In the grim darkness of the AI boom everyone is watching high profitability industries move into areas characterized by having few providers due to incredibly poor margins and high expenses and declare it business genius.
Yes… and AWS is critical to the cash flow and valuation of its parent Amazon. The market looking like it’s trending towards oversupply does not paint a good picture in that context. AWS has to operate at these margins as most of the rest of Amazon is a very low margin business.
Google, Microsoft and others have separate cash cows. For Amazon, AWS is the cash cow. They can’t afford a blip here.
xAI has shown this to be quite lucrative. And it seems to even make some sense - if the contract can be ended on relatively short notice, you basically have the capacity on stand-by if you ever need it yourself (accumulating GPUs is not trivial), but can monetize it if you don't need it.
Though it's probably a bad sign generally that you can't capitalize on all the GPUs you've acquired.
That doesn't mean they're lucrative, though. They cost tens of billions to build and huge sums to run. Even if I believe that they are now running profitably, that doesn't mean they're "quite lucrative" overall.
SpaceX's numbers show xAI increasing its losses over time. In 2025, xAI lost $6.4 billion on $3.2 billion. In 2026, xAI posted a $2.47 billion operating loss in the first quarter alone.
If you mean Alexandr Wang, the number is off by an order of magnitude. Forbes estimates he made over $3 billion from selling his startup to Meta and getting hired as chief AI officer.
I'm excited for the opportunities that an abundance of compute brings, even in the absense of AI applications. Big compute is the right direction, if we can afford it. There are always things to simulate or discover or automate in some way shape or form.
For one thing, I still don't understand Meta as a business. It seems like Zuck refused to accept owning a boring ad business and keeps trying to act like a tech business.
For another, this seems like a move that happens when the bubble pops.
The dotcom bubble went from having tons of business raising fortunes on the promise of business models the internet unlocks to having few internet businesses left at that scale and a bunch of unused fiber.
Now we're starting to see all the companies claiming AI products will make fortunes to them trying to sell unused hardware capacity.
He might be scared that his ad business won't be as profitable in a few years. He'll still be rich, but perhaps that's not enough for him. He has all the correct number and knows exactly how many of their users are bots and how much of the ad revenue comes from scams. Maybe he sees a potential risk that we can't.
Could also be a self-image thing. Facebook isn't exactly young and fashionable anymore, it's more akin to a weird mid-90s idea of a close walled Internet in some sense. Instagram users I'd guess is mostly younger women. None of these really align with Zuckerbergs macho persona.
The difference is the hardware will become obsolete in a few years, unlike the dark fibre (really, rights of way) that provided cheap connectivity for years after the crash.
The hardware will become obsolete but we may end up with an oversupply of cheap power - overbuild on things like Nuclear and solar. That would be nice.
So 10% failure rate per gpu year. Some might be repairable and there might be a failure rate acceleration as time passes etc. So very approximate numbers.
It's an inherently different thing. Fibre is infrastructure. It'll still function decades later. GPUs at this scale are consumables. I've heard 3-5 years lifespan before they fail out. This might be a low estimate, but even if you double it - they're 50% of the cost of datacenters. We're flushing entire countries worth of economic value down an Nvidia shaped toilet.
It's not that they fail (which they also do), it's that they become obsolete. They consume more energy to do fewer operations relative to newer nodes. This is the basis behind Moore's law.
Unfortunately he has too much cash since he owns some of the largest social networks and thus receives insane amounts of ad spend… this will not end soon
It’s a pretty consistent story in tech. Meta built a successful ads platform. That success went to their head and the company starts acting like it’s just great at tech and innovation and tried to do a bunch of other stuff. Turns out it’s pretty crap at most other things and can’t figure out how to make money from anything other than selling ads. Company is very headstrong and refuses to admit it’s just not good at other stuff. It has enough cash from ads to not go bankrupt and so just carries on with this nonsense on cycle after cycle of crazy side projects that go nowhere.
> For one thing, I still don't understand Meta as a business
Zuck got spooked by apple ads change a few years ago, which crashed the meta's stock to double digit. So Zuck is trying to own a platform. They want to be a version of openai selling ads
Btw I am just quoting someone else. It's in the news in for a long time. Besides ads monetization, Zuck's moves in recent years are mostly about creating their own platform so the Apple incident never happen again
It contextualizes all the strange Metaverse and VR-goggles stuff quite well. It always seemed like a weirdly forced product without any particular market fit - who is clamoring for a version of Teams where you need a bulky helmet, and can theoretically buy pretend digital land in order to have a... less useful corporate website? Who is interested in buying dorky goggles that instantly mark you as a weird creep who is probably recording everything you see?
Owning A Platform being the point instead of any actual use-case tracks perfectly with all these weird, forced pitches. These are all products for nobody.
I think we're going to look back on this as one of the things that really exacerbates this bubble. It's one thing for these companies to have high expectations of their future compute needs and therefore overbuild. But what the SpaceX/Anthropic deal showed was even if you fall short, you can profit by selling off your overcapacity. This means there is essentially no incentive to limit how much capacity you're building, anything you don't use you'll still profit from.
The problem is that only works if aggregate demand is higher than supply. At some point, all these new data centres are going to come on line, and the frothy "throw AI at everything with no regard to costs" is going to drawn down across the industry. And this supply will be much larger - because everyone thought the down side was limited.
At that point you're going to have all of these companies trying to dump their excess capacity on the market and it suddenly won't be true that you can just sell capacity to your competitors.
Obviously this won't bankrupt Meta - it'll just eat into their profits from their ads business. But it likely will drive a bunch of neo-clouds out of business very quickly, and the technology providers like Nvidia etc will suddenly come back to reasonable P/Es.
Meta is just another company that, like xAI/Grok, is at last starting to face reality: there are going to be two or three winners in this space, and they are not one of them.
Excess AI capacity? Huh? Wait a minute does that mean they've built more data centers than there's demand for? Oh! How did that happen? Well, I'm a complete idiot, a layman, not even a developer, what do I know about economics and technology...
It’s a hard thing to measure right? Google under bet on demand for their AI infra and left lots of money on the table with Anthropic/ has to buy computer from SpaceX for awhile, Meta over built.
Even then though at their scale - billions of DAU with pretty elite consumer distribution in terms of InstaGram and WhatsApp the future is hard to predict.
Like they have much lower cost of capital than OpenAI or Anthropic for example and far more direct to consumer muscle, if they could've gotten their models into a better position it's very possible they could've beaten out OpenAI on consumer AI. Probably a smaller prize to win than B2B AI but still huge.
And then if the case where AI is panning out but they are failing to execute (where we appear to be now) they have the option of becoming a SpaceX style NeoCloud that is pretty solid, even though most of their moves were pretty bad along the way.
So like likes of AWS are spending crazy sums building capacity citing amazing demand they can’t meet. Meanwhile those with capacity are saying hey maybe we don’t need so much and are building cloud businesses to compete with AWS and flood the market with excess capacity.
I don't really think this necessarily means demand is limited.
It just means demand is consolidating to winners. There's no denying openai and anthropic and maybe Google have a ton of demand. The fact that xai and meta do not does not mean no demand exists.
Everyone slamming the brakes on tokenmaxing and companies putting strict AI budgets in place is surely having at least some impact on demand.
And the point is not that “no” demand exists but the likes of AWS are building to the moon citing massive future need but that simply doesn’t jive with all this unused capacity from others coming to the market.
And building out massive capacity to sell a product that’s likely to trend to a low margin market commodity isn’t a great business.
Most of the demand for data center GPUs comes from the big tech companies themselves, to develop products/services.
Facebook and Twitter are both now signaling that the resources they funded with vast amounts of capital would be worth more to shareholders if they were leased to the highest bidder.
This is either a realization that their internally developed product/service roadmap is not panning out, or that the lease demand is so high that they’d be fools not to take advantage.
Good news. The more providers, the more competition and lower prices.
