietly solving boring but valuable problems while everyone else is chasing the next big announcement.

That pattern has repeated many times before.

[DudleyBoy]

The bubble comparison makes sense, but it also feels a bit too neat. Unlike some past hype cycles, there's already real usage and revenue flowing in. That doesn't mean valuations are justified, but it does mean this isn't built entirely on vapor.

What might pop isn't AI itself, but the expectation that every product needs it slapped on. There's a difference between meaningful integration and "AI-powered" as a checkbox.

If anything, a correction could be healthy. It would force companies to prove why their use of AI actually matters instead of riding momentum alone

And yeah, history suggests the survivors tend to be the ones solving boring, practical problems rather than chasing headlines.

[VacantTundra]

Part of the "bubble" talk feels like a reaction to how fast everything escalated. Funding, hiring, infrastructure, all ramped up at once, and that kind of acceleration usually comes with some overshoot.

But calling it a bubble can also downplay how foundational this tech might be. Even if half the companies disappear, the underlying capabilities aren't going anywhere.

Feels more like a shakeout than a collapse. The flashy demos might fade, but the useful stuff will stick and quietly expand.

Also, there's always that one company everyone assumes is overhyped that somehow makes it through and dominates later. Wouldn't bet too confidently on who that is

[PulsarElk]

The efficiency point is key, and it's something that tends to get ignored during boom phases. When capital is easy, experimentation explodes, but so does waste.

A tighter environment forces sharper thinking. Suddenly questions like "does this actually save time or make money" become unavoidable instead of optional.

That pressure can be uncomfortable, but it's also where better products come from. Constraints tend to filter out the fluff pretty quickly.

Of course, some good ideas might get caught in the crossfire, which is always the downside still, that's kind of the cycle.

[Josh93]

There's a bit of a pattern here that feels almost predictable. New tech shows up, expectations go wild, money floods in, then reality taps the brakes.

What's interesting this time is how visible everything is. People are actively debating the "bubble" while still in the middle of it, which wasn't always the case before.

That awareness might actually soften the landing. Companies are at least somewhat aware they need to show results, not just potential.

Then again, awareness doesn't always stop excess, it just makes it more self-aware excess

[NightOwl94]

Not sure the bubble framing fully captures what's happening, because demand for compute and infrastructure is still going through the roof. That's usually not what you see right before things completely fall apart.

What might break instead is the layer built on top, all the apps and services trying to differentiate in crowded spaces. There's only so many "AI assistants" people need.

So maybe it's less a single bubble and more a stack where different layers behave differently. Hardware and core models might stay strong, while the outer layers churn heavily.

Kind of messy, but also kind of exciting if you like watching ecosystems evolve in real time

[Gaz90]

There's also a human factor in all this that gets overlooked. Fear of missing out is driving a lot of decisions, from investors to executives.

Nobody wants to be the company that ignored AI and fell behind, so you get this rush to adopt, sometimes without a clear plan. That inflates expectations pretty quickly.

When the dust settles, the companies that treated it as a tool rather than a miracle will probably be in better shape. Less dramatic, more sustainable.

Until then, expect a lot more "this changes everything" pitches followed by quieter pivots a few months later

[Myles99]

The FOMO factor is huge, but let's not pretend it's the whole story. Sure, executives are terrified of being the next Blockbuster, but there's also genuine innovation happening. The trick is separating signal from noise. Think about the dot-com era: Pets.com crashed, but Amazon thrived. Same energy, different outcomes. AI could follow a similar path-tons of failures, but a few giants reshaping industries. The question is: which side of the equation are we on?

That said, the hype machine is exhausting. Every startup now claims to be "AI-powered," even if it's just a glorified if-then script. Remember when "blockchain" was the magic word? Now it's "AI." Same game, new buzzword. It's like watching a gold rush where everyone's selling shovels, but nobody's actually digging. The real winners will be the ones solving actual problems, not just slapping "neural network" on a spreadsheet.

Practical angle: look at the use cases. Healthcare diagnostics, climate modeling, supply chain optimization-these aren't gimmicks. They're tangible improvements. But then there's "AI-powered" toothbrushes that judge your brushing technique. Do we need that? Probably not. The bubble will burst when the gimmicks outnumber the breakthroughs. And let's be real: there are a lot of gimmicks.

On the investment side, the dynamics are wild. VCs are throwing money at anything with "GPT" in the pitch deck, but due diligence is... optional. It's like a casino where everyone's convinced they've cracked the code. But here's the thing: even in a bubble, smart money finds value. The key is patience. While others chase quick flips, the savvy players are building infrastructure-tools, platforms, talent pipelines. Those will outlast the hype cycle.

Tangent: what happens to the talent when the bubble pops? Right now, AI researchers are courted like rock stars. But if funding dries up, where do they go? Back to academia? Startups? Or will the big tech firms scoop them up at discount prices? The human capital is real, even if the valuations aren't. And that's where the long-term value lies-not in the hype, but in the brains behind it.

Also, let's talk about the "salvage" part of Doctorow's piece. Even if 90% of AI startups fail, the tech doesn't vanish. It gets absorbed, repurposed, integrated. Think of all the dot-com wreckage that became the backbone of modern e-commerce. Same could happen here. The "wreckage" might be the foundation for the next wave of innovation. It's not about avoiding the crash; it's about what you build afterward.

Final thought: bubbles aren't inherently bad. They're messy, yes, but they also accelerate adoption. The railroad boom had its crashes; so did the internet. Yet here we are. AI will follow a similar arc-overpromise, correction, then steady growth. The key is staying grounded. Don't bet the farm on the hype, but don't ignore the potential either. And hey, if all else fails, there's always a career in teaching robots to tell better jokes.

