An analysis by Exponential View and Epoch AI delves into OpenAI's unit economics, specifically focusing on the profitability of running GPT-5. The study examines revenue, compute costs, and other operational expenses. It aims to determine if AI models are financially sustainable, given the massive investments in their development and deployment. The findings offer insights into the broader AI industry's financial health.

The study reveals that while OpenAI's gross profit margins from GPT-5 were decent, including all operational costs likely resulted in an operating loss. The short lifespan of models like GPT-5 raises questions about recouping R&D expenses. The report estimates the inference compute costs were $3.2 billion. This raises questions about the long-term viability of current AI business models.

Despite potential losses, the analysis suggests that developing and running these models can still be a worthwhile investment. The progress made attracts customers and fuels further development. The report also highlights that the R&D invested in models like GPT-5 can inform future generations. Microsoft's deal with OpenAI also impacts these economics.

Ultimately, the report indicates that running AI models is profitable in terms of gross margins, but the overall profitability depends on factors like operational costs and model lifecycles. The AI industry is still evolving, and these financial dynamics will continue to shift. The long-term success of companies like OpenAI hinges on their ability to manage these costs effectively.