Hacker News new | ask | show | jobs
Stability AI reportedly ran out of cash to pay its cloud GPU bills (theregister.com)
91 points by obelix150 804 days ago
5 comments

This makes me worry that the 20 bux a month I pay for GPT4 could end up being more like 200 bux on the backend.

When they stop running gpt4 as a loss leader, I'll probably lose access.

I used to think Adobe was in trouble with Stable Diffusion and Midjourney.

Now I realize that Adobe knows how to make money. They'll just add the AI models to Photoshop, Premier, etc and charge an extra $50/month and voila.

However, Stable Diffusion and Midjourney does not have a solid vehicle to making money.

Maybe in the long term, no one will care about using Photoshop anymore. But we're not even close to there yet. AI models seem to make Photoshop even more valuable.

GPT4 is different. OpenAI has made monetization its goal from the start by charging $20/month for GPT4 in ChatGPT and charging loads of money for the API. Also, they have the backing of Microsoft.

I’ve attended two local Adobe Create Now events where they demoed their new AI functionality. It’s so well integrated- I’m impressed despite usually disliking them. The key point I got from both events was their emphasis on training exclusively on content they already had licensed through adobe stock so you wouldn’t need to worry about accidentally infringing so much
Yeah this is my experience as well, gen AI is so well integrated into Photoshop. It's the only software I've used to far where AI integration actually improved the experience. Removing things from pictures used to be a painful process where I'd use the spot heal tool as much as possible then touch up the bad bits manually. Now it's a selection and a sentence, and I have multiple variations I could never have hoped to replicate manually. Same with adding things into photos. Both generating them and blending them in.
What evidence do you see indicating that the 20/month is actually profitable? You make the claim, but you have not backed it up.

They could easily be eating a loss on GPT+ and the GPT4 AI. They have billions in runway to do that. I have not seen any reports about their expenses or profitability. Perhaps you can link me to some hard facts.

Until I see real reason to think otherwise, I'll assume there is a reasonable risk of a price hike

Right now they could hike prices, and I'd have nowhere else to go, because nothing else competes with GPT4.

Anthropic’s Claude Opus gives GPT4 a run for its money.
Which is why things like local LLMs are important.
The good local LLMs are also loss leaders, a company spends a lot of money training a model then gives it away for free in the hopes they can eventually pivot to an actual business model. The work being done with local LLMs seems to take for granted that Mistral and co will just keep shoveling investor money into the furnaces forever to keep the free LLMs coming, and there is no contingency plan for when that money runs out.

Likewise local image generation is almost entirely dependent on the charity of one company, and per the OP that company is on the verge of going broke.

I am not worried about GPT4 like this but GPT6 or GPT7.

On the other hand, I think of how at the individual level if they price me out then they don't have a mass scale product.

I suppose I worry far more about this with Sora too.

I have also spent about $300 on Midjourney in the last 18 months but I don't know how to spend money on Stability.

There will be an enshittified ad supported tier. You’ll never know when the model is hallucinating and when it’s actively trying to manipulate you. It’ll be great.
The real fun begins when the manipulations are driven by hallucinations.
Thus we'll circle right back to affiliate spam blogs
$11M in revenue on $153M in cloud spend + salaries...

Is the AI bubble going to pop hard soon?

>Is the AI bubble going to pop hard soon?

No. I think it's going to get way bigger. When it pops, it will still be bigger than February 2024.

Right now, the companies making money are Microsoft, OpenAI, AI chip makers. Google might join soon if they get their act together. Apple as well.

Basically, only existing large tech companies + OpenAI are making money so far.

However, I expect newer AI-only startups, to start making money in the near future. For example, I can see a company like Devin AI get loads of revenue very quick such as when ChatGPT launched.

As for open source AI companies like Stability, they're in trouble.

Unfortunately, no one has figured out how to make money with open source AI. It used to be that you'd open source your tool, get adoption, then offer a cloud. With open source AI so far, this hasn't come true for whatever reason. Perhaps we're just early.

AI is not a bubble. Anymore than iphone / the internet was a bubble.
You have no way of knowing that. Remember, the parent poster is referring to the current ridiculous hype + push of current AI. Current AI models were trained on copyrighted material. Lawsuits are taking place.

Many different members of the US government have indicated that just because AI is new and fancy, does not mean companies get to violate established law, including copyright. The head of the FTC was the most recent person to state this.

OpenAI could very likely be sued out of existence. When that happens, all who invested in them could lose billions. The US government very likely won't intervene because there are multiple lessons to be learned from this.

Courts are there to enforce the law, regardless of how "amazing" a technology may appear.

To add something I forgot to say, yes, AI as a concept is here to stay. The current hype surrounding existing companies will not, however.
The internet was a bubble too. Remember the dot com era?

Just because the tech is legit doesn’t mean it can’t have a bubble.

> Anymore than iphone / the internet was a bubble.

By that do you mean they're technology platforms that outlive their obvious peak hype investment dice rolls?

As a broad advancement that's true, but on individual companies there's going to be a ton that don't make it (which is the case for all companies, but for AI there's a significant number of new ones to fail within the next 10 years).
Do you know what a bubble is?
Except that they're not talking technology, they're talking economics. It may be significantly to earn a profit on selling an iPhone than selling continued AI access at a reasonable price.
So where's the revenue?
At companies with a business plan like Midjourney, which has 200M in revenue, 40 employees, and no investors.
MS is pushing hard with copilot for $30/user/month (annual commitment... I see they've learned from Adobe, sigh).
OpenAI, Microsoft, Google, and other companies who have existing products and added AI, such as Notion and Adobe.
Then where's the use case?
and yet there was a dotcom bubble.
I think it will just shift away from relying on cloud providers for the underlying infrastructure.
If the big AI companies go under there could be a lot of cheap second hand H100s on the market.
We’re very far away from H100 firesales. Pretty sure capacity is booked out for years in advance still so even if it is a bubble and pops they’re unlikely to feature on eBay
Yeah. Looks like the gods demand one last AI winter.
Rookie numbers. How much did Uber lose? Wirecard?
Wirecard was a fraud, not a bubble.
*In a ZIRP environment.
Related:

StabilityAI chief resigns, raising doubts about AI start-up’s future

https://news.ycombinator.com/item?id=39894334

Super easy to rack up a hefty bill when you're training AI.

The company I work for is trying to actually tackle this. We all have an AI background and were super sick of the costs. Here's the concept we're building: https://www.trybuild.ai/

In the meantime, we're solving this somewhat manually. Hit me up if you need compute, we can usually find a cheaper alt. https://www.trybuild.ai/m

Another one of those “what good reason was there for flagging this” moments.
Because Forbes is a clickbait domain.