OpenAI delays GPT-4.5 release due to GPU shortage
On February 27, 2025, OpenAI announced the large-scale language model ' GPT-4.5 ', but it was revealed that its release had been delayed due to a shortage of GPUs.
OpenAI has run out of GPUs, says Sam Altman — GPT-4.5 rollout delayed due to lack of processing power | Tom's Hardware
https://www.tomshardware.com/tech-industry/artificial-intelligence/openai-has-run-out-of-gpus-says-sam-altman-gpt-4-5-rollout-delayed-due-to-lack-of-processing-power
When announcing GPT-4.5, OpenAI CEO Sam Altman posted on X (formerly Twitter) 'GPT-4.5 is ready! The good news is that GPT-4.5 is the first model that makes me feel like I'm talking to a thoughtful person. I've been surprised several times by the fact that I actually got good advice from an AI while sitting in my chair. The bad news is that GPT-4.5 is a huge and expensive model. We actually wanted to release GPT-4.5 to ChatGPT Plus and ChatGPT Pro users at the same time, but we've grown so fast that we don't have enough GPUs. We'll be adding tens of thousands of GPUs next week and making it available to ChatGPT Plus users as well. (We'll be adding hundreds of thousands of GPUs soon, but we're sure our users will use all the GPUs we've collected.)' He noted that due to the chronic GPU shortage, GPT-4.5 is only available to some users.
GPT-4.5 is ready!
— Sam Altman (@sama) February 27, 2025
good news: it is the first model that feels like talking to a thoughtful person to me. i have had several moments where i've sat back in my chair and been astonished at getting actually good advice from an AI.
bad news: it is a giant, expensive model. we…
According to OpenAI, GPT-4.5 costs $75 (about 11,000 yen) per million input tokens and $150 (about 22,600 yen) per million output tokens. In contrast, GPT-4o costs only $2.5 (about 376 yen) per million input tokens and $10 (about 1500 yen) per million output tokens.
Regarding OpenAI's chronic GPU shortage, technology media Tom's Hardware pointed out that 'this GPU shortage is clearly what is driving OpenAI to work with Broadcom to develop its own AI silicon.'
OpenAI plans to develop and manufacture its first chips in collaboration with Broadcom and TSMC, scaling back ambitions for its own manufacturing - GIGAZINE
On the other hand, he pointed out that it will take 'years' for OpenAI to be able to release its own chips, so 'for the time being, we have no choice but to rely on semiconductor manufacturers such as NVIDIA to meet demand.'
Therefore, Tom's Hardware predicts that the AI boom will continue to boost NVIDIA. In fact, it seems that the new GPUs using NVIDIA's just-announced GPU architecture ' Blackwell ' are already sold out until October 2025. In addition, he pointed out that 'NVIDIA will continue to do well for the next few years' as multiple organizations and individuals plan large-scale data center expansions.
In fact, Microsoft and OpenAI are working on the 'Stargate' project to set up an AI data center, and Elon Musk is reportedly planning to add more than a million GPUs to Colossus .
OpenAI and SoftBank announce 'Stargate' project to establish AI data center with investment of over 70 trillion yen - GIGAZINE
Related Posts:
in Note, Posted by logu_ii