Meta are laying their own sub-sea cables globally at present:
https://www.bbc.co.uk/news/articles/ckgrgz8271go
In the grim darkness of the AI boom everyone is watching high profitability industries move into areas characterized by having few providers due to incredibly poor margins and high expenses and declare it business genius.
AWS have 35% profit margins.
Google Cloud recently told Meta they couldn't supply all the ai compute they want so Meta has no choice but to secure their supply?
https://memeburn.com/google-limits-metas-gemini-ai-access-as...
Yes… and AWS is critical to the cash flow and valuation of its parent Amazon. The market looking like it’s trending towards oversupply does not paint a good picture in that context. AWS has to operate at these margins as most of the rest of Amazon is a very low margin business.
Google, Microsoft and others have separate cash cows. For Amazon, AWS is the cash cow. They can’t afford a blip here.
And much like the 41st millenium, there are no good guys, just the eternal laughter of hungry shareholders.
xAI has shown this to be quite lucrative. And it seems to even make some sense - if the contract can be ended on relatively short notice, you basically have the capacity on stand-by if you ever need it yourself (accumulating GPUs is not trivial), but can monetize it if you don't need it.
Though it's probably a bad sign generally that you can't capitalize on all the GPUs you've acquired.
> (accumulating GPUs is not trivial)
Also like accumulating milk rather than gold.
A approximate failure rate I have been able to find is: https://www.tomshardware.com/tech-industry/artificial-intell...
Which would imply 42% loss over 5 years if linear attrition at 10% failure rate per gpu year.
(((1-(148+72+19+17+6)÷16 384)^(1÷54))^(365)
Dunno how many faults are repairable in practice. The failure rates of CPUs were alot lower for comparison.
>xAI has shown this to be quite lucrative.
The demand is likely short term till the renter can ramp up their own centres/find better deals.
> xAI has shown this to be quite lucrative
How do you know this? Have they released any numbers?
I think the SpaceX IPO showed them to rent the clusters for billions a month
That doesn't mean they're lucrative, though. They cost tens of billions to build and huge sums to run. Even if I believe that they are now running profitably, that doesn't mean they're "quite lucrative" overall.
SpaceX's numbers show xAI increasing its losses over time. In 2025, xAI lost $6.4 billion on $3.2 billion. In 2026, xAI posted a $2.47 billion operating loss in the first quarter alone.
> xAI has shown this to be quite lucrative
Come back when you have actual numbers on dollars spent, interest rates and the time spent.
To be fair to meta, they've been running data centers for almost two decades now, so they'd be in a different position to XAI.
So have they just given up on llama then? What happened to the 25 year old Zuck paid $250M for?
If you mean Alexandr Wang, the number is off by an order of magnitude. Forbes estimates he made over $3 billion from selling his startup to Meta and getting hired as chief AI officer.
Nah, it's a guy named Matt Deitke and he was 24 at the time.
25 year old has been doing the rounds giving podcast and other interviews so you can listen to him.
Probably got laid off already
I'm excited for the opportunities that an abundance of compute brings, even in the absense of AI applications. Big compute is the right direction, if we can afford it. There are always things to simulate or discover or automate in some way shape or form.
Big computer isn't the right direction if we have to sacrifice LOCAL COMPUTE for the big cloud future.
For one thing, I still don't understand Meta as a business. It seems like Zuck refused to accept owning a boring ad business and keeps trying to act like a tech business.
For another, this seems like a move that happens when the bubble pops.
The dotcom bubble went from having tons of business raising fortunes on the promise of business models the internet unlocks to having few internet businesses left at that scale and a bunch of unused fiber.
Now we're starting to see all the companies claiming AI products will make fortunes to them trying to sell unused hardware capacity.
He might be scared that his ad business won't be as profitable in a few years. He'll still be rich, but perhaps that's not enough for him. He has all the correct number and knows exactly how many of their users are bots and how much of the ad revenue comes from scams. Maybe he sees a potential risk that we can't.
