They are talking about Meta, AWS, Microsoft, Google, Nvidia and so on.
Although Anthropic supposedly has a good margin on each token, so the inference is substantially less expensive than what they charge customers. Training is eating into that.
Although Anthropic supposedly has a good margin on each token, so the inference is substantially less expensive than what they charge customers. Training is eating into that.
I am still to see any evidence of this being true.
The actual data at hand suggests otherwise, with Anthropic's losses increasing with more usage, and they relying on shenanigans to give an appearance of being healthy. They also need to raise money like crazy and seem desperate to offload this hot potato through an IPO. Not really the sort of thing that screams "sustainable" to me.
Such as having a "profitable" quarter on one of their funny metrics when they were not paying for compute, or leaking completely obscure ARR, "EBTIT", etc.
What data? To my knowledge the only data point that exists is a 2025 document, which is completely irrelevant considering how massively they hiked the API token price.
To my knowledge, the subscriptions are the only part that may still be profiting from vc subventions. And if not, then their API margins are gigantic
- The suspiciously high amount of money they have to keep raising at a rapid pace
- The fact that they have to play around with silly metrics and leak bullshit to pretend they are healthy
- Their desperation for an IPO.
I feel kind of silly having to repeat myself, it was all there in my previous reply of you bothered to read. I point at that whenever people spread this bullshit of Anthropic being "profitable" especially when it comes to subscriptions. Everything points at it being ruinously unprofitable actually.
Profitability only matters if we consider that they are grifters willing to cash in as quickly as possible before this jenga tower falls apart. It matters less if the interpretation is that the fucking weirdos at Anthropic are death cultists that want to conjure their death god into existence.
Anthropic can be profitable and massively overvalued, AND deeply into debt from running too fast and loose with their money in previous years.
That situation would have them behave exactly the same: hurrying to get the IPO done while the stocks are still skyhigh to get over the debt burden and then buckle up for the profitability to weather them through coming correction...
It is WELL known that the margins on inference are extreme across the board, something like 95%.
It is an outlandish claim, I expect outlandish evidence too.
Genuine question: why are people still saying the plans are heavily subsidised?
The only real comparison point that matters for the argument of subsidization is how much money it cost costs Anthropic to serve you those tokens.
OpenAI released an AI video app to the masses for free, made a bunch of promises, and then just shut it down.
So there’s a growing body of evidence that while neither of these are in great financial shape, Anthropic looks to be the more disciplined of the two.
Video is costly and wasn't going to get them to RSI like math and coding will. While I enjoyed Sora and it showed just how far AI has come, their decision makes sense.
And to think just how amazing agents are at coding compared to even 6 months ago... I'm not complaining!
When you max out your Claude Max plan every billing period, are you consuming more tokens than you could realistically pay for with the same money spent on API inference? Definitely! This works because many Claude Max subscribers don't use it to the max (and you probably aren't using it to the max every single time either), so overall subscriptions might still be profitable. Technically, the subscribers subsidize each other, but this is entirely sustainable (see any other subscription model that is based on its consumers not maximizing their usage, e.g. gyms).
What people mean (or want to say) when they say inference is subsidized is that it is subsidized in an unsustainable way (e.g. with investments that will not last forever - if this is true at scale it should worry us!). What happens now is that they take the fact that an individual subscription used to the max by itself is a financial drain for the company offering it, attach the "subsidized" term to it, and infer from that a bunch of stuff that is usually associated with the term (such as unsustainability) but doesn't really apply here.
Some people then assume that this price means the subscriptions are deeply unprofitable at least for users that make heavy use of their usage limit.
If you look at estimates based on serving Kimi K3, margins at API pricing must be well north of 95%.
And that's assuming frontier labs don't serve inference more efficiently than Kimi K3.
I believe most people dramatically overestimate the inference costs.
It doesn't help that in many discussions about profitability, people mix many different metrics, including other costs like R&D.