OpenAI has commenced the phased rollout of its latest model, GPT-4.5, amidst a significant GPU shortage. The rollout began on Thursday for subscribers to ChatGPT Pro, with ChatGPT Plus customers expected to gain access the following week. This strategic deployment highlights OpenAI’s ongoing challenges in meeting the demand for advanced computational resources.
Increased Costs and Computational Demands
The costs associated with GPT-4.5 are notably higher than its predecessor, the GPT-4o model. OpenAI charges $75 per million tokens (~750,000 words) fed into GPT-4.5 and $150 per million tokens generated. This pricing structure reflects the model’s increased computational demands, as GPT-4.5 requires 30 times more cost to input and 15 times more to output compared to GPT-4o. OpenAI CEO Sam Altman has described the model as “giant” and “expensive,” necessitating “tens of thousands” more GPUs before expanding access.
“We’ve been growing a lot and are out of GPUs.” – Sam Altman
To address these challenges, OpenAI plans to add tens of thousands of GPUs next week, enabling a broader rollout to ChatGPT Plus customers. The company is actively working on developing its own AI chips to mitigate future GPU shortages, an issue that has delayed product launches and impacted user access. OpenAI’s rapid growth has exacerbated GPU scarcity, making it difficult to predict and prepare for computational demands.
In response to these challenges, OpenAI is constructing a massive network of datacenters aimed at bolstering its computing capacity. This ambitious endeavor aligns with the company’s long-term goal of improving computing resources while reducing reliance on external GPU availability. Sam Altman emphasized that OpenAI does not want its operations to hinge on fluctuating GPU supplies.
As the company navigates these obstacles, it remains committed to enhancing its technological capabilities. OpenAI aims to significantly improve its computing capacity in the coming years, ensuring more robust and reliable access to its AI models.
What The Author Thinks
While OpenAI’s GPT-4.5 rollout showcases impressive advancements in AI, the company’s reliance on external GPUs and the resulting resource constraints point to potential long-term issues. The transition to in-house AI chips and the development of a vast datacenter network could help address these problems, but it remains to be seen whether OpenAI can keep up with the ever-growing demand for AI capabilities. OpenAI’s challenge will be ensuring that its infrastructure evolves in line with the rapid pace of innovation, without sacrificing access or affordability for users.
Featured image credit: Jobba
Follow us for more breaking news on DMR