> If Anthropic is losing a fortune now with $13 billion in costs how are they going to break even spending $100+ billion per year? Don’t worry, their godlike AI will figure that out?
It's wild that this is an accurate summation of their position here. It's equally wild that a company building and selling a technology as actually useful as an LLM - yes, I think they're useful, sue me - can't begin to think about how to make money with one.
I agree with the article - they'd better hope they are messianic super geniuses, because if not, they're gonna seem real stupid when we have all moved in in a few years. The technology will stay, but that doesn't mean they will.
> I agree with the article - they'd better hope they are messianic super geniuses
It sounds remarkable, and I didn't believe it at first. But after more reading, I am convinced an influential part of Anthropic leadership really is something like a religious movement
Surely this is a "hook 'em with low prices, then raise them later" scheme--the biggest one yet.
I wonder where the split between self-hosted LLMs (w new more powerful LLM hardware) and hosted LLMs (big iron from an LLM-provider or at your office) will fall. Will there be a future for both? LLMs and other ML tools just with different levels of capability? etc.
No, the demand is effectively infinite at the moment and models are getting cheaper to serve. At least for a while price will go down when keeping capabilities constant.
I think "locally hosted LLMs" and "frontier LLMs hosted by frontier labs" will both end up as less than 1% of "Open Source LLMs hosted remotely by a lowest-bidder commodity compute provider"
We're one well-publicised security pop from locally hosted being a non-negotiable for some large set of buyers.
They've convinced every company in the country, including in previously security or compliance-centric environments (finance, medical, government), to hand them hyper-sensitive data.
We've already seen the leakage scenario, with the academics who found the labs scooping them on the research they were using the LLMs to assist. What's to say they-- or someone who infiltrated their infrastructure-- won't do the same thing with your propriatery business details?
I'm not sure open LLMs are a great play for commodity vendors. A lot of "lowest bidder commodity services" work because they can wildly oversell. That VPS hosting a small-biz website or 1000-messages-a-week internal email can run on a Raspberry Pi 1B, so you can pretty easily sublet a $1000 server to 50 customers at $10 per month each. Current LLM hardware plays tend to need more expensive gear and seem to be running at higher utilization rates, so the price that actually fits the underlying business might not look as appealing.
Their TAM basically presumes a good percentage of businesses will just be defacto subsidiaries of their products and they will essentially own the profit revenue stream, like a drug dealing pimp who gets their hookers addicted to crack to keep them working and inured to their loss of self suffiency.
Just look at palanetirs forward deployed engineers for how that'll start.
I had the experience of working at a shop where the business side would use that sort of justification for all sorts of antics.
The favorite I was not involved in, was when they spent many months (at least 3, maaaaybe more than 6?) working on a thing, with verbal promises on the other side, but then the owner of the business that kept promising to sign a contract once we got to a certain point died and at that point the whole thing got quietly killed.
Again, not my project; the ones I was involved in (different product,) at least half of which did materialize but most of them never made up the volume for the pain of the work, either on the dev side or the business side [0].
[0] - My favorite was when we did an integration with Tesla. they decided to use a VIN as the primary key for their salesforce integration. So, if a customer got an inspection on a trade-in, waited, then decided to try again... oops!
What never checks out (aside from most of it) is how they will raise prices to a market that shrinks due to unemployment. Can’t sell apples to people without money.
Here are my biggest risks for Anthropic, OpenAI and SpaceX that don't seem to get the attention they deserve:
1. What happens to your existing hardware investments when NVidia releases new hardware that allows buyers to run the same models at a fraction of the cost in energy times? IMHO performance-per-Watt is going to be the only significant metric here. Yes, your existing hardware will still work but efficiency is going to be the name of the game;
2. How are you going to deal with the threat of open weight models that can be run locally and/or or cloud GPU infrastructure? We already have relatively affordable hardware (ie Mac Studios) that can run models as large as 500B-1T parameters with 3-4 bit quantization. That's only going to get better; and
3. The only use cases for AI really are:
a) Raising prices;
b) Cutting costs, particularly labor; and
c) Military applications such as targeting and autonomous drone and missile operation.
A society where even 20% of people end up unemployed is one that catastrophically fails.
‘Anthropic is making a massive bet that AI will transform the global economy more profoundly than industrialization, electricity and the internet, according to its IPO prospectus seen by Reuters.’
I mean, if you’re gonna aim big may as well go all out. Next week OpenAI will claim to transform the global economy more than life itself.
All horseshit, of course. Still haven’t found a benefit to AI in my work or personal life. I hope the crash is a doozy.
I think AI will transform things but the Anthropic financials are crazy. I suspect you'll get your crash. It's like Webvan saying the internet would transform commerce. Yes but...
> If Anthropic is losing a fortune now with $13 billion in costs how are they going to break even spending $100+ billion per year? Don’t worry, their godlike AI will figure that out?
It's wild that this is an accurate summation of their position here. It's equally wild that a company building and selling a technology as actually useful as an LLM - yes, I think they're useful, sue me - can't begin to think about how to make money with one.
I agree with the article - they'd better hope they are messianic super geniuses, because if not, they're gonna seem real stupid when we have all moved in in a few years. The technology will stay, but that doesn't mean they will.
Matt Levine's summary of Sam Altman's (and most of AI industry's) business plan remains accurate: "We will create God and then ask it for money."
