Meanwhile in Australia we have a "AI data centre" startup being valued at $1.9 billion and given $330 million to play with, having not built anything yet. It's co-founded by a guy that went to prison for insider trading. His wife is also an investor, who happens to be a prominent Australian influencer. The company previously focused on Bitcoin mining but have pivoted to the AI boom, claiming their cooling systems to be 60% better than competitors. Their first project will kick off soon in the Australian state of Tasmania, where there is nothing much more than sheep and tourists.
Signs of a bubble in this sector are everywhere. OpenAI is trying to fundraise a $5tn infrastructure buildout on $15bn in annual revenue. That's insane.
What's worse is that OpenAI and the other AI companies are all intertwined. The chipmakers are invested in the datacenter operators are invested in the software guys. When the bubble implodes - and it will implode - the good will go down with the bad, and that's what makes a financial crash a true crash.
That and the 3-way "OpenAI pays Oracle $300B in cloud computing on their $15B of revenue", "Oracle buys $40B of NVidia chips for their new datacenter", "NVidia invests $100B in OpenAI". The money's just moving around and around, propping up revenue numbers for Wall Street, but the consumer benefits are dubious.
We do, because they've already done that stuff. You're talking about whether any of the 3 people will be able to get more money in afterwards, which seems like a separate issue to this.
It's more like you offering to invest $1bn in Fred's company in return for a promise to spend $1bn for sink fixing services. You could question if there is a demand or $1bn of sink fixing in the same way that you could wonder about the demand for $100bn of AI slop.
Investing is just an equivalent value exchange. My point is that while you could say in my example that money just moved around in a circle, it turns out that money moving around in a circle is actually fine.
I agree money in a circle is fine. But in your example there is a straightforward situation of an end customer paying money for a service, eg. getting a sink fixed. The worry with the data centers is will there be enough end customers willing to pay what the data centers cost.
Sure - but nothing driving up revenue value is actually being created[1]. What's missing in that system is a way that money is entering the system. These deals are (in my cynical opinion at least) being inked to create the appearance of large continued investment and market excitement to pump or sustain valuations. Oracle actually spotlit the arrangement as future sales in their recent earnings and that seemed to be what mostly drove their valuation up.
Performative actions to drive up valuation and try and attract more investors absolutely feels bubbly to me.
1. Discounting products that are not only currently operating at a loss but are priced well below actual resourcing required to produce.
Customers pay for the compute. There are tons of CSPs selling the capacity to small and large consuming entities alike (both OpenAI/Anthropic + other small outfits we've not heard of).
The fair criticism of the infra $ is where the non-VC non-bank-loan cash stream is, but there could be a lot of B2B deals and e.g. Meta, TikTok and other behemoths do tend to make plenty of money and pay their bills, and have extreme thirst for more AI capacity.
Take Oracle for example (as a whole, not just OCI) - tons of customers who are paying for AI-enhanced enterprise products.
It's still the early days, as the cost of creating software continues to approach zero the rules will change in ways which are hard to predict. The effect this will have on other white collar industries is even more challenging to reason about.
That line of reasoning conveniently left out the explosive datacenter revenue growth that generates huge free cash flows. Even disregarding AI, it's double-digit compounding growth.
NVIDIA's stock may eventually get decimated (but the company itself will be fine, they have a relatively low employee account and insane margins), the Coreweaves of the world are definitely leveraged plays on compute and may indeed end up being DotCom style busts, but a key difference is that the driving forces at the very top - the Microsofts and Amazons of the world - have huge free cash flows, real compute demand growth beyond the AI space, and fortress balance sheets.
I think that's a fair point and sort of speaks to one of the indicators that say this possible bubble may be different than the dotcom bubble. I think that end-user revenue for AI is a pipe-dream - but the companies interested in compute have a whole lot of resources and so long as they are willing to divert those resources to prop up AI it can keep going for quite a while (at a smaller scale though).
There is a commonly held belief that there is a level of compute (vaguely referred to as AGI) that would be extremely valuable and those companies may continue to rationally fund AI research as R&D though if the VC and loan funding dries up there will probably be serious fights with the accounting departments. It is good to point out that companies with huge war chests do seem poised to continue investing in this even if VC/etc dries up due to the lack of end-user profitability - it'll be an interesting shift but probably not as disastrous as the dot-com bubble burst was.
>What's missing in that system is a way that money is entering the system.
Or maybe not enough money soon enough, and at this scale that could be more of a disaster than it had to be.
So far it's not looking like a business boom much at all compared to the massive investment boom which is undeniable, and that's where a good amount of remaining prosperity is emanating from.
If you were a financial person wouldn't you figure there were a lot bigger bonuses by getting involved with the amount of cash flow being invested rather than the amount resulting from profits being made in AI right now?
And how much of the money even exists. I remember earlier in the year Altman saying they were spending $500bn on the Stargate project and Must saying he had less than $10bn in actual funds.
It wouldn't surprise me if much of the $1tn most doesn't turn up and the bubble bursts before 1/10th or that becomes real.
There's nVidia that we know (primarily graphics cards) and more like an investment firm "nVidia" these days. The stocks have grown so much that they are trying to turn their fortunes around (to sustain growth) by investing everywhere.
nVidia has invested in so many companies in the ecosystem and beyond.
What about the NVidia/OpenAI deal? Such a giant investment in a huge customer looks a heck of a lot like "circular dealing". That is, invest $100 million so your biggest customer can spend a ton of that on your own chips. You get to report skyrocketing revenue but that revenue was bought with your own money.
If the AMD/OpenAI deal means that OpenAI will put serious work into making AMD GPUs finally be more amenable to general purpose computing, it's actually a huge win for AMD. None of the major numerical computing frameworks (PyTorch, TensorFlow, JAX, etc.) work as well on AMD GPUs versus NVIDIA GPUs, and that's really holding AMD back from making inroads into the machine learning space.
If this means that compute is considerably cheaper for OpenAI, it's a win for them too. But that remains to be seen.
They should hurry up and build the DCs at double-time then! The last tech crash left us with lots of dark fiber that enabled really cool applications and very low costs.
I can only imagine what could grow out of an oversupply of rack-space and electrical power generation, post-crash.
Electrical power and maybe rack space, but the glass strands of the dark fiber were still perfectly good a number of years later, once demand catches up. Buildings and such still need continuous maintenance. In addition, the actual computational equipment (such as the GPUs that have given nvidia the highest market cap in the world) will be totally worthless.
Tasmania doesn't do much sheep farming. They are more into salmon. Maybe you were making a derogatory reference to the local human population? Fair enough.
Please keep arguing about this, it is relieving a lot of stress to watch this. There have to be more facts than just that. I'm serious. This helps me take focus off the rest of the world burning. I'm serious.
oooh look who's punching down now haha yes sir, I have an extremely unfortunate life, I'm the sort of individual who does not have good luck. I try not to bring it up because its simply not on topic for hn, me circling the drain is really nobody's business but mine and the drain, oh yeah, yours too I guess. questions?
improved literacy rates (https://mississippitoday.org/2025/01/24/jim-barksdale-100-mi...), misspent welfare money with brett farve involved (old news), state auditor doxing private individuals for their political speech, new jesmyn ward novel came out a couple years ago, mississippi filmmaker al warren released a funny movie "dogleg", the band mspaint had a good album a couple years ago, etc
data centers being built but people fighting it. new rezoning in taylor, MS being fought by locals bc they are trying to reclassify agriculture land for heavy industry so they can build an asphalt plant. federal government troops in nearby memphis.
When you come in dead last (or in the bottom 5 states) in virtually every possible positive metric, people are going to make negative assumptions. I understand the historical reasons for that, but it doesn’t change what it is.
fair enough, but their assumption wasn't even about that. a better joke on that quote would've been about vermont.
i hear people online punch down on mississippi all the time, and often they don't know anything about the state except whatever metric they've heard about from a headline. the rest of america isn't very far behind, and if you think the state of mississippi isn't a product of america as a whole then you are extremely mistaken. without the industrialized north you have no plantation economy and without the civil war you have no "dead last (or in the bottom 5 states) in virtually every possible positive metric."
i grew up there, attended public school all the way through my BA, and then spent significant time as a young adult there. based on the stereotypical assumptions, it might be shocking to the big brains on hacker news that somebody from mississippi is an audience of this website.
> i hear people online punch down on mississippi all the time, and often they don't know anything about the state except whatever metric they've heard about from a headline. the rest of america isn't very far behind, and if you think the state of mississippi isn't a product of america as a whole then you are extremely mistaken.
Agreed, people usually just say ‘lol MS is full of idiots, they’re bad at school’ instead of taking the time to understand why. It was more isolated than GA and LA (and AL), there was a higher ratio of slaves to freedmen in the antebellum period, the Delta was undeveloped so lots of impoverished people from across the south moved there to try and develop the land, to name a few reasons.
> i grew up there, attended public school all the way through my BA, and then spent significant time as a young adult there. based on the stereotypical assumptions, it might be shocking to somebody on hacker news that somebody on a similar enough intelligence level to be an audience of this website is from mississippi.
I can say that I don’t assume everyone from Mississippi is stupid, but the generalization about Mississippi that you related seems to be more common than it should be. I think a lot of it has to do with a lack of exposure to people from Mississippi or Mississippi itself.
Thanks for taking the time to respond, I appreciate the discussion.
There was a recent post at the top of HN done by someone who went to USM in Hattiesburg who lives in New Orleans now (and which inspired a show HN https://news.ycombinator.com/item?id=45183050 )
Isn't diversity a positive metric? Mississippi has the highest black percentage of all the states. If one wants to use statistics, use all of the statistics.
Is there a lot of power generation in Tasmania? That is the main criteria for AI data centers from what I heard. Latency is less critical than cost of power.
There's pretty national parks, convict history and some pretty tasty whiskeys and wines, but not a whole lot of reasons to build a data centre. The power argument isn't particularly compelling either, because it's much more sunny on the Australian mainland.
> The power argument isn't particularly compelling either, because it's much more sunny on the Australian mainland.
But the vast majority of Tasmania's power is hydroelectric. Hydro is a much more desirable renewable than solar because it essentially is its own built in battery.
Might need to fact check yourself there mate. Tasmania has issues with enough power for some industries. There was a factory that could not get enough power to a planned upgrade.
God knows how a datacentre would do down there....
Regardless of a bubble or not, don't some operators do exceptionally well financially, without a lot of recognition, when things are murky compared to transparent or having clarity?
And how long has this kind of thing probably been going on?
So the real risk with that data center is the fact that Tasmania is a culdesac on the internet. We have three fiber cables coming to the island, all from Victoria (directly to the north). The whole state has lost internet before due to someone in Victoria digging a trench through the cable (yes a silly mistake but we're all humans).
It's not a bad idea to put data centers here, but we really need a few more links out to the world from here.
However Zitron seems to have forgotten that Google exists or makes TPUs. He mentions Google only 10 times in the entire article, always in a minor way.
Any reason to believe Google's unit economics on AI are any different than the other players here?
And Google is an advertising company. Mostly in search, and increasingly dependent on YouTube. Everything else is a net money loser, including Waymo, Gemini etc.
Almost doesn’t matter what Google’s short term unit economics on AI are as long as their shareholders and board are OK with it. Google makes money hand over fist and can afford it. Microsoft is fine, Oracle is fine, TSMC is fine. These other shops are leveraged to the hilt, so they’ll be in a world of hurt if people stop buying, financing, or investing.
Oracle is not fine. They are borrowing money hoping to get paid back by money losing OpenAI propped up by VC funding.
No business is going to run workloads on OCI outside of ones running Oracle. They a
They are a way distance fourth in cloud. I’ve been working in cloud consulting for five years including the first three directly at AWS (Professional Services). No one worried about having talking points about competing against Oracle.
Microsoft, Google and Amazon have both internal products that can benefit from inference and cloud hosting.
> Any reason to believe Google's unit economics on AI are any different than the other players here?
Only when it comes to their TPUs, and sometimes that one thing may just be the difference to push them over the hump.
Per-token cost-wise, TPUs (& specialized processors in general) will beat GPUs every time. The efficiency difference between the 2 types is never to be ignored, & is likely why they can shotgun it everywhere.
> And Google is an advertising company. Mostly in search, and increasingly dependent on YouTube. Everything else is a net money loser, including Waymo, Gemini etc.
1) Each venture should be treated as a (relatively) isolated vertical slice
2) 9 out of 10 times, a venture just doesn't break even. That's just the nature of the business.
You should listen to the latest episode of the Acquired podcast about Google.
Google also has GCP and unlike OpenAI who is dependent on VC funding and Oracle who is borrowing money. Google throws off cash like crazy and self funds its infrastructure which is already better than everyone else’s
In almost all scenarios, a setup with this incentive structure will lead to massive adoption. It's too tempting, and with most jobs / political positions being short term (<5yrs) ones, people optimize for their time in that timeframe, not longer.
Boards will pursue stock buybacks (short term growth, long term may cause trouble if there's a downturn), banks will lend out subprime mortgages (hit your sales numbers in the short term, at the cost of long term risk), etc etc.
This situation is no different. There's money flowing in and there's less red tape since everyone is being pressured to allow it. It might work out in the long term, it might not, but it will 100% benefit those who push it in the short term. People will get promoted for driving a new data center, politicians can promote more jobs being added, everybody wins... for now.
The future economic aspect becomes irrelevant when the short term candy is sweet enough.
IBGYBG. “I’ll be gone, you’ll be gone”. Infamous email sign-off associated with the run-up to the Great Financial Crisis. Used by Wall Streeters asking analysts to inflate credit ratings for undeserving securities, backed by risky mortgages.
Every time I read one of these articles the main issue I have is that it doesn't take into account the huge shortages of compute that are going on all the time. Anthropic and Google especially have been incredibly unreliable, struggling to keep up with demand.
Each of the main providers could easily use 10x the compute tomorrow (albeit arguably inefficiently) by using more thinking for certain tasks.
Now - does that scale to the 10s of GWs of deals OpenAI is doing? Probably not right now, but the bigger issue as the article does point out in fairness is the huge backlog of power availability worldwide.
Finally, AI adoption outside of software engineering is incredibly limited at work. This is going to rapidly change. Even the Excel agent Microsoft has recently launched has the potential to result in hundred fold increases in token consumption per user. I'm also suspect of the AI sell through rate being an indicator that it's not popular for Microsoft. The later versions of M365 copilot (or whatever it is called today) are wildly better than the original ones.
It all sort of reminds me of Apple's goal of getting 1% in cell phone market share, which seemed laughably ambitious at one point - a total stretch goal. Now they are up to 20% and smartphone penetration as a whole is probably close to 90% globally of those that have a phone.
One potential wild card though for the whole market is someone figuring out a very efficient ASIC for inference (maybe with 1.58bit). GPUs are mostly overkill for inference and I would not be surprised if 10-100x efficiency gains could be had on very specialised chips.
the huge demand exists right now because the cost of a token is near zero. and companies have figured out one weird hack to gaining value in the stock market, which is to brag about how many tokens are being crammed into all manner of places that they may or may not belong.
customer value must eventually flow out of those datacenters in the opposite direction to the the energy and capex that are flowing in
do people actually want all this AI? I see studio ghibli portraits, huge amounts of internet spam, workslop... where is the value?
> Each of the main providers could easily use 10x the compute tomorrow (albeit arguably inefficiently) by using more thinking for certain tasks.
That's true for everyone with regard to any resource.
The question is whether the 10x increase in resources results in 10x or more increase in profit.
If it doesn't then it doesn't make sense to pay for the extra resources. For AI right now, the constraint is profit per resource unit, not number or resource units.
"The later versions of M365 copilot (or whatever it is called today) are wildly better than the original ones."
I find AI agents work very poorly within the Microsoft ecosystem. They can generate great HTML documents (because it's an open source format maybe?) but for word documents, the formatting is so poor I'd had to turn it off and just do things manually.
Opposing anecdote: I got consistent performance out of Grok and Qwen (17 providers on Openrouter) throughout the day but Gemini gets slow and dumb at times.
BlackRock did not buy “a” data center, it bought a data center company with 78 data centers. I have no comment on whether or not it was a good deal, but your framing is silly.
Unless the AI industry can figure out how to be profitable soon (to which basically nobody has a clear path to profitability besides maybe ad revenue) it's hard to see this not blowing up in a year or two. The bills for all of this are going to come due eventually and the AI CEOs can only convince investors to keep letting them burn billions for so long.
It's incredible how much money is being lit on fire without anyone having an answer to the question "how will you generate revenue". Meanwhile alternative energy discussions here are met with harsh criticisms of their bottom lines while they have some of the most firm demand of any industry. Society should be prudent with their investments, but only when it's used for R&D into essential stuff and not a bunch of toys for rich psychopaths?
"If we end up misspending a couple of hundred billion dollars, I think that is going to be very unfortunate, obviously. But I actually think the risk is higher on the other side . If you build too slowly and superintelligence is possible in 3 years, but you built it out assuming it is possible in 5 years, then you are out of position on the most important technology."
His assumption is that superinteligence is close, its just a question of whether it is 3 or 5 years!
No, the first sentence makes it clear that he is aware AGI is not a given. His position is that there is a possibility we can reach super intelligence, and given that possibility they want to be on the bleeding edge and are investing accordingly, given that even total failure won’t cripple their business and all their competitors are doing the same.
"If you build too slowly and superintelligence is possible in 3 years, but you built it out assuming it is possible in 5 years, then you are out of position on the most important technology"
This made me think Zuck sees it as a question of when rather than if. I.e its more a question of 3 vs 5 years rather than possible vs non possible.
And yet, imagine it's 100 million jobs, at 100K per job. That's 10 trillion dollars a year, well worth the investment! Except that it won't be 100K per job that AI companies will capture, it'll be 2K. So that's not 10 trillion dollars a year, it's 200 billion. Which is about what a big tech company makes in advertising already.
>> Cheerleaders such as Sam Altman, OpenAI’s boss, argue that the risks of underbuilding are at least as serious as those of overbuilding, because of the long-term economic potential of generative AI.
* questionable demand for AI products make the DC investments risky (makes sense)
* DCs being built in “remote” areas with cheap land may become obsolete/replaced by other DCs making use of said cheap land (questionable; DCs can upgrade cheaper than building a new one)
* financing for DCs used to be from Big Tech but is now spread out among private equity, sovereign wealth funds etc increasing the exposure of the economy to failure of these investments (again, questionable, unless they are being financed by bank loans)
The most salient concern seems to be a lack of demand. I don’t see why that would change in the future.
