OpenAI delays GPT-4.5 expansion due to GPU shortage.

According to TomsHardware , the CEO of OpenAI recently announced on platform X (formerly Twitter) that the company is having to postpone the rollout of GPT-4.5 due to a shortage of GPUs (graphics cards). OpenAI plans to add tens of thousands of GPUs next week to deploy this model for users of the Plus plan, which costs $20 per month. He also confirmed that hundreds of thousands more GPUs will be added in the near future.

OpenAI has launched GPT-4.5 but cannot yet deploy it widely due to GPU limitations, which are a core element for running AI.

PHOTO: SCREENSHOT FROM TECHPOWERUP

The GPU shortage isn't just an issue for OpenAI; it reflects the growing demand for AI hardware. To reduce its reliance on external suppliers like Nvidia, OpenAI has partnered with Broadcom to develop its own line of AI chips. However, this will take years before the company can produce chips independently. For now, OpenAI still needs to rely on Nvidia and other suppliers to stay afloat.

Nvidia is currently benefiting greatly from the wave of investment in AI. The company confirmed that its latest Blackwell GPU lineup is already fully ordered through October of this year. The increased demand is also driven by plans for large-scale data center expansions globally. For example, OpenAI and Microsoft are collaborating on a $100 billion AI supercomputer, while billionaire Elon Musk wants to expand the Colossus supercomputer in Memphis, Tennessee, to over one million GPUs.

In addition, many investors are also focusing on the AI ​​infrastructure sector. The South Korean government recently approved a 3 GW (gigawatt) data center project, while another research group is experimenting with storing data on the Moon . However, Microsoft CEO Satya Nadella warns that people are building too many AI systems as the models become more advanced and resource-intensive.

GPT-4.5 is a clear example of this trend. According to OpenAI CEO Sam Altman, it is a "huge and expensive" model, costing up to $75 per million tokens input and $150 per million tokens output. Meanwhile, GPT-40 – another OpenAI model – costs only $2.50 and $10 respectively. However, Altman stated that GPT-4.5 doesn't focus on reasoning ability or high performance in tests, but instead offers a different type of intelligence, such as "emotional intelligence".

Source: https://thanhnien.vn/openai-tri-hoan-mo-rong-gpt-45-do-thieu-nguon-gpu-185250301222736674.htm