OpenAI CEO Sam Altman: Company 'Out of GPUs'

OpenAI's GPT-4.5 Rollout Delayed Due to GPU Shortage

According to Sam Altman, CEO of OpenAI, the deployment of their latest language model, GPT-4.5, is being implemented in phases. This is a direct result of a current shortage of GPUs within the company’s infrastructure.

Limited Initial Access

Altman communicated via X (formerly Twitter) that GPT-4.5, characterized as a significantly large and costly model, necessitates the addition of “tens of thousands” of further GPUs before broader access can be granted to ChatGPT users.

The initial release of GPT-4.5 will prioritize subscribers to ChatGPT Pro, beginning on Thursday. Following this, access will be extended to ChatGPT Plus customers in the subsequent week.

Significant Cost Associated with GPT-4.5

The substantial size of GPT-4.5 translates into considerable operational expenses. OpenAI has established pricing at $75 for every million tokens inputted into the model—approximately 750,000 words.

Furthermore, generating output from the model is priced at $150 per million tokens. This represents a 30-fold increase in input costs and a 15-fold increase in output costs when compared to OpenAI’s widely used GPT-4o model.

Growth and Infrastructure Challenges

“We’ve been experiencing substantial growth and are currently limited by GPU availability,” Altman stated. “We anticipate adding tens of thousands of GPUs next week, which will enable us to extend access to the Plus tier.”

He acknowledged that this phased rollout is not the ideal operational approach, but that accurately forecasting growth surges and preventing GPU shortages is a complex undertaking.

Long-Term Solutions

Altman has previously indicated that limitations in computing power are hindering the progress of OpenAI’s product development.

To address this, OpenAI is actively pursuing the development of proprietary AI chips and the construction of a large-scale network of data centers. These initiatives are intended to mitigate future capacity constraints.

Key Takeaways

• The rollout of GPT-4.5 is being staggered due to a lack of available GPUs.
• GPT-4.5 is significantly more expensive to use than GPT-4o.
• OpenAI is investing in its own hardware to address long-term computing needs.