Stability AI Reportedly Ran Out of Cash To Pay Its Bills For Rented Cloud GPUs (theregister.com) 45
An anonymous reader writes: The massive GPU clusters needed to train Stability AI's popular text-to-image generation model Stable Diffusion are apparently also at least partially responsible for former CEO Emad Mostaque's downfall -- because he couldn't find a way to pay for them. According to an extensive expose citing company documents and dozens of persons familiar with the matter, it's indicated that the British model builder's extreme infrastructure costs drained its coffers, leaving the biz with just $4 million in reserve by last October. Stability rented its infrastructure from Amazon Web Services, Google Cloud Platform, and GPU-centric cloud operator CoreWeave, at a reported cost of around $99 million a year. That's on top of the $54 million in wages and operating expenses required to keep the AI upstart afloat.
What's more, it appears that a sizable portion of the cloudy resources Stability AI paid for were being given away to anyone outside the startup interested in experimenting with Stability's models. One external researcher cited in the report estimated that a now-cancelled project was provided with at least $2.5 million worth of compute over the span of four months. Stability AI's infrastructure spending was not matched by revenue or fresh funding. The startup was projected to make just $11 million in sales for the 2023 calendar year. Its financials were apparently so bad that it allegedly underpaid its July 2023 bills to AWS by $1 million and had no intention of paying its August bill for $7 million. Google Cloud and CoreWeave were also not paid in full, with debts to the pair reaching $1.6 million as of October, it's reported.
It's not clear whether those bills were ultimately paid, but it's reported that the company -- once valued at a billion dollars -- weighed delaying tax payments to the UK government rather than skimping on its American payroll and risking legal penalties. The failing was pinned on Mostaque's inability to devise and execute a viable business plan. The company also failed to land deals with clients including Canva, NightCafe, Tome, and the Singaporean government, which contemplated a custom model, the report asserts. Stability's financial predicament spiraled, eroding trust among investors, making it difficult for the generative AI darling to raise additional capital, it is claimed. According to the report, Mostaque hoped to bring in a $95 million lifeline at the end of last year, but only managed to bring in $50 million from Intel. Only $20 million of that sum was disbursed, a significant shortfall given that the processor titan has a vested interest in Stability, with the AI biz slated to be a key customer for a supercomputer powered by 4,000 of its Gaudi2 accelerators. The report goes on to mention further fundraising challenges, issues retaining employees, and copyright infringement lawsuits challenging the company's future prospects. The full expose can be read via Forbes (paywalled).
What's more, it appears that a sizable portion of the cloudy resources Stability AI paid for were being given away to anyone outside the startup interested in experimenting with Stability's models. One external researcher cited in the report estimated that a now-cancelled project was provided with at least $2.5 million worth of compute over the span of four months. Stability AI's infrastructure spending was not matched by revenue or fresh funding. The startup was projected to make just $11 million in sales for the 2023 calendar year. Its financials were apparently so bad that it allegedly underpaid its July 2023 bills to AWS by $1 million and had no intention of paying its August bill for $7 million. Google Cloud and CoreWeave were also not paid in full, with debts to the pair reaching $1.6 million as of October, it's reported.
It's not clear whether those bills were ultimately paid, but it's reported that the company -- once valued at a billion dollars -- weighed delaying tax payments to the UK government rather than skimping on its American payroll and risking legal penalties. The failing was pinned on Mostaque's inability to devise and execute a viable business plan. The company also failed to land deals with clients including Canva, NightCafe, Tome, and the Singaporean government, which contemplated a custom model, the report asserts. Stability's financial predicament spiraled, eroding trust among investors, making it difficult for the generative AI darling to raise additional capital, it is claimed. According to the report, Mostaque hoped to bring in a $95 million lifeline at the end of last year, but only managed to bring in $50 million from Intel. Only $20 million of that sum was disbursed, a significant shortfall given that the processor titan has a vested interest in Stability, with the AI biz slated to be a key customer for a supercomputer powered by 4,000 of its Gaudi2 accelerators. The report goes on to mention further fundraising challenges, issues retaining employees, and copyright infringement lawsuits challenging the company's future prospects. The full expose can be read via Forbes (paywalled).
Could have saved a bunch of money (Score:4, Interesting)
Could have saved a lot of money by buying hardware instead of renting it. Not as sexy on the books as a low capital growth plan, but works better on the long run.
Re: Could have saved a bunch of money (Score:1)
Given the ongoing super expensive GPU situation, there's likely no winning for either buying or renting.
Re: (Score:2)
Given the ongoing super expensive GPU situation, there's likely no winning for either buying or renting.
Sounds like the BitCoin Bonanza all over again.
Re: (Score:2)
yes, Emad was in crypto before AI, how did you guess?
Re: (Score:2)
Sometimes truth is stranger than fiction
Re: (Score:2)
Re: (Score:3)
It's also insurance against disruption. For example, the research on 1-trit models seems to have been upheld in followup research, meaning that we might at some point see AI hardware entirely break from traditional GPUs and switch to trinary AI accelerators, getting an order-of-magnitude increase in performance and 1 1/2 order of magnitude increase in efficiency. Any hint that this is about to happen would send resale values of AI-targeted GPUs crashing. The same thing would happen with e.g. PCNs + analo
Re: (Score:2)
Question is over what time interval.
Often renting a resource via cloud provider you end up spending more than purchase price within one year, maybe two. It's doubtful that even the most optimistic breakthroughs would be able to make it to market in volume with a viable ecosystem to utterly obsolete equipment within that time frame.
Re: (Score:2)
Re: (Score:2)
I mean, it's all probably just a big grift anyways, so may as well leave someone else holding the bag.
Re:Could have saved a bunch of money (Score:4, Insightful)
The problem is that Nvidia is picky who they let buy their chips, and will pick and choose who gets to buy how many chips. Moreover everyone knows that Emad was a scam artist, especially the people who started the business with him, because he was using money meant for covid research, to fund stable diffusion research at EleutherAI. Anyone who had enough time to actually do a background check (like Nvidia does), would have inevitably discovered that he was a bullshit artist, and had no actual ability to lead a company doing this line of work.
Re: (Score:2)
The shame of it is... someone knew what they were doing, because they got a usable product out the door. It kinda sucked at certain things, but that's generally true of every first release. So with a reasonable business plan, they might have actually benefitted from their success. Now they're broke, just somewhat later than if they'd not bothered at all.
Errr no. (Score:3)
Sorry but this is an ill thought out meme at this point. The cost of buying hardware vs renting it is a decision for a long term stable company with good finances and a well funded IT team. Datacentres of this scale cost money. LOTS OF MONEY. I think you don't really appreciate what is required to build this level of computing power. We're not talking about a company deciding to forgo a server or two to host a Exchange server instance here.
Startups cannot jump in and build a massive datacentre of this scale
Re: Errr no. (Score:2)
"Cloud" is the new "Microsoft". So many people operating on received wisdom. Cloud providers have been trying to scare companies away from IT for a decade. It's all marketing bullshit, but lots of people fall for it.
Re: (Score:2)
They could have just rented hardware - still a tiny fraction of what it costs to use the same hardware in AWS/GCP. There are umpteen datacentre vendors just itching to fill racks with hardware - if you have a budget of $99mil/year, you could throw them a few and end up with racks and racks of hardware just for you - and some goons to protect it, and some geeks to swap out stuff when it breaks.
I know it happens, and I know why it happens, but anyone spending more than a few tens of thousand a month on 'fixed
Cloud is just someone else's computer. (Score:3)
Better be ready to pay premium for it, and in perpetuity.
Wait a second (Score:3)
You can't just use the letters A and I and things will work out? You mean you have to have money to operate? That doesn't sound right.
Re:Wait a second (Score:4, Funny)
You can't just use the letters A and I and things will work out?
Yes, even Sesame Street lied to us.
Re: (Score:2)
Re: (Score:3)
Maybe it could be called synthetic brainpower or automated thinking or something like that. It might be really useful if it could be done.
Well that answers that (Score:5, Insightful)
Re: (Score:3, Insightful)
Truthfully, the practical applications for this technology are 10% clever novelty and 90% pure, utterly society-destroying evil, so I'd say we all dodged a bullet on this one, if I wasn't sure some other jackasses were just going to take up the charge instead.
Re: (Score:2)
Do you think that everything Stability has produced will just vanish if they go bankrupt?
You know that they release their model weights, right?
Re: (Score:2)
It's generally a good business strategy to release your *lower-end* models free but charge for your higher-end models via APIs, so as to build interest in your models and get free volunteer development work. There can also be a case for making non-commercial usage free but charge for commercial usage, as there's usually more money in that.
Re: (Score:2)
Will this be the end of free high quality uncensored generative AI for the masses?
Almost certainly not. It's likely open-source distributed training using people's spare computing in a manner similar to BOINC (if not within it) will pop-up for free-model enthusiasts to continue developing such models.
Bad business (Score:4, Insightful)
A poorly run business that fails does not imply a poorly run industry that will fail.