Could also be a self-image thing. Facebook isn't exactly young and fashionable anymore, it's more akin to a weird mid-90s idea of a close walled Internet in some sense. Instagram users I'd guess is mostly younger women. None of these really align with Zuckerbergs macho persona.
The difference is the hardware will become obsolete in a few years, unlike the dark fibre (really, rights of way) that provided cheap connectivity for years after the crash.
The hardware will become obsolete but we may end up with an oversupply of cheap power - overbuild on things like Nuclear and solar. That would be nice.
Do we have any good data on how many of the GPU burn out given hard usage?
Are we talkinga bout loosing 50% of the hardware as it fails? Or far less?
I tried to do a calculation.
https://news.ycombinator.com/item?id=48762428
So 10% failure rate per gpu year. Some might be repairable and there might be a failure rate acceleration as time passes etc. So very approximate numbers.
It's an inherently different thing. Fibre is infrastructure. It'll still function decades later. GPUs at this scale are consumables. I've heard 3-5 years lifespan before they fail out. This might be a low estimate, but even if you double it - they're 50% of the cost of datacenters. We're flushing entire countries worth of economic value down an Nvidia shaped toilet.
Anyone got any good testing data? Surely someone knows rough failure rate from constant usage?
It's not that they fail (which they also do), it's that they become obsolete. They consume more energy to do fewer operations relative to newer nodes. This is the basis behind Moore's law.
Ah gotcha, ya that will be interesting to see, but does seem like a more likely life of 7 to 10 years is possible.
I am curious on actual hardware failure rates due to heat etc.
It's obsolete for training but is it obsolete for inference?
The factories won't be obsolete, even if the chips are.
The fabs for current nodes will be obsoleted by newer nodes.
not if they completely crash manufacturing with their push for fascism; be more optimistic!
Unfortunately he has too much cash since he owns some of the largest social networks and thus receives insane amounts of ad spend… this will not end soon
And he can't be fired by shareholders.
He just has that much money and his 'whatever ecosystem this is' continues to work, it has to be a kind a playground for him.
He could invest in paying back society but he cares more about continuing his ignorance.
Just imagine how his ego got stroked when he was agreeing with trump like a bootlicker but still sitting at a table at the Whitehouse.
Even with this clown this has to count for something in his books.
It’s a pretty consistent story in tech. Meta built a successful ads platform. That success went to their head and the company starts acting like it’s just great at tech and innovation and tried to do a bunch of other stuff. Turns out it’s pretty crap at most other things and can’t figure out how to make money from anything other than selling ads. Company is very headstrong and refuses to admit it’s just not good at other stuff. It has enough cash from ads to not go bankrupt and so just carries on with this nonsense on cycle after cycle of crazy side projects that go nowhere.
> For one thing, I still don't understand Meta as a business
Zuck got spooked by apple ads change a few years ago, which crashed the meta's stock to double digit. So Zuck is trying to own a platform. They want to be a version of openai selling ads
This is an interesting angle. I had not considered how fundamentally exposed Meta is, should Apple and Google decide to break omerta.
Btw I am just quoting someone else. It's in the news in for a long time. Besides ads monetization, Zuck's moves in recent years are mostly about creating their own platform so the Apple incident never happen again
It contextualizes all the strange Metaverse and VR-goggles stuff quite well. It always seemed like a weirdly forced product without any particular market fit - who is clamoring for a version of Teams where you need a bulky helmet, and can theoretically buy pretend digital land in order to have a... less useful corporate website? Who is interested in buying dorky goggles that instantly mark you as a weird creep who is probably recording everything you see?
Owning A Platform being the point instead of any actual use-case tracks perfectly with all these weird, forced pitches. These are all products for nobody.
How is that different from Facebook
Actually his basket case investments with VR is what led to a loss of investor trust
Good take! I buy it.
> It seems like Zuck refused to accept owning a boring ad business
Zuck sees a real problem: not owning the platform.
Hence the VR and other attempts. He sees what he has today as a dead end. The fear grips him.