Matt never fails to deliver
> I agree with the article - they'd better hope they are messianic super geniuses
It sounds remarkable, and I didn't believe it at first. But after more reading, I am convinced an influential part of Anthropic leadership really is something like a religious movement
Surely this is a "hook 'em with low prices, then raise them later" scheme--the biggest one yet.
I wonder where the split between self-hosted LLMs (w new more powerful LLM hardware) and hosted LLMs (big iron from an LLM-provider or at your office) will fall. Will there be a future for both? LLMs and other ML tools just with different levels of capability? etc.
This is what many have been forecasting (myself included), but the open-weights models coming out of China pose a serious problem for that approach
That's precisely the issue. LLMs are replicable technology. There is insufficient special sauce to build any kind of moat here.
No, the demand is effectively infinite at the moment and models are getting cheaper to serve. At least for a while price will go down when keeping capabilities constant.
I think "locally hosted LLMs" and "frontier LLMs hosted by frontier labs" will both end up as less than 1% of "Open Source LLMs hosted remotely by a lowest-bidder commodity compute provider"
We're one well-publicised security pop from locally hosted being a non-negotiable for some large set of buyers.
They've convinced every company in the country, including in previously security or compliance-centric environments (finance, medical, government), to hand them hyper-sensitive data.
We've already seen the leakage scenario, with the academics who found the labs scooping them on the research they were using the LLMs to assist. What's to say they-- or someone who infiltrated their infrastructure-- won't do the same thing with your propriatery business details?
I'm not sure open LLMs are a great play for commodity vendors. A lot of "lowest bidder commodity services" work because they can wildly oversell. That VPS hosting a small-biz website or 1000-messages-a-week internal email can run on a Raspberry Pi 1B, so you can pretty easily sublet a $1000 server to 50 customers at $10 per month each. Current LLM hardware plays tend to need more expensive gear and seem to be running at higher utilization rates, so the price that actually fits the underlying business might not look as appealing.
Their TAM basically presumes a good percentage of businesses will just be defacto subsidiaries of their products and they will essentially own the profit revenue stream, like a drug dealing pimp who gets their hookers addicted to crack to keep them working and inured to their loss of self suffiency.
Just look at palanetirs forward deployed engineers for how that'll start.
Even worse than that, they have to be the only messianic super geniuses.
if they were making profits now, it would set an upper limit on the number they IPO at. Losing money however? Sky's the limit.
> their godlike AI will figure that out?
Maybe it could if they stop crippling it so much
(Opus 5.5 thought a user was trying to hack it because their prompt contained the word "reasoning")
I recently tripped the safeguards by referring to a fake deployment tool. I started referring to it as a virtual one and it was fine.
Reminds me of the Amazon joke in the 90s (before they barely avoided bankruptcy): Amazon loses $5 on each sale but makes it up in volume.
I had the experience of working at a shop where the business side would use that sort of justification for all sorts of antics.
The favorite I was not involved in, was when they spent many months (at least 3, maaaaybe more than 6?) working on a thing, with verbal promises on the other side, but then the owner of the business that kept promising to sign a contract once we got to a certain point died and at that point the whole thing got quietly killed.
Again, not my project; the ones I was involved in (different product,) at least half of which did materialize but most of them never made up the volume for the pain of the work, either on the dev side or the business side [0].
[0] - My favorite was when we did an integration with Tesla. they decided to use a VIN as the primary key for their salesforce integration. So, if a customer got an inspection on a trade-in, waited, then decided to try again... oops!
The play that finally sunk my old job (headed by a serial entrepeneur) was literally selling a $49 product with a $80-per-transaction commission.
We were supposed to make it all up on being able to hard-sell the user on addons and multi-pack buys.
What never checks out (aside from most of it) is how they will raise prices to a market that shrinks due to unemployment. Can’t sell apples to people without money.
Anth revenue has probably crossed 70 billion.
The way they calculate that is, last month's revenue x 12.
The leaked numbers are old.
Does anyone have the actual doc or are we now getting articles about articles.
Here are my biggest risks for Anthropic, OpenAI and SpaceX that don't seem to get the attention they deserve:
1. What happens to your existing hardware investments when NVidia releases new hardware that allows buyers to run the same models at a fraction of the cost in energy times? IMHO performance-per-Watt is going to be the only significant metric here. Yes, your existing hardware will still work but efficiency is going to be the name of the game;
2. How are you going to deal with the threat of open weight models that can be run locally and/or or cloud GPU infrastructure? We already have relatively affordable hardware (ie Mac Studios) that can run models as large as 500B-1T parameters with 3-4 bit quantization. That's only going to get better; and
3. The only use cases for AI really are:
a) Raising prices;
b) Cutting costs, particularly labor; and
c) Military applications such as targeting and autonomous drone and missile operation.
A society where even 20% of people end up unemployed is one that catastrophically fails.
> > The more they lose, the more they win!
American zeitgeist.
Related:
Anthropic's IPO prospectus shows AI vision, surging costs
https://news.ycombinator.com/item?id=49886005
‘Anthropic is making a massive bet that AI will transform the global economy more profoundly than industrialization, electricity and the internet, according to its IPO prospectus seen by Reuters.’
I mean, if you’re gonna aim big may as well go all out. Next week OpenAI will claim to transform the global economy more than life itself.
All horseshit, of course. Still haven’t found a benefit to AI in my work or personal life. I hope the crash is a doozy.
I think AI will transform things but the Anthropic financials are crazy. I suspect you'll get your crash. It's like Webvan saying the internet would transform commerce. Yes but...