redwattlebird comments on ChatGPT is down worldwide, conversations disappeared for users

[–] redwattlebird@lemmings.world 4 points 9 months ago (4 children)

These companies have BILLIONS in revenue and millions of customers, and you're saying very few want to pay...

Yep, I am. Just follow the money. Here's an example:

https://www.theregister.com/2025/10/29/microsoft_earnings_q1_26_openai_loss/

not saying this is an easy problem to solve, but you're making it sound no one wants it and they can never do it.

... That's all in your head, mate. I never said that nor did I imply it.

What I am implying is that the uptake is so small compared to the investment that it is unlikely to turn a profit.

If OpenAI can build a datacenter that re-uses all it's heat for example to heat a hospital nearby, that's another step towards reaching profitability.

😐

I've worked in the building industry for over 20 years. This is simply not feasible both from a material standpoint and physics standpoint.

I know it's an example, but this kind of rhetoric is exactly the kind of wishful thinking that I see in so many people who want LLMs to be a main staple of our everyday lives. Scratch the surface and it's all just fantasy.

permalink

fedilink

source

parent

hideshow 4 child comments

[–] NotMyOldRedditName@lemmy.world -3 points 9 months ago* (last edited 9 months ago) (3 children)

You > the trends show that very few want to pay for this service.

Me > These companies have BILLIONS in revenue and millions of customers, and you’re saying very few want to pay

Me > ... but you’re making it sound no one wants it

You > … That’s all in your head, mate. I never said that nor did I imply it.

Pretty sure it's not all in my head.

The heat example was just one small example of things these large data centers (not just AI ones) can do to help lower costs, and they are a real thing that are being considered. It's not a solution to their power hungry needs, but it is a small step forward on how we can do things better.

https://www.bbc.com/news/articles/cew4080092eo

1Energy said 100 gigawatt hours of energy would be generated through the network each year, equivalent to the heat needed for 20,000 homes.

Edit: Another that is in use: https://www.itbrew.com/stories/2024/07/17/inside-the-data-center-that-heats-up-a-hospital-in-vienna-austria

This system “allows us to cover between 50% and 70% of the hospital’s heating demand, and save up to 4,000 tons of CO2 per year,” he said, also noting that “there are virtually no heat losses” since “the connecting pipe is quite short.”

permalink

fedilink

source

parent

hideshow 3 child comments

[–] redwattlebird@lemmings.world 1 point 9 months ago* (2 children)

And how, pray tell, will doing all of that return a profit?

I'm from Australia, so I can only speak to the Australian climate and industry. I can confidently say that the model shown in Vienna is not feasible in our country. We simply don't have much use for excess heat and we are highly susceptible to droughts. DCs use a lot of water to cool down and having these all over the country for private enterprise is bonkers. So, that's instantly a market that isn't profitable. Furthermore, it's not feasible to build a pipe and re-route the heat across large distances with minimal heat loss.

However, even when or if they implement this throughout all of Austria, it won't return a profit (which is what I thought your attachment was here, not the feasibility. We are talking about profitability, right?). This project cost $3.5m Euro and partially funded by tax. It's not a great example of profitability but a good example of sustainability measures.

Also, reading comprehension assistance: not feasible != Impossible.

permalink

fedilink

source

parent

hideshow 2 child comments

[–] NotMyOldRedditName@lemmy.world 0 points 9 months ago* (last edited 9 months ago) (1 child)

Australia isn't the greatest spot to run a data centre in general in terms of heat, but I do understand the need for sovereign data centres, so this obviously can't work everywhere.

What makes you think $3.5 million can't be profitable? A mid sized hospitals heating bill can get into the many hundreds of thousands or into the millions even. Especially if it's in a colder environment. A 5-6 year payback on that wouldn't be terrible and would be worth an upfront investment. Even a 10 year payback isn't terrible.

These colder locations are the ideal locations for the data centres in the first place because they generally want a cooler climate to begin with, so they will gravitate to them when possible.

Edit: And if you build a data centre with this ability to recoup heat, you could start building further commercial things in the area and keep the heat redistribution very close. You don't need to travel very long distances. You do need to put some thought into where they go through and whats around or will be built around.

permalink

fedilink

source

parent

hideshow 1 child comment

[–] redwattlebird@lemmings.world 2 points 9 months ago

Ok. We're deviating off the point of LLM profitability here and have driven this conversation off into the weeds. So I'll make this one last comment, and then I'm done. This debate has been interesting but exhausting.

Final counterpoints:

$3.5mil is the cost of the connection footed by the energy provider and tax payer, and provides no ROI to investors like NVIDIA, hence no profit to LLM and "AI" in general.
As far as I can tell, the biggest method of external income for LLM companies are subscriptions and there is simply not enough uptake in subscriptions to get ROI, so they try to force consumers to use it which ends up pushing away your customer base since you're taking away their power of choice.
For them to obtain ROI, literally the entire planet needs to use it which isn't feasible because, as a consumer, you need income to consume and the larger driver of investment into LLMs is to reduce the cost of labour.

LLMs have long since gone beyond the scope of interesting science project to something driven by pure parasitic greed.

permalink

fedilink

source

parent