Peter Smith
⬤ Fresh estimates break down OpenAI's GPT-5 economics from its August launch through the December GPT-5.2 period, showing the model was profitable at the core level but deep in the red once you factor in everything else. The analysis tracks revenue and costs across OpenAI's entire product lineup during those months.
⬤ GPT-5 pulled in about $6.1 billion in total revenue. Running the model cost roughly $3.2 billion in compute expenses, leaving around $2.9 billion in gross profit—that's a healthy 45% margin. At the inference level, the model clearly worked financially and found strong market demand.
⬤ Things flipped once the full cost picture came into view. Salaries and operating expenses hit approximately $1.