In theory Google and Apple could ban Meta’s apps and it would be game over.
lol that’s charitable
The real problem for Zuck is he thinks he’s some grand master who can play 5-d chess, except he can’t
I think we're going to look back on this as one of the things that really exacerbates this bubble. It's one thing for these companies to have high expectations of their future compute needs and therefore overbuild. But what the SpaceX/Anthropic deal showed was even if you fall short, you can profit by selling off your overcapacity. This means there is essentially no incentive to limit how much capacity you're building, anything you don't use you'll still profit from.
The problem is that only works if aggregate demand is higher than supply. At some point, all these new data centres are going to come on line, and the frothy "throw AI at everything with no regard to costs" is going to drawn down across the industry. And this supply will be much larger - because everyone thought the down side was limited.
At that point you're going to have all of these companies trying to dump their excess capacity on the market and it suddenly won't be true that you can just sell capacity to your competitors.
Obviously this won't bankrupt Meta - it'll just eat into their profits from their ads business. But it likely will drive a bunch of neo-clouds out of business very quickly, and the technology providers like Nvidia etc will suddenly come back to reasonable P/Es.
> But what the SpaceX/Anthropic deal showed was even if you fall short, you can profit by selling off your overcapacity.
Do you have proof that the deal is profitable (and for how long?) for SpaceX ?
[dupe] Earlier on source: https://news.ycombinator.com/item?id=48745966
Meta is just another company that, like xAI/Grok, is at last starting to face reality: there are going to be two or three winners in this space, and they are not one of them.
"Bubble? What Bubble? I don't see no steenkin' Bubble."
Excess AI capacity? Huh? Wait a minute does that mean they've built more data centers than there's demand for? Oh! How did that happen? Well, I'm a complete idiot, a layman, not even a developer, what do I know about economics and technology...
It’s a hard thing to measure right? Google under bet on demand for their AI infra and left lots of money on the table with Anthropic/ has to buy computer from SpaceX for awhile, Meta over built.
> Meta over built.
In retrospect, this seems pretty easy for Meta to have done as heretofore they were a single captive customer AI shop.
Even then though at their scale - billions of DAU with pretty elite consumer distribution in terms of InstaGram and WhatsApp the future is hard to predict.
Like they have much lower cost of capital than OpenAI or Anthropic for example and far more direct to consumer muscle, if they could've gotten their models into a better position it's very possible they could've beaten out OpenAI on consumer AI. Probably a smaller prize to win than B2B AI but still huge.
And then if the case where AI is panning out but they are failing to execute (where we appear to be now) they have the option of becoming a SpaceX style NeoCloud that is pretty solid, even though most of their moves were pretty bad along the way.
So like likes of AWS are spending crazy sums building capacity citing amazing demand they can’t meet. Meanwhile those with capacity are saying hey maybe we don’t need so much and are building cloud businesses to compete with AWS and flood the market with excess capacity.
Oh yeah, this is gonna end well.
I don't really think this necessarily means demand is limited.
It just means demand is consolidating to winners. There's no denying openai and anthropic and maybe Google have a ton of demand. The fact that xai and meta do not does not mean no demand exists.
Everyone slamming the brakes on tokenmaxing and companies putting strict AI budgets in place is surely having at least some impact on demand.
And the point is not that “no” demand exists but the likes of AWS are building to the moon citing massive future need but that simply doesn’t jive with all this unused capacity from others coming to the market.
And building out massive capacity to sell a product that’s likely to trend to a low margin market commodity isn’t a great business.
Yeah today’s demand at existing and higher prices won’t persist.
Firms are currently experimenting and figuring out the correct way to use these tools in the production function.
Most of the demand for data center GPUs comes from the big tech companies themselves, to develop products/services.
Facebook and Twitter are both now signaling that the resources they funded with vast amounts of capital would be worth more to shareholders if they were leased to the highest bidder.
This is either a realization that their internally developed product/service roadmap is not panning out, or that the lease demand is so high that they’d be fools not to take advantage.
> Oh yeah, this is gonna end well.
Hope so. It also would put hardware prices down.