由 zq4034698 
A Closer Look at OpenAI’s Hardware Woes and GPT-4.5’s Market Impact
In the world of artificial intelligence, OpenAI has been a dominant force. But lately, it’s facing a big problem – a shortage of GPUs. This has led to the 分批上线 (phased rollout) of its new GPT-4.5 model.
Sam Altman, the CEO of OpenAI, said GPT-4.5 is large and costly. Because there aren’t enough GPUs, the model can’t be launched to all ChatGPT users at once. Instead, it will be released to different user groups in stages.
On February 28, Altman took to social media to share the issue. He said the company was running out of GPUs due to rapid business growth. As a result, they had to release GPT-4.5 gradually.
Starting this Thursday, only ChatGPT Pro users can access GPT-4.5. ChatGPT Plus users will get access next week. Altman said they’d add tens of thousands of GPUs next week to meet the demand.
OpenAI’s hardware problems didn’t start recently. Last October, during a Reddit AMA, Altman admitted that limited computing power was a major obstacle to product launches. He said they had to carefully allocate limited computing resources because AI models were getting more and more complex.
OpenAI has been working hard to solve this problem. Reports say it’s focusing on designing its own AI chips. Since last year, it has been collaborating with Broadcom. The new chips might be available as early as 2026.
Another thing that has caught people’s attention is GPT-4.5’s pricing. OpenAI charges $75 for every million input tokens (about 750,000 words) and $150 for every million output tokens. Compared to GPT-4, the input cost has increased 30 times, and the output cost has gone up 15 times.
In the AI industry, hardware costs are rising fast. GPT-4.5’s high price is likely a way for OpenAI to cover these increased hardware costs.