I can't find the reference now unfortunately, but I read it was about 50:50, so half is the data center land / building / cooling / power generation, and half is in the GPUs themselves.
This is notably very different from the dot-com build out of dark fibre, where digging the holes cost the vast majority of the money, and the fibres and network equipment cost very little in comparison.
Last weekend we had dinner at a friend to celebrate their child's birthday. They work at OCI in a very senior position and were heading out next day to troubleshoot that exact datacenter in Abilene, TX next day, to presumably yell at various people about cabling and schedule and whatnot. They were not looking forward to it because a) TX and b) it's Abilene TX and c) the overall situation.
We spoke quite a bit about this project, circular money and things and the impression I got is that OCI is just grabbing money while the perpetuum mobile is unrolling.
I am looking forward to hearing how the trip went when I see them next.
Oracle Sinks on Report Its Cloud Margins Are Lower Than Expected
Jeran Wittenstein
October 7, 2025 at 6:06 PM GMT+2
Oracle Corp. shares tumbled after a report that the software maker’s profit margin in its cloud computing business is lower than many on Wall Street have been estimating.
While Oracle generated roughly $900 million in revenue from the rental of servers powered by Nvidia Corp. chips during the three months ended in August, the company only managed about $125 million in gross profit, the Information reported, citing internal corporate documents.
Cloud has failed (by becoming more expensive then on prem) so they're buying up everything in a last ditch effort to centralize compute. Really silly since all those GPUs will be obsolete soon.
Cloud was always going to be more expensive than on prem, that wasn't what was being sold. Cloud sells that they are able to give you high availability across multiple regions without having to do an expensive capital investment. Amazon/Microsoft/Google is selling the talented engineering for getting these data centers working around the world at a high level of availability to companies. Most companies like recurring costs compared to capital expenditures. Companies also prefer lower head counts and paying for services. It is BETTER from an MBA pov to pay Amazon to run a Data center since they are experts than for you to run a data center as a finance company.
This is the engineering perspective, not the finance perspective. As an engineer holding an MBA, I've made the argument countless times in BigCo to move from cloud deployments to on-prem. When you're a startup, you often simply don't have the cash on hand to make the capital expenditure to build out datacenter capacity, especially with an uncertain (but hopefully high) expected rate of growth. When you're a BigCo, the script flips; you have plenty of cash on hand and you want to improve overall profitability, which is done by using capital expenditures to reduce operating expenditures, i.e. funding datacenter build-outs to reduce cloud bills.
> experts
Companies can hire experts and can still out-source to colos if they prefer. This is a question of political will and risk analysis.
> Most companies like recurring costs compared to capital expenditures.
Actually big companies prefer capex to opex and cloud pushed them in the opposite direction of what they'd naturally prefer. But the other advantages you cite + hype overruled the liability of switching from onprem capex to opex.
Also, this was remarkably hard to find. Might I suggest if the logged in user is looking over their own comments, eg me viewing https://news.ycombinator.com/threads?id=fragmede, that a flagged dead comment with a moderator responding shouldn't be collapsed by default? If that's intentional then nevermind, but as I am truly grateful for this site and you and DanG's (and everyone else's) hard work, I want to point out that it was hard to find and wouldn't have come across it if I didn't have a program running that hits the Firebase API and notifies me on replies. Someone who didn't have such a program could easily miss mod responses.
https://archive.is/5CY26
Meanwhile in Australia we have a "AI data centre" startup being valued at $1.9 billion and given $330 million to play with, having not built anything yet. It's co-founded by a guy that went to prison for insider trading. His wife is also an investor, who happens to be a prominent Australian influencer. The company previously focused on Bitcoin mining but have pivoted to the AI boom, claiming their cooling systems to be 60% better than competitors. Their first project will kick off soon in the Australian state of Tasmania, where there is nothing much more than sheep and tourists.
https://www.smartcompany.com.au/startupsmart/firmus-raises-3...
Signs of a bubble in this sector are everywhere. OpenAI is trying to fundraise a $5tn infrastructure buildout on $15bn in annual revenue. That's insane.
What's worse is that OpenAI and the other AI companies are all intertwined. The chipmakers are invested in the datacenter operators are invested in the software guys. When the bubble implodes - and it will implode - the good will go down with the bad, and that's what makes a financial crash a true crash.
The AMD/OpenAI deal sounded like the music stopping to me.
That and the 3-way "OpenAI pays Oracle $300B in cloud computing on their $15B of revenue", "Oracle buys $40B of NVidia chips for their new datacenter", "NVidia invests $100B in OpenAI". The money's just moving around and around, propping up revenue numbers for Wall Street, but the consumer benefits are dubious.
https://www.nbcnews.com/business/economy/openai-nvidia-amd-d...
It's like you paying me to fix your sink, me paying Fred to build me a shed, and Fred paying you to fix his wifi.
No, the problem here would be we don't know if anybody actually need wifi, a shed or his sink to be repaired.
We do, because they've already done that stuff. You're talking about whether any of the 3 people will be able to get more money in afterwards, which seems like a separate issue to this.
It's more like you offering to invest $1bn in Fred's company in return for a promise to spend $1bn for sink fixing services. You could question if there is a demand or $1bn of sink fixing in the same way that you could wonder about the demand for $100bn of AI slop.
Investing is just an equivalent value exchange. My point is that while you could say in my example that money just moved around in a circle, it turns out that money moving around in a circle is actually fine.
I agree money in a circle is fine. But in your example there is a straightforward situation of an end customer paying money for a service, eg. getting a sink fixed. The worry with the data centers is will there be enough end customers willing to pay what the data centers cost.
If the money is leaving the system slow enough (via Employee wages and materials for chip manufacturing) I suppose it can go on for quite some time.
Sure - but nothing driving up revenue value is actually being created[1]. What's missing in that system is a way that money is entering the system. These deals are (in my cynical opinion at least) being inked to create the appearance of large continued investment and market excitement to pump or sustain valuations. Oracle actually spotlit the arrangement as future sales in their recent earnings and that seemed to be what mostly drove their valuation up.
Performative actions to drive up valuation and try and attract more investors absolutely feels bubbly to me.
1. Discounting products that are not only currently operating at a loss but are priced well below actual resourcing required to produce.
Customers pay for the compute. There are tons of CSPs selling the capacity to small and large consuming entities alike (both OpenAI/Anthropic + other small outfits we've not heard of).
The fair criticism of the infra $ is where the non-VC non-bank-loan cash stream is, but there could be a lot of B2B deals and e.g. Meta, TikTok and other behemoths do tend to make plenty of money and pay their bills, and have extreme thirst for more AI capacity.
Take Oracle for example (as a whole, not just OCI) - tons of customers who are paying for AI-enhanced enterprise products.
It's still the early days, as the cost of creating software continues to approach zero the rules will change in ways which are hard to predict. The effect this will have on other white collar industries is even more challenging to reason about.
That line of reasoning conveniently left out the explosive datacenter revenue growth that generates huge free cash flows. Even disregarding AI, it's double-digit compounding growth.
NVIDIA's stock may eventually get decimated (but the company itself will be fine, they have a relatively low employee account and insane margins), the Coreweaves of the world are definitely leveraged plays on compute and may indeed end up being DotCom style busts, but a key difference is that the driving forces at the very top - the Microsofts and Amazons of the world - have huge free cash flows, real compute demand growth beyond the AI space, and fortress balance sheets.
I think that's a fair point and sort of speaks to one of the indicators that say this possible bubble may be different than the dotcom bubble. I think that end-user revenue for AI is a pipe-dream - but the companies interested in compute have a whole lot of resources and so long as they are willing to divert those resources to prop up AI it can keep going for quite a while (at a smaller scale though).
There is a commonly held belief that there is a level of compute (vaguely referred to as AGI) that would be extremely valuable and those companies may continue to rationally fund AI research as R&D though if the VC and loan funding dries up there will probably be serious fights with the accounting departments. It is good to point out that companies with huge war chests do seem poised to continue investing in this even if VC/etc dries up due to the lack of end-user profitability - it'll be an interesting shift but probably not as disastrous as the dot-com bubble burst was.
> as the cost of creating software continues to approach zero
"continues " is inaccurate. The cost of creating software is nowhere near approaching zero
In Sam's dreams, perhaps.
>What's missing in that system is a way that money is entering the system.
Or maybe not enough money soon enough, and at this scale that could be more of a disaster than it had to be.
So far it's not looking like a business boom much at all compared to the massive investment boom which is undeniable, and that's where a good amount of remaining prosperity is emanating from.
If you were a financial person wouldn't you figure there were a lot bigger bonuses by getting involved with the amount of cash flow being invested rather than the amount resulting from profits being made in AI right now?
Money leaves the system rather fast in energy costs if nothing else.
Just now on NPR - Open AI has inked $1T over this year. Remember the times when $1B were a serious money?
Remember a year and a half ago when we all scoffed at the prospect of multi-trillion dollar global AI infrastructure build out? Now it's happening.
How much is dumb money, though? That's the real question which remains highly relevant post dotcom bust.
And how much of the money even exists. I remember earlier in the year Altman saying they were spending $500bn on the Stargate project and Must saying he had less than $10bn in actual funds.
It wouldn't surprise me if much of the $1tn most doesn't turn up and the bubble bursts before 1/10th or that becomes real.
Does that matter anymore? The stock market no longer serves as a tool to help allocate resources, it operates more like a Casino now.
> The money's just moving around and around
There's nVidia that we know (primarily graphics cards) and more like an investment firm "nVidia" these days. The stocks have grown so much that they are trying to turn their fortunes around (to sustain growth) by investing everywhere.
nVidia has invested in so many companies in the ecosystem and beyond.
Do they diversify? Is Nvidia buying VXUS?
What about the NVidia/OpenAI deal? Such a giant investment in a huge customer looks a heck of a lot like "circular dealing". That is, invest $100 million so your biggest customer can spend a ton of that on your own chips. You get to report skyrocketing revenue but that revenue was bought with your own money.
If the AMD/OpenAI deal means that OpenAI will put serious work into making AMD GPUs finally be more amenable to general purpose computing, it's actually a huge win for AMD. None of the major numerical computing frameworks (PyTorch, TensorFlow, JAX, etc.) work as well on AMD GPUs versus NVIDIA GPUs, and that's really holding AMD back from making inroads into the machine learning space.
If this means that compute is considerably cheaper for OpenAI, it's a win for them too. But that remains to be seen.
The pivot all the companies made to TikTok style video shitpost apps felt like that too. The ideas are running out and they need a money maker now.
This also has gigantic impact potential. Imagine if Lehman Brothers only had one competitor, and they were also at risk to be seriously impacted.
They should hurry up and build the DCs at double-time then! The last tech crash left us with lots of dark fiber that enabled really cool applications and very low costs.
I can only imagine what could grow out of an oversupply of rack-space and electrical power generation, post-crash.
Electrical power and maybe rack space, but the glass strands of the dark fiber were still perfectly good a number of years later, once demand catches up. Buildings and such still need continuous maintenance. In addition, the actual computational equipment (such as the GPUs that have given nvidia the highest market cap in the world) will be totally worthless.
Tasmania doesn't do much sheep farming. They are more into salmon. Maybe you were making a derogatory reference to the local human population? Fair enough.
Tasmania has 38 sheep per square kilometer and is only surpassed by Victoria with 60 sheep km2. WA has 4 sheep per km2, QLD 1.44, SA 10, NSW 3.
In population terms, Tasmania has 4.5 sheep per person, whereas Victoria has 1.9 sheep per person. NSW 0.28, SA 5.2, WA 3.3, QLD 0.4
[1] https://www.ga.gov.au/scientific-topics/national-location-in... [2] https://www.abs.gov.au/statistics/people/population/national... [3] https://www.wool.com/market-intelligence/sheep-numbers-by-st...
* Population numbers are one head per person, so actual numbers may vary for Tasmania ;-)
Please keep arguing about this, it is relieving a lot of stress to watch this. There have to be more facts than just that. I'm serious. This helps me take focus off the rest of the world burning. I'm serious.
Understandable, it's counting sheep after all.
What are the salmon numbers?
