OpenAI Delays GPT-4.5 Extension Due to GPU Resource Shortage

According to TomsHardware , the CEO of OpenAI just announced on the X platform (formerly Twitter) that the company is having to delay expanding the reach of GPT-4.5 due to the shortage of GPUs (graphics cards). OpenAI plans to add tens of thousands of GPUs next week to deploy this model to Plus users for $20/month. He also confirmed that hundreds of thousands of GPUs will be added in the near future.

OpenAI launches GPT-4.5 but cannot be widely deployed due to GPU limitations, which are the core elements for running AI

PHOTO: TECHPOWERUP SCREENSHOT

The GPU shortage isn’t just an OpenAI problem, it’s a reflection of the growing demand for AI hardware. To reduce its reliance on outside suppliers like Nvidia, OpenAI has partnered with Broadcom to develop its own line of AI chips. However, it will be years before the company can produce its own chips. For now, OpenAI will still need to rely on Nvidia and other suppliers to sustain its operations.

Nvidia is benefiting from the wave of AI investment, confirming that its latest Blackwell GPUs are fully booked through October. The demand is also being fueled by massive data center expansions around the world. OpenAI and Microsoft, for example, are teaming up to build a $100 billion AI supercomputer, while billionaire Elon Musk wants to expand the Colossus supercomputer in Memphis, Tennessee, to over a million GPUs.

In addition, many investors are also focusing on the field of AI infrastructure. The South Korean government has just approved a 3 GW (gigawatt) data center project, while another research group is testing data storage on the Moon . However, Microsoft CEO Satya Nadella warned that humans are building too many AI systems while the models are becoming more advanced and consuming more resources.

GPT-4.5 is a clear example of this trend. According to Open AI CEO Sam Altman, this is a "huge and expensive" model, with usage costs of up to $75 for each million input tokens and $150 for each million output tokens. Meanwhile, GPT-4o - another OpenAI model - costs only $2.50 and $10 for the corresponding levels. However, Altman said GPT-4.5 does not focus on reasoning ability or high performance on tests, but instead brings 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