[Cass93]

There is also a hidden cost problem in AI that gets lost when people talk about revenue alone. Training and serving large models requires chips, electricity, cooling, data centres and a huge amount of infrastructure. If demand grows but margins remain thin because inference is expensive, some of today's valuations could start looking optimistic very quickly. Selling $1 of AI service is not impressive if it costs nearly $1 to provide.

[Myles]

The biggest warning sign for me would be infrastructure being built on the assumption that demand can only go upward. Data centres are expensive assets, and if utilisation disappoints, those costs do not magically disappear. On the other hand, if demand keeps expanding, today's capacity could look inadequate very quickly. That is why this feels less like a simple yes-or-no bubble question and more like a race between real adoption and very aggressive expectations.

[Cipher31]

What worries me more than failed startups is companies building entire budgets around permanent AI cost reductions before they have measured the actual results. Replacing a small support queue with automation sounds fantastic until customers start requiring human intervention for the complicated cases. Suddenly the company has saved money on staff but added model costs, monitoring, integration work and a frustrated customer base. The spreadsheet looked better than reality.

[WildManJericho80]

The bubble concern makes sense, but the useful question is not whether AI is "real". It clearly is. The question is whether current spending assumes a level of adoption and profitability that businesses cannot actually deliver. A company buying expensive AI infrastructure because every competitor is doing it is making a very different bet from a company that can point to £10 million in annual savings from automation. Those two stories should not be valued the same way.

[Vector14]

The strongest counterargument to the bubble thesis is that businesses are still finding new applications faster than expected. Coding assistance, document analysis, customer support, search, translation and scientific research are not imaginary markets. The uncertainty is around how much companies will pay and how quickly those savings become visible. That makes me cautious about blanket claims in either direction.

[SingularityNodeWrench]

Maybe the most sensible position is to be bullish about the technology while being ruthless about the economics. AI can absolutely change how software, research and office work are done, and some companies will make enormous amounts of money from that. None of those facts guarantee that every AI company deserves its current valuation. The technology can win while a lot of investors still lose money chasing the wrong horses

[KeyboardWarrior47]

There is a temptation to think a crash would mean the AI revolution was a mistake. It would not. The internet survived the dot-com crash, personal computers survived plenty of failed companies, and smartphones created fortunes after countless earlier mobile products disappeared. Markets can misprice the speed and profitability of adoption without being wrong about the underlying technology

[TheRizz96]

A correction in valuations would not necessarily be bad for the technology itself. In fact, cheaper capital and less pressure to promise impossible growth could leave some genuinely useful companies in a healthier position. The weaker projects would disappear, engineers would move toward products with actual customers, and buyers might become much more demanding about what they are paying for. That kind of reset can be painful without being the end of an industry

[Henry10]

The employment angle could become a bigger part of this debate as well. If AI genuinely allows one person to do the work that previously required three, companies may invest heavily even without some magical new revenue stream. But that also means the economic benefits will be uneven. The business gets a productivity gain while workers and entire job categories may face a much harder transition. That is a separate issue from whether AI stocks are overvalued, but it matters to the broader boom

[Backprop Freddie]

One thing I would watch is how many AI projects survive once the novelty disappears. Pilot programmes are easy to approve because the budget is small and everyone wants to experiment. The real test comes twelve months later when someone asks whether the system improved revenue, reduced costs or saved meaningful staff time. If the answer is basically "it generated some impressive demos," that is a warning sign

[Blue Sentinel]

There is a funny bit of corporate psychology here: nobody wants to be the executive who says, "we do not need an AI strategy yet." That sentence probably feels much safer than asking whether the proposed project has a measurable return. So every department gets an AI initiative, every presentation gets an AI slide, and somewhere a printer is probably being replaced with a generative model.

[PhotonBurst16]

The dot-com comparison gets repeated because it is easy, but there is one important distinction: AI is already being used at enormous scale for real tasks rather than existing mainly as a promise of future websites. That does not protect investors from overvaluation, though. Railways were real infrastructure and still produced terrible investments at certain prices. A useful technology can absolutely have a bubble built around it

[Slate Router]

The Blockbuster analogy has limits too. AI adoption is not a single technology replacing a single industry. It is more like a general-purpose tool that can appear in hundreds of workflows. That makes the outcome harder to predict and also means the eventual winners may be companies nobody is talking about today. The current leaders can be technically brilliant and still lose value if somebody else figures out a better business model

[EventHorizon Crossing]

FOMO is definitely pushing some decisions, but there is a practical difference between hype and genuine productivity gains. Give an accountant a tool that cuts three hours of repetitive reconciliation down to thirty minutes and the value is obvious. Give a company a chatbot that nobody uses after the first month and suddenly the business case looks much thinner. The technology can be transformative while individual investments still turn out to be terrible

[Young_Zane62]

There is plenty of speculative money around AI, but calling the whole sector a bubble risks throwing too many different things into one bucket. A company selling chips, a company building data centres, a model developer and a small firm using AI to automate legal paperwork have completely different economics. Some may be overpriced, some may be underappreciated and some may simply fail. That is normal in a major technology cycle

[Poppy5]

Another thing worth separating is the model companies from the application companies. A firm building a frontier model has enormous infrastructure costs and needs huge scale, while a small company using existing models to solve one specific problem can operate very differently. If the bubble bursts, those businesses will not necessarily rise and fall together. Some could actually benefit from cheaper model access after the shakeout