They rise and fall as the orange bellied parrots fly over, salmon leaping to catch the parrots is a sight to behold.
I better get my maugean skates on...
NSW ~ 30.2 sheep / sq km.
> where there is nothing much more than sheep and tourists
kind of like Mississippi but without the tourists part
people have the weirdest/wrong perceptions of mississippi
It's more like "it's like the US, but without the education"
Yea you might need to update your knowledge. They are killing it in schools now
It’s still the 48th worse place for kids.
https://www.clarionledger.com/story/news/2025/06/11/mississi...
punching down on a state you don't understand or know anything about is such a overdone trope. at least be funny!
Luckily our government is full bore on banning the collection of statistics so you can feel better about your state.
They are serious - MS is doing very well in early reading and vaccination. We in Louisiana don’t get to say last except for MS anymore.
ok how bout this, its all relative except for mississippi, there's too much going on between relatives there.
get it? hahahaha
unfortunate life you must live to be so unfunny
oooh look who's punching down now haha yes sir, I have an extremely unfortunate life, I'm the sort of individual who does not have good luck. I try not to bring it up because its simply not on topic for hn, me circling the drain is really nobody's business but mine and the drain, oh yeah, yours too I guess. questions?
[flagged]
So what’s going on in Mississippi these days?
improved literacy rates (https://mississippitoday.org/2025/01/24/jim-barksdale-100-mi...), misspent welfare money with brett farve involved (old news), state auditor doxing private individuals for their political speech, new jesmyn ward novel came out a couple years ago, mississippi filmmaker al warren released a funny movie "dogleg", the band mspaint had a good album a couple years ago, etc
data centers being built but people fighting it. new rezoning in taylor, MS being fought by locals bc they are trying to reclassify agriculture land for heavy industry so they can build an asphalt plant. federal government troops in nearby memphis.
just like anywhere else in america.
When you come in dead last (or in the bottom 5 states) in virtually every possible positive metric, people are going to make negative assumptions. I understand the historical reasons for that, but it doesn’t change what it is.
fair enough, but their assumption wasn't even about that. a better joke on that quote would've been about vermont.
i hear people online punch down on mississippi all the time, and often they don't know anything about the state except whatever metric they've heard about from a headline. the rest of america isn't very far behind, and if you think the state of mississippi isn't a product of america as a whole then you are extremely mistaken. without the industrialized north you have no plantation economy and without the civil war you have no "dead last (or in the bottom 5 states) in virtually every possible positive metric."
i grew up there, attended public school all the way through my BA, and then spent significant time as a young adult there. based on the stereotypical assumptions, it might be shocking to the big brains on hacker news that somebody from mississippi is an audience of this website.
True. As an AI engineer from Gurajatar, Uttar Pradesh I can relate to your experience in this auspicious website sir. We are top human capital.
> i hear people online punch down on mississippi all the time, and often they don't know anything about the state except whatever metric they've heard about from a headline. the rest of america isn't very far behind, and if you think the state of mississippi isn't a product of america as a whole then you are extremely mistaken.
Agreed, people usually just say ‘lol MS is full of idiots, they’re bad at school’ instead of taking the time to understand why. It was more isolated than GA and LA (and AL), there was a higher ratio of slaves to freedmen in the antebellum period, the Delta was undeveloped so lots of impoverished people from across the south moved there to try and develop the land, to name a few reasons.
> i grew up there, attended public school all the way through my BA, and then spent significant time as a young adult there. based on the stereotypical assumptions, it might be shocking to somebody on hacker news that somebody on a similar enough intelligence level to be an audience of this website is from mississippi.
I can say that I don’t assume everyone from Mississippi is stupid, but the generalization about Mississippi that you related seems to be more common than it should be. I think a lot of it has to do with a lack of exposure to people from Mississippi or Mississippi itself.
Thanks for taking the time to respond, I appreciate the discussion.
There was a recent post at the top of HN done by someone who went to USM in Hattiesburg who lives in New Orleans now (and which inspired a show HN https://news.ycombinator.com/item?id=45183050 )
Isn't diversity a positive metric? Mississippi has the highest black percentage of all the states. If one wants to use statistics, use all of the statistics.
Is there a lot of power generation in Tasmania? That is the main criteria for AI data centers from what I heard. Latency is less critical than cost of power.
Plenty of power and it’s almost always renewable: https://explore.openelectricity.org.au/energy/tas1/?range=7d...
There’s also plenty more there than sheep and tourists (and not really that many of those)
There's pretty national parks, convict history and some pretty tasty whiskeys and wines, but not a whole lot of reasons to build a data centre. The power argument isn't particularly compelling either, because it's much more sunny on the Australian mainland.
> The power argument isn't particularly compelling either, because it's much more sunny on the Australian mainland.
But the vast majority of Tasmania's power is hydroelectric. Hydro is a much more desirable renewable than solar because it essentially is its own built in battery.
Might need to fact check yourself there mate. Tasmania has issues with enough power for some industries. There was a factory that could not get enough power to a planned upgrade.
God knows how a datacentre would do down there....
Regardless of a bubble or not, don't some operators do exceptionally well financially, without a lot of recognition, when things are murky compared to transparent or having clarity?
And how long has this kind of thing probably been going on?
So the real risk with that data center is the fact that Tasmania is a culdesac on the internet. We have three fiber cables coming to the island, all from Victoria (directly to the north). The whole state has lost internet before due to someone in Victoria digging a trench through the cable (yes a silly mistake but we're all humans).
It's not a bad idea to put data centers here, but we really need a few more links out to the world from here.
Where would you send them to achieve useful path diversity?
Adelaide, Sydney and New Zealand. https://www.submarinecablemap.com/submarine-cable/bass-strai...
Ed Zitron writes about this constantly: https://www.wheresyoured.at/the-case-against-generative-ai/
All of the big players - Nvidia, OpenAI, Oracle, Microsoft - are in insane circular financing agreements that would make Enron executives blush.
Wow this is quite a rant.
However Zitron seems to have forgotten that Google exists or makes TPUs. He mentions Google only 10 times in the entire article, always in a minor way.
Any reason to believe Google's unit economics on AI are any different than the other players here?
And Google is an advertising company. Mostly in search, and increasingly dependent on YouTube. Everything else is a net money loser, including Waymo, Gemini etc.