Many AI software firms are going to go bankrupt very soon, but the underlying industry will continue and strengthen.
Dot com fails were followed by Google and Amazon.
Crypto coin and exchange scams have mostly fallen away leaving Bitcoin stronger than ever.
AI gold rush companies will fall to the trailside leaving tomorrowâ(TM)s leaders to thrive.
Re: (Score:2)
The problem here is that the companies who can afford to operate at massive losses will until they are entrenched and any upstart competition has been eliminated or purchased by a big player. Not to mention the fact that if you want to compete in this space you'll also have to rent hardware from your competitors, because they're the only ones with the infrastructure big enough to scale up quickly if you grow.
Also, Bitcoin is a weird example, and I don't think it supports your point. Bitcoin is another examp
Re: (Score:1)
Crypto coin and exchange scams have mostly fallen away leaving Bitcoin stronger than ever.
Bwaahaa!! ROFL!!
Bitcoins are one bad news story away from crashing down like the 1930's stock market. Bitcoins will end up being as worthless as NFTs and CB radios.
Re: (Score:3)
LOL Bitcoin's "strength" has nothing to do with a solid underlying industry or the failures of exchanges.
You're dot.com example was good, but seriously don't ruin your otherwise well thought out point by shoehorning bitcoin into it.
Great Accounting (Score:2)
The funny thing is, a few years ago this business model would've seemed perfectly sane! Just keep throwing money on the fire!
The beginning of consolidation and failure (Score:5, Insightful)
These guys are blowing over $8 million a month just on gpu rentals plus another $4.5 million per month on staff and other standard expenses.
Crazy.
If the others are blowing money anywhere near this scale, they're all fucked.
The only winner here is the cloud companies renting them hardware.
Re: (Score:3)
If the others are blowing money anywhere near this scale, they're all fucked.
Others are blowing money far worse than this scale. ChatGPT is on track to cost OpenAI over $200m in infrastructure costs ($700k per day).
The question is how many of them have the good investor money. StabilityAI didn't have powerful partners, unlike OpenAI.
Re: (Score:2)
OpenAI is an interesting case as they're not really an independently run company anymore. They're basically a Microsoft front. Through Azure Microsoft is buying gpus not renting so when they finally fully consume OpenAI their infrastructure cost structure will be large scale traditional data center with large capex and much smaller opex, all of it written off as a line item in Microsoft's quarterly budget for running the Azure business unit. And those capex buys get written down by the finance guys, too,
Re: (Score:2)
If the others are blowing money anywhere near this scale, they're all fucked.
Others are blowing money far worse than this scale. ChatGPT is on track to cost OpenAI over $200m in infrastructure costs ($700k per day).
The question is how many of them have the good investor money. StabilityAI didn't have powerful partners, unlike OpenAI.
I'm paying OpenAI $20/month, and considering the fact that I use it speed up my software development work I think the cost is definitely worth it, and I'm sure lots of shops are buying their devs subscriptions based on the same hypothesis. I can definitely see them having a viable business model.
StabilityAI? There's surely some market for AI image generation, but I don't think it's nearly as large and the models are probably more costly. Add to that bad management will kill any business.
Re: (Score:2)
I'm paying OpenAI $20/month
Some people paid StabilityAI as well, and OpenAI are doing just as much image work as StabilityAI is. On the flip side I don't pay OpenAI anything and still use ChatGPT.
Most of OpenAI's revenue comes in terms of corporate partnerships.
This is a capital intensive business (Score:2)
A startup will have trouble competing with TSMC in the chip space without billions in financial backing. AI space is like that. Unless you are building a wrapper around ChatGPT or some other LLM you can't build a business creating a ChatGPT or Claude competitor.
$11 Million in sales, $1 Billion in valuation (Score:1)
aside from pump and dump shenanigans... what serious person can think this makes ANY sense?
SHOCKING (Score:2)
Re: (Score:3)
Developing foundational models is expensive (though the price to develop a given model drops exponentially over time, both due to hardware advances but also architectural advances). Running models, particularly those which can run on consumer-grade GPUs, is dirt cheap per query. A query that takes 5-10 seconds on a consumer-grade GPU is like playing a video game for 5-10 seconds. It's a nothing cost.
Heartfelt thanks to Stability AI (Score:1)
Apple Could Shrink (Score:2)
it's Apple Car down to chihuahua size bot.
Get the Richard Scarry design team on the phone, STAT!
https://www.richardscarry.com/... [richardscarry.com]
Re: (Score:2)
Dang it; wrong tab.