Hacker News new | past | comments | ask | show | jobs | submit login

> Even if we accept the CEO’s lie that running the API is costly

compute isn't free man. it probably costs millions of dollars a year just to pay the cloud providers to support those apps. i doubt that's a lie...




Well simple logic would say that if they get $.12 per user/month, the costs to serve each user (in terms of compute) is probably below $.12 or else they would be losing money, no?


$.12 is MAU, which visit sites few times a week.

Apollo devs claim his users make 250 reqs/d, which is very different users and there is likely huge gap in costs numbers between cohorts.


MAU as a figure cannot be a dollar amount, but instead only an integer no?

Silly.

Maybe you’re thinking of ARPU?

Even if the ARPU was an order of magnitude higher, it would still be half as much as Reddit is charging per user to 3PAs.

It’s not about compute cost, because compute cost is almost definitely below revenue and the fees being charged don’t seem to have any direct link to it anyway.

Compute cost per user must generally be below ARPU, right? So even if a 3PA user cost 5x in terms of compute, at most that would be $.75 if they’re making no profit. Significantly less than the $2.50 3PAs are being charged per user per month.

tldr: reddit is charging significantly more for API access than is reasonable to expect it costs them even in lost revenue, let alone raw compute


> So even if a 3PA user cost 5x in terms of compute

in case of apollo my belief is that it will be much more than 5x

it is also not just about pure compute, but also paying salaries.


> more than 5x

doubt. the backend was open sourced to dispel this rumor.

> salaries

Salaries are a fixed cost, so not super relevant to API costs given what we already know. ARPU is ARPU. The amortized cost of a user to reddit is below or around ARPU (including R&D), unless they’re making the decision to still leave profit on the table in exchange for growth. However, given recent moves I doubt that.


> doubt. the backend was open sourced to dispel this rumor.

how its even relevant? Now that backend with unproven track record and unknown quality needs to handle petabytes of reddit data under significant load and with low latency.

you can just do the math I proposed to do above: mau visits site few times a month, apollo users generate 250 reqs every day, how we can come to just 5x difference here?

> Salaries are a fixed cost

its fixed but significant and adds to final number.


> mau visits site few times a month

To you, does MAU stand for monthly active user (singular noun?)

> 250 reqs every day

Do you know the difference between a request to the API and a site visit?


> To you, does MAU stand for monthly active user (singular noun?)

correct.

> Do you know the difference between a request to the API and a site visit?

sure, you can incorporate your view on difference into you calculations of how you could arrive to 5x estimation.


> sure, you can incorporate your view on difference into you calculations of how you could arrive to 5x estimation.

??

Regardless, Reddit wouldn’t prioritize a singular monthly active user (MAU) mentioned in this context. It’s more valuable for Reddit to have users who consistently visit, whether through Apollo or the official Reddit app. As long as advertising is effective and API pricing is appropriate, Reddit can generate profit.

From this perspective, what matters is the ratio of calls per view between different clients, not the type of user. It’s absurd to compare an inactive user on the official app with an active user on a third-party app. The issue at hand is unrelated to the cost of serving views or the potential revenue loss associated with those views.

In my opinion, the motivation behind these actions is 70% in response to low-effort low-value moderators (LLMs) and 30% due to the convenient timing to remove third-party apps (3PAs).


I asked you several times to show your math how you came to 5x. Please show if you really interested in meaningful discussion.


> So even if a 3PA user cost 5x in terms of compute ...

> I asked you several times to show your math how you came to 5x. Please show if you really interested in meaningful discussion.

Consider this: The Apollo app could generate significantly more requests (5x, 10x, 20x, ... ?) to serve the exact same content and still not justify the new pricing it is being charged. In other words, the app would have to be incredibly inefficient, which doesn't align with Occam's razor. It seems more plausible that Reddit has made allegations that are only loosely based on fact, rather than Apollo being so inefficient despite its high-quality reputation.

There is no specific mathematical calculation to present. I did not conduct benchmarks for you, if that's what you expected.

Your argument revolves around comparing an active 3PA user with an inactive Reddit user, which is fundamentally flawed and easily refutable. I focused on that aspect because it is not a valid comparison.


> The Apollo app could generate significantly more requests (5x, 10x, 20x, ... ?) to serve the exact same content and still not justify the new pricing it is being charged.

my hypothesis is that apollo users generate more requests because they are some power users like mods, or more engaged and active users compared to reddit MAU, so they bothered researching and installing 3p app.

> Your argument revolves around comparing an active 3PA user with an inactive Reddit user, which is fundamentally flawed and easily refutable.

no, it is people who brought that $0.15 per MAU number do such comparison.

I would compare daily active users, which are about 12x times less for reddit, and you would come to number close to $2/user.


> my hypothesis is that apollo users generate more requests because they are some power users like mods, or more engaged and active users compared to reddit MAU, so they bothered researching and installing 3p app.

Yes. It’s still silly to compare power users and non power users though in this context as I’ve pointed out.

I’m very confused about what you are attempting to say in the latter half of your reply.

However, to sum up, I wanted to reference this post from the apollo dev that came out today [0]. People who come to this thread in the future might benefit from the more detailed breakdowns and explanations with references provided there.

0. https://www.reddit.com/r/apolloapp/comments/14dkqrw/i_want_t...


> Yes. It’s still silly to compare power users and non power users though in this context as I’ve pointed out.

sure, and that's people like you who constantly bring those $0.12 of reddit revenue per user doing such silly comparison.


i thought they _were_ losing money?


Reddit likely could’ve taken profit in recent years, but was instead focused on growth. Reddit doubled its headcount in 2021.

There’s no reason to think that compute is a constraining cost for reddit imo. They produce what is at this point legacy tech.




Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: