Semiwiki on LinkedIn Semiwiki on Facebook Semiwiki on Twitter Register / Log in 
Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/threads/openai-ceo-sam-altman-says-the-company-is-out-of-gpus.22178/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021770
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

OpenAI CEO Sam Altman says the company is 'out of GPUs'

Daniel Nenni

Daniel Nenni

Admin
Staff member
Sam Altman
Image Credits:Eugene Gologursky / Getty Images for the New York Times
  • Sam Altman says the company is ‘out of GPUs’
OpenAI CEO Sam Altman said that the company was forced to stagger the rollout of its newest model, GPT-4.5, because OpenAI is “out of GPUs.”
In a post on X, Altman said that GPT-4.5, which he described as “giant” and “expensive,” will require “tens of thousands” more GPUs before additional ChatGPT users can gain access. GPT-4.5 will come first to subscribers to ChatGPT Pro starting Thursday, followed by ChatGPT Plus customers next week.

Perhaps in part due to its enormous size, GPT-4.5 is wildly expensive. OpenAI is charging $75 per million tokens (~750,000 words) fed into the model and $150 per million tokens generated by the model. That’s 30x the input cost and 15x the output cost of OpenAI’s workhorse GPT-4o model.

“We’ve been growing a lot and are out of GPUs,” Altman wrote. “We will add tens of thousands of GPUs next week and roll it out to the Plus tier then … This isn’t how we want to operate, but it’s hard to perfectly predict growth surges that lead to GPU shortages.”

Altman has previously said that a lack of computing capacity is delaying the company’s products. OpenAI hopes to combat this in the coming years by developing its own AI chips and by building a massive network of data centers.

techcrunch.com

OpenAI CEO Sam Altman says the company is 'out of GPUs' | TechCrunch

OpenAI CEO Sam Altman said that the company was forced to stagger the rollout of its newest model, GPT-4.5, because OpenAI is 'out of GPUs.'
techcrunch.com techcrunch.com
 
As Huang said the fact the chip gets designed doesn't mean it gets deployed. It's good to have some backup from their perspective but it's hard to build anything comparable to the cutting edge nvidia stuff.
 
Jumper said:
As Huang said the fact the chip gets designed doesn't mean it gets deployed. It's good to have some backup from their perspective but it's hard to build anything comparable to the cutting edge nvidia stuff.
Click to expand...
Building close to Nvidia HW is not the challenge the challenge is SW AMD has better HW Spec than NV it's the Sw that matters
 
siliconbruh999 said:
Building close to Nvidia HW is not the challenge the challenge is SW AMD has better HW Spec than NV it's the Sw that matters
Click to expand...
That's oversimplification. Nvidia also has advantage in networking, strong partnerships with hardware manufacturers etc. Yes software is big part of their success but they still are also behind in some hardware aspects.
 
Daniel Nenni said:
Good news for TSMC since they own the GPU business. The AI datacenter buildout continues!
Click to expand...

But if it doesn't the great bet on making huge, expensive, low volume dies will backfire on TSMC spectacularly.

When TSMC was a high volume, mass market oriented follower node foundry, they were able to weather market cyclicity very easily. When they only get 3-4 customers for a very very very expensive cutting edge node, the departure of just a few of them will be devastating.
 
Daniel Nenni said:
Sam Altman
Image Credits:Eugene Gologursky / Getty Images for the New York Times
  • Sam Altman says the company is ‘out of GPUs’
OpenAI CEO Sam Altman said that the company was forced to stagger the rollout of its newest model, GPT-4.5, because OpenAI is “out of GPUs.”
In a post on X, Altman said that GPT-4.5, which he described as “giant” and “expensive,” will require “tens of thousands” more GPUs before additional ChatGPT users can gain access. GPT-4.5 will come first to subscribers to ChatGPT Pro starting Thursday, followed by ChatGPT Plus customers next week.

Perhaps in part due to its enormous size, GPT-4.5 is wildly expensive. OpenAI is charging $75 per million tokens (~750,000 words) fed into the model and $150 per million tokens generated by the model. That’s 30x the input cost and 15x the output cost of OpenAI’s workhorse GPT-4o model.

“We’ve been growing a lot and are out of GPUs,” Altman wrote. “We will add tens of thousands of GPUs next week and roll it out to the Plus tier then … This isn’t how we want to operate, but it’s hard to perfectly predict growth surges that lead to GPU shortages.”

Altman has previously said that a lack of computing capacity is delaying the company’s products. OpenAI hopes to combat this in the coming years by developing its own AI chips and by building a massive network of data centers.

techcrunch.com

OpenAI CEO Sam Altman says the company is 'out of GPUs' | TechCrunch

OpenAI CEO Sam Altman said that the company was forced to stagger the rollout of its newest model, GPT-4.5, because OpenAI is 'out of GPUs.'
techcrunch.com techcrunch.com
Click to expand...
The big problem with OAI is that people are not excited and not impressed with these big, expensive models just rolled out now.
 
kevin01 said:
The big problem with OAI is that people are not excited and not impressed with these big, expensive models just rolled out now.
Click to expand...
It all seemed to me really dubious from back when gpt-3 or 3.5 was announced I heard that Altman said it would get exponentially better. Even then it was clear A: that even these smaller models where eating lot of resources and B: even though it seems internet is infinite with some automated crawling they would in a few years run out of meaningful quantities of new data.
 
Paul2 said:
But if it doesn't the great bet on making huge, expensive, low volume dies will backfire on TSMC spectacularly.

When TSMC was a high volume, mass market oriented follower node foundry, they were able to weather market cyclicity very easily. When they only get 3-4 customers for a very very very expensive cutting edge node, the departure of just a few of them will be devastating.
Click to expand...
But those 3-4 customers don’t have any other options other than TSMC so where would they go?
 
Well that's similar situation to Nvidia if something happened and AI market would go bust or slow down significantly it would cause a lot of headaches but events like that are pretty rare. I'm sure TSMC is looking on diversification into the future when they catch up with demand.
Paul2 said:
They would go bankrupt
Click to expand...
 
Jumper said:
I'm sure TSMC is looking on diversification into the future when they catch up with demand.
Click to expand...

I am sure their internal economists know this will not go on forever, and I am also sure they are pissing molten lead now from the realisation that to bring these super expensive nodes to mass market, they had to got the entirely different route for the last 10 years.

They have basically accepted that with way, way more expensive dies, they can tolerate issues like stochastic defects being left unsolved, and latest node steppers being slow as snails.

They cannot make money making LED blinkers on 3nm.
 
Paul2 said:
I am sure their internal economists know this will not go on forever, and I am also sure they are pissing molten lead now from the realisation that to bring these super expensive nodes to mass market, they had to got the entirely different route for the last 10 years.

They have basically accepted that with way, way more expensive dies, they can tolerate issues like stochastic defects being left unsolved, and latest node steppers being slow as snails.

They cannot make money making LED blinkers on 3nm.
Click to expand...
But like the even if the demand slows down there are still lot of markets that will use even the more expensive nodes. Nvidia will surely shift their production to gaming and other segments for example. AMD has also lot of other segments. Apple can eat lot of their expensive silicon. It's the same as when Nvidia is investing so much into ai. Like yeah they are very exposed but they'd be crazy not to try to eat up as much as possible.
 
hist78 said:
Can you please be more specific about it? Thanks.
Click to expand...
Deepseek did some clever GPU optimization tricks to those nerfed H800 to perform at the level of the original H100.

Imagine you put the same tricks in H100 or Blackwell to make them run even faster.
 
horace said:
Deepseek did some clever GPU optimization tricks to those nerfed H800 to perform at the level of the original H100.

Imagine you put the same tricks in H100 or Blackwell to make them run even faster.
Click to expand...
Some concrete examples/papers?

I have not seen anything they did what other people didn't.
 
Jozo035 said:
Some concrete examples/papers?

I have not seen anything they did what other people didn't.
Click to expand...
Well not only that, you can find their research on internet. According to what I read and heard is they have very good engineers that can do optimizations at level or even below cuda. These are deeper optimizations that will be beneficial for whole industry.
 
You must log in or register to reply here.
Back
Top