I'm surprised I don't see a thread about their IPO yet. Their S1 came out last week, and it's better than I thought it would be, but it still leaves a lot of questions, IMO. One big one is their Q3 numbers, as many people are speculating that their revenue growth stalled out after token maxing ended and companies started looking for ways to reduce their AI spend.

We don't know the IPO price yet, but I suspect we'll have the typical IPO pop as soon as it goes live, before it drops down back. Basically, following a similar curve as SPCX. Long term (like 5 years out), I just don't see how they can make their revenue commitments. I do see a big value in AI, especially for coding and agents, but the open models can do what most people need, which will make it hard to justify all the spend. They should be able to raise enough capital, plus their current cash, to keep things going for at least 5 years, but after that I have big concerns with the company, unless they figure out how to dramatically cut their costs (or can somehow get regulations to block competition).

Also, I think all the ridiculous warnings in the S1 about the harms of AI could come back to haunt them. If there is a major security breach, and they get sued, they won't really be able to blame it on the user as they've basically said their own products are likely to cause this stuff.

I'd argue it's the opposite on the risk factors. That language is there to protect them.

Every company has to list what could go wrong before it sells stock. Drug companies say the pills might hurt you, airlines admit planes crash, banks say they might get hacked, and so on. The whole point is that if you warned people and it happens anyway, they can't claim you hid it.

And customers who get hurt wouldn't be suing over the S-1 anyway. They'd sue over their contract, and every software contract already says you're responsible for how you use the thing.

If anything, the unusually detailed warnings tell me their lawyers are trying very hard to make sure nobody can say investors weren't warned.

I guess I wasn't meaning investors suing, I was meaning the victim of something like a cyber security attack (who has no contact with Anthropic). Take the recent Hugging Face attack. OpenAI was able to say, "We didn't know our models could do that." Anthropic won't be able to say that if one of their models breaks out and hacks someone, because they say they know their models can do that. They will have to explain why, despite knowing it was a possibility, they didn't put more guard rails into it. The same if someone uses their models to help generate some sort of weapon that is used in an attack. They specifically called that out as a possibility, so if that happens, the victims will be able to sue saying that Anthropic knew that could happen, yet continued on.

I see what you're saying. I still think if they can show they saw the risks coming and were taking steps to deal with them, that's a pretty good CYA, even if those steps didn't hold up. The test is usually whether you took reasonable care, not whether it worked.

AustinAg2K said:

---

I'm surprised I don't see a thread about their IPO yet. Their S1 came out last week, and it's better than I thought it would be, but it still leaves a lot of questions, IMO. One big one is their Q3 numbers, as many people are speculating that their revenue growth stalled out after token maxing ended and companies started looking for ways to reduce their AI spend.

We don't know the IPO price yet, but I suspect we'll have the typical IPO pop as soon as it goes live, before it drops down back. Basically, following a similar curve as SPCX. Long term (like 5 years out), I just don't see how they can make their revenue commitments. I do see a big value in AI, especially for coding and agents, but the open models can do what most people need, which will make it hard to justify all the spend. They should be able to raise enough capital, plus their current cash, to keep things going for at least 5 years, but after that I have big concerns with the company, unless they figure out how to dramatically cut their costs (or can somehow get regulations to block competition).

Also, I think all the ridiculous warnings in the S1 about the harms of AI could come back to haunt them. If there is a major security breach, and they get sued, they won't really be able to blame it on the user as they've basically said their own products are likely to cause this stuff.

---

This is why Alex Karp thinks the government is ultimately going to take this over. The liability is too great. They've arguably already lost control of it.

I'm no way saying that the government will somehow do a better job of this but the government can absorb liability that the private industry can't.

I'm staying away from it as an individual investment until we know what's going to happen although I assume I will end up with some in a fund, regardless.

Seems the better avenue would be government limiting their liability?

A takeover would be terrible.

Facebook, 'x', etc. all had their liability limited if I recall correctly.

Red Fishing Ag93 said:

---

Seems the better avenue would be government limiting their liability?

A takeover would be terrible.

Facebook, 'x', etc. all had their liability limited if I recall correctly.

---

Not sure about the social media sites but their liability doesn't seem too limited with the massive lawsuits Meta is currently losing. And while we can certainly argue the intangible destruction of social media it's a different deal compared to AI which can destroy something immediately - like draining bank accounts, taking stuff offline, etc.

Section 230 of the Communications Decency Act

This is the main U.S. federal law that limits liability for social media platforms and other online services for content posted by users.

Red Fishing Ag93 said:

---

Section 230 of the Communications Decency Act

This is the main U.S. federal law that limits liability for social media platforms and other online services for content posted by users.

---

Yeah, that's still different. Meta is losing billions in lawsuits right now for crap they did as a company; not what users did.

AI's potential for destruction is far more immediate and impactful than social media.

I'll give it to Anthropic, chefs kiss on getting their financials IPO-ready. I especially like how they let the good stuff sit out for a week before showing Q3 growth declined and almost $900Bn in future compute commitments with half due by 2030. I don't want to hear **** about operating leverage and inference efficiency. This company will have to step up to borderline $Trns in fixed costs (with generating a fraction of it) before any true, steady-state operating leverage. All while there is future pricing pressure. This **** is a fairy tale where AI will actually have to replace white collar workers and cybersecurity goes berserk with spend and hardware / robotics fly off the shelf.

AustinAg2K said:

---

I'm surprised I don't see a thread about their IPO yet. Their S1 came out last week, and it's better than I thought it would be, but it still leaves a lot of questions, IMO. One big one is their Q3 numbers, as many people are speculating that their revenue growth stalled out after token maxing ended and companies started looking for ways to reduce their AI spend.

We don't know the IPO price yet, but I suspect we'll have the typical IPO pop as soon as it goes live, before it drops down back. Basically, following a similar curve as SPCX. Long term (like 5 years out), I just don't see how they can make their revenue commitments. I do see a big value in AI, especially for coding and agents, but the open models can do what most people need, which will make it hard to justify all the spend. They should be able to raise enough capital, plus their current cash, to keep things going for at least 5 years, but after that I have big concerns with the company, unless they figure out how to dramatically cut their costs (or can somehow get regulations to block competition).

Also, I think all the ridiculous warnings in the S1 about the harms of AI could come back to haunt them. If there is a major security breach, and they get sued, they won't really be able to blame it on the user as they've basically said their own products are likely to cause this stuff.

---

When people comment about the spend, I'm curious their actual perspective. I feel like this is a YouTube comment that gets thrown around a lot. What do you mean precisely?

The average user doesn't need to use fable, that's true. I haven't read their S1, so I would be curious to know the breakdown between the more expensive models and basic ones.

Is a lot of their revenue made up of people using fable to edit their word doc? If so, ya, revenue will go down. If the main users of the latest and greatest most expensive models still need the best and it correlates to value in their business they will pay for it, even if the price increases. The technology is that good.

The main normal models for every day business use are immaterial right now from a cost standpoint. We are a real estate developer, team of 17. We probably will spend 10,000 for the year with Anthropic. There is no going back.

If they keep the price stable and there isn't a clear cut better alternative, then we will happily pay for a long time. Probably would if the price was double.

Prices will go up across the board. We're in a sweet spot era right now where access to these models is far cheaper than it will ever be.

Proposition Joe said:

---

Prices will go up across the board. We're in a sweet spot era right now where access to these models is far cheaper than it will ever be.

---

Are we sure about this?

I mean, they're clearly heavily subsidized right now. On an Anthropic ultra plan currently, you're getting like $5000 or $6000 worth of compute a month for $200 bucks. That's not sustainable long term.

mass adoption will likely apply lower-price pressure over time

For "what planet is the coldest?" queries, sure.

But for "take this API documentation, read it over and develop me an app that does X, Y and Z" ? That ain't gonna be a $50/month plan. Do your devving now.

FriendlyAg said:

---

Proposition Joe said:

---

Prices will go up across the board. We're in a sweet spot era right now where access to these models is far cheaper than it will ever be.

---

Are we sure about this?

---

Well, Elon just did it for Grok yesterday.

I think the only reason I question is that we are assuming the status quo in engineering. If they are able to figure out how to reduce the LLM ro specialized models for specific tasks, does that keep cost down? How does GPU/Hardware improvements factor in?

I'm by no means an expert.

Our whole team uses Claude, we use skills, and use a lot of the stuff that creates excels, PPTs, PDFs, distills large real estate related data sets, etc. we would be happy to pay 3-4x what we are now, but it feels like competition for this less intense use case will keep the price low-ish?

I would assume the people using the most compute are anyone that is creating lots of images, videos, large development, etc, but I could be totally off.

Well I'd also be happy to keep paying less because I don't need the advancements instantly.

If someone comes along and can offer a model that just does excel stuff then I could cut a more expensive license and would go with the paired down model for 70-80% of work.

Basically what I'm suggesting is, of course if there is a monopoly/duopoly on the LLMs with the most services, this is like cable, but then cord cutting options will come along where I can pick a choose services that are more efficient from a cost and compute stand point.

It's like other softwares, that's going to become a thin business because everyone's going to be able to eventually code up alternatives without the massive corporate cost structure behind it and will be able to undercut the market.

To be fair, I also 100% acknowledged with something like Uber that they did low cost for a long time until they had adoption and then they changed their price. Thats fair.

I'm just simply asking the question. I think there are a lot of variables in play.

At some point they will want to make money on you. Regardless if that's tighter quotas, higher token costs, higher subscriptions, or another mechanism. Right now, they are not making money on you. The part that I'm most curious about is how much of Anthropic's compute has been subsidized by their equity.

Is there an index or aggregated price somewhere I can look at the average price of compute? Or, forgive my ignorance, is that something that could even be done?

Edit: I can see compute turning into a commodity like input like the price of oil. If it's able to be distilled down to a single number.

Capt. J. Yossarian said:

---

Is there an index or aggregated price somewhere I can look at the average price of compute? Or, forgive my ignorance, is that something that could even be done?

Edit: I can see compute turning into a commodity like input like the price of oil. If it's able to be distilled down to a single number.

---

I can't wait for compute futures