Almost doesn’t matter what Google’s short term unit economics on AI are as long as their shareholders and board are OK with it. Google makes money hand over fist and can afford it. Microsoft is fine, Oracle is fine, TSMC is fine. These other shops are leveraged to the hilt, so they’ll be in a world of hurt if people stop buying, financing, or investing.
Oracle is going to get screwed if they build a bunch of data centers and the customers (mostly OpenAI) all declare bankruptcy.
Oracle has started creating bonds with really long maturity 40years...?
Oracle is not fine. They are borrowing money hoping to get paid back by money losing OpenAI propped up by VC funding.
No business is going to run workloads on OCI outside of ones running Oracle. They a They are a way distance fourth in cloud. I’ve been working in cloud consulting for five years including the first three directly at AWS (Professional Services). No one worried about having talking points about competing against Oracle.
Microsoft, Google and Amazon have both internal products that can benefit from inference and cloud hosting.
> Any reason to believe Google's unit economics on AI are any different than the other players here?
Only when it comes to their TPUs, and sometimes that one thing may just be the difference to push them over the hump.
Per-token cost-wise, TPUs (& specialized processors in general) will beat GPUs every time. The efficiency difference between the 2 types is never to be ignored, & is likely why they can shotgun it everywhere.
> And Google is an advertising company. Mostly in search, and increasingly dependent on YouTube. Everything else is a net money loser, including Waymo, Gemini etc.
1) Each venture should be treated as a (relatively) isolated vertical slice
2) 9 out of 10 times, a venture just doesn't break even. That's just the nature of the business.
You should listen to the latest episode of the Acquired podcast about Google.
Google also has GCP and unlike OpenAI who is dependent on VC funding and Oracle who is borrowing money. Google throws off cash like crazy and self funds its infrastructure which is already better than everyone else’s
Well there's no circular financing for one thing.
Feedback loops. Always with the feedback loops.
It's a long-term gamble but a short-term win.
In almost all scenarios, a setup with this incentive structure will lead to massive adoption. It's too tempting, and with most jobs / political positions being short term (<5yrs) ones, people optimize for their time in that timeframe, not longer.
Boards will pursue stock buybacks (short term growth, long term may cause trouble if there's a downturn), banks will lend out subprime mortgages (hit your sales numbers in the short term, at the cost of long term risk), etc etc.
This situation is no different. There's money flowing in and there's less red tape since everyone is being pressured to allow it. It might work out in the long term, it might not, but it will 100% benefit those who push it in the short term. People will get promoted for driving a new data center, politicians can promote more jobs being added, everybody wins... for now.
The future economic aspect becomes irrelevant when the short term candy is sweet enough.
IBGYBG. “I’ll be gone, you’ll be gone”. Infamous email sign-off associated with the run-up to the Great Financial Crisis. Used by Wall Streeters asking analysts to inflate credit ratings for undeserving securities, backed by risky mortgages.
Every time I read one of these articles the main issue I have is that it doesn't take into account the huge shortages of compute that are going on all the time. Anthropic and Google especially have been incredibly unreliable, struggling to keep up with demand.
Each of the main providers could easily use 10x the compute tomorrow (albeit arguably inefficiently) by using more thinking for certain tasks.
Now - does that scale to the 10s of GWs of deals OpenAI is doing? Probably not right now, but the bigger issue as the article does point out in fairness is the huge backlog of power availability worldwide.
Finally, AI adoption outside of software engineering is incredibly limited at work. This is going to rapidly change. Even the Excel agent Microsoft has recently launched has the potential to result in hundred fold increases in token consumption per user. I'm also suspect of the AI sell through rate being an indicator that it's not popular for Microsoft. The later versions of M365 copilot (or whatever it is called today) are wildly better than the original ones.
It all sort of reminds me of Apple's goal of getting 1% in cell phone market share, which seemed laughably ambitious at one point - a total stretch goal. Now they are up to 20% and smartphone penetration as a whole is probably close to 90% globally of those that have a phone.
One potential wild card though for the whole market is someone figuring out a very efficient ASIC for inference (maybe with 1.58bit). GPUs are mostly overkill for inference and I would not be surprised if 10-100x efficiency gains could be had on very specialised chips.
the huge demand exists right now because the cost of a token is near zero. and companies have figured out one weird hack to gaining value in the stock market, which is to brag about how many tokens are being crammed into all manner of places that they may or may not belong.
customer value must eventually flow out of those datacenters in the opposite direction to the the energy and capex that are flowing in
do people actually want all this AI? I see studio ghibli portraits, huge amounts of internet spam, workslop... where is the value?
> Each of the main providers could easily use 10x the compute tomorrow (albeit arguably inefficiently) by using more thinking for certain tasks.
That's true for everyone with regard to any resource.
The question is whether the 10x increase in resources results in 10x or more increase in profit.
If it doesn't then it doesn't make sense to pay for the extra resources. For AI right now, the constraint is profit per resource unit, not number or resource units.
Why have spot H100 prices been going down then? It was roughly $3/hr a year ago and now it is closer to $2.2/hr.
Reminds me of the fiber boom
"The later versions of M365 copilot (or whatever it is called today) are wildly better than the original ones."
I find AI agents work very poorly within the Microsoft ecosystem. They can generate great HTML documents (because it's an open source format maybe?) but for word documents, the formatting is so poor I'd had to turn it off and just do things manually.
Opposing anecdote: I got consistent performance out of Grok and Qwen (17 providers on Openrouter) throughout the day but Gemini gets slow and dumb at times.
BlackRock just bought a data center for 40bn last week. BlackRock. The animals that bought all the housing once. They must be stupid I guess.
BlackRock did not buy “a” data center, it bought a data center company with 78 data centers. I have no comment on whether or not it was a good deal, but your framing is silly.
Unless the AI industry can figure out how to be profitable soon (to which basically nobody has a clear path to profitability besides maybe ad revenue) it's hard to see this not blowing up in a year or two. The bills for all of this are going to come due eventually and the AI CEOs can only convince investors to keep letting them burn billions for so long.
It's incredible how much money is being lit on fire without anyone having an answer to the question "how will you generate revenue". Meanwhile alternative energy discussions here are met with harsh criticisms of their bottom lines while they have some of the most firm demand of any industry. Society should be prudent with their investments, but only when it's used for R&D into essential stuff and not a bunch of toys for rich psychopaths?
Their only answer is "if we invent AGI and fire everyone, the profits will be limitless".
Zuck has made it clear this is a Pascal's Gamble:
"If we end up misspending a couple of hundred billion dollars, I think that is going to be very unfortunate, obviously. But I actually think the risk is higher on the other side . If you build too slowly and superintelligence is possible in 3 years, but you built it out assuming it is possible in 5 years, then you are out of position on the most important technology."
His assumption is that superinteligence is close, its just a question of whether it is 3 or 5 years!
No, the first sentence makes it clear that he is aware AGI is not a given. His position is that there is a possibility we can reach super intelligence, and given that possibility they want to be on the bleeding edge and are investing accordingly, given that even total failure won’t cripple their business and all their competitors are doing the same.
"If you build too slowly and superintelligence is possible in 3 years, but you built it out assuming it is possible in 5 years, then you are out of position on the most important technology"
This made me think Zuck sees it as a question of when rather than if. I.e its more a question of 3 vs 5 years rather than possible vs non possible.
And yet, imagine it's 100 million jobs, at 100K per job. That's 10 trillion dollars a year, well worth the investment! Except that it won't be 100K per job that AI companies will capture, it'll be 2K. So that's not 10 trillion dollars a year, it's 200 billion. Which is about what a big tech company makes in advertising already.
That's been the story for the last 20 years of VC culture, and the answer has so often been, ruin the service with advertising.
>> Cheerleaders such as Sam Altman, OpenAI’s boss, argue that the risks of underbuilding are at least as serious as those of overbuilding, because of the long-term economic potential of generative AI.
Pure hucksterism.
To summarize:
* questionable demand for AI products make the DC investments risky (makes sense)
* DCs being built in “remote” areas with cheap land may become obsolete/replaced by other DCs making use of said cheap land (questionable; DCs can upgrade cheaper than building a new one)
* financing for DCs used to be from Big Tech but is now spread out among private equity, sovereign wealth funds etc increasing the exposure of the economy to failure of these investments (again, questionable, unless they are being financed by bank loans)
The most salient concern seems to be a lack of demand. I don’t see why that would change in the future.
How much of this capex is infrastructure that can support future generation of compute vs current-gen GPUs that will become obsolete?
I can't find the reference now unfortunately, but I read it was about 50:50, so half is the data center land / building / cooling / power generation, and half is in the GPUs themselves.
This is notably very different from the dot-com build out of dark fibre, where digging the holes cost the vast majority of the money, and the fibres and network equipment cost very little in comparison.
I wonder how much of this round tripping is because there are only a few companies that have the demands and can meet those demands...
Last weekend we had dinner at a friend to celebrate their child's birthday. They work at OCI in a very senior position and were heading out next day to troubleshoot that exact datacenter in Abilene, TX next day, to presumably yell at various people about cabling and schedule and whatnot. They were not looking forward to it because a) TX and b) it's Abilene TX and c) the overall situation.
We spoke quite a bit about this project, circular money and things and the impression I got is that OCI is just grabbing money while the perpetuum mobile is unrolling.
I am looking forward to hearing how the trip went when I see them next.
The market can stay irrational longer than you can stay solvent.
https://www.bloomberg.com/news/articles/2025-10-07/oracle-si...
Oracle Sinks on Report Its Cloud Margins Are Lower Than Expected
Jeran Wittenstein October 7, 2025 at 6:06 PM GMT+2
Oracle Corp. shares tumbled after a report that the software maker’s profit margin in its cloud computing business is lower than many on Wall Street have been estimating.
While Oracle generated roughly $900 million in revenue from the rental of servers powered by Nvidia Corp. chips during the three months ended in August, the company only managed about $125 million in gross profit, the Information reported, citing internal corporate documents.
* Oracle sinks 7%, after having jumped 36% earlier in September.
Sure seems like a combination of TINA and massive amounts of hope.
Cloud has failed (by becoming more expensive then on prem) so they're buying up everything in a last ditch effort to centralize compute. Really silly since all those GPUs will be obsolete soon.
Cloud was always going to be more expensive than on prem, that wasn't what was being sold. Cloud sells that they are able to give you high availability across multiple regions without having to do an expensive capital investment. Amazon/Microsoft/Google is selling the talented engineering for getting these data centers working around the world at a high level of availability to companies. Most companies like recurring costs compared to capital expenditures. Companies also prefer lower head counts and paying for services. It is BETTER from an MBA pov to pay Amazon to run a Data center since they are experts than for you to run a data center as a finance company.
> since they are experts
This is the engineering perspective, not the finance perspective. As an engineer holding an MBA, I've made the argument countless times in BigCo to move from cloud deployments to on-prem. When you're a startup, you often simply don't have the cash on hand to make the capital expenditure to build out datacenter capacity, especially with an uncertain (but hopefully high) expected rate of growth. When you're a BigCo, the script flips; you have plenty of cash on hand and you want to improve overall profitability, which is done by using capital expenditures to reduce operating expenditures, i.e. funding datacenter build-outs to reduce cloud bills.
> experts
Companies can hire experts and can still out-source to colos if they prefer. This is a question of political will and risk analysis.
> Most companies like recurring costs compared to capital expenditures.
Actually big companies prefer capex to opex and cloud pushed them in the opposite direction of what they'd naturally prefer. But the other advantages you cite + hype overruled the liability of switching from onprem capex to opex.
>a last ditch effort to centralize compute.
I started with mainframes and I'm not going back.
[flagged]
Please don't do this here.
We detached this comment from https://news.ycombinator.com/item?id=45510207 and marked it off topic.
Apologies. Thanks for your hard work keeping the site quality high!
Also, this was remarkably hard to find. Might I suggest if the logged in user is looking over their own comments, eg me viewing https://news.ycombinator.com/threads?id=fragmede, that a flagged dead comment with a moderator responding shouldn't be collapsed by default? If that's intentional then nevermind, but as I am truly grateful for this site and you and DanG's (and everyone else's) hard work, I want to point out that it was hard to find and wouldn't have come across it if I didn't have a program running that hits the Firebase API and notifies me on replies. Someone who didn't have such a program could easily miss mod responses.