Sam Altman: OpenAI's GPU is running out, GPT-4.5 release will be phased in
Mar 03,2025
OpenAI has recently released its latest model GPT-4.5. However, it is currently limited to Pro users who pay $200 per month. OpenAI CEO Sam Altman stated on February 28th that the release of the model had to be postponed because "... (OpenAI) is developing at a fast pace and the GPU is already running out Then he added, 'We will add tens of thousands of GPUs next week and promote them to Plus.' This means that users only need to pay $20 to OpenAI to access its state-of-the-art models thereafter.
According to rumors, the shortage of GPUs will prompt OpenAI to collaborate with Broadcom to develop its own AI chips. But as the company needs several years to release its own chips, it currently has to rely on Nvidia and other suppliers to meet its needs.
This indicates that Nvidia is in a favorable position, with the chip manufacturer stating that its latest Blackwell GPU will not be able to deliver orders until October of this year. As institutions and individuals plan to expand their data centers on a large scale, Nvidia will continue to be successful in the coming years. For example, OpenAI and Microsoft are developing a large-scale AI supercomputer that costs $100 billion, while Elon Musk hopes to expand his Colossus supercomputer in Memphis, Tennessee to over 1 million GPUs.
Other investors have also joined the competition for data centers, with a 3GW facility approved by the South Korean government and another team attempting to store data on the moon. However, all these expansions of AI infrastructure have led Microsoft CEO Satya Nadella to argue that even as AI models become more advanced and require more computing power, the construction of AI systems has become excessive.
OpenAI's GPT-4.5 is a good example. Sam Altman said, 'This is a huge and expensive model,' with GPT-4.5 costing $75 per million input tokens and $150 per million output tokens. In contrast, the cost of GPT-4o per million input tokens is only $2.50, and the cost per million output tokens is only $10. Despite its outrageously high price, Sam Altman stated that it is "not an inference model and will not achieve overwhelming results in benchmark tests". However, he added, "This is a different type of intelligence, it has a magic that I have never felt before
According to rumors, the shortage of GPUs will prompt OpenAI to collaborate with Broadcom to develop its own AI chips. But as the company needs several years to release its own chips, it currently has to rely on Nvidia and other suppliers to meet its needs.
This indicates that Nvidia is in a favorable position, with the chip manufacturer stating that its latest Blackwell GPU will not be able to deliver orders until October of this year. As institutions and individuals plan to expand their data centers on a large scale, Nvidia will continue to be successful in the coming years. For example, OpenAI and Microsoft are developing a large-scale AI supercomputer that costs $100 billion, while Elon Musk hopes to expand his Colossus supercomputer in Memphis, Tennessee to over 1 million GPUs.
Other investors have also joined the competition for data centers, with a 3GW facility approved by the South Korean government and another team attempting to store data on the moon. However, all these expansions of AI infrastructure have led Microsoft CEO Satya Nadella to argue that even as AI models become more advanced and require more computing power, the construction of AI systems has become excessive.
OpenAI's GPT-4.5 is a good example. Sam Altman said, 'This is a huge and expensive model,' with GPT-4.5 costing $75 per million input tokens and $150 per million output tokens. In contrast, the cost of GPT-4o per million input tokens is only $2.50, and the cost per million output tokens is only $10. Despite its outrageously high price, Sam Altman stated that it is "not an inference model and will not achieve overwhelming results in benchmark tests". However, he added, "This is a different type of intelligence, it has a magic that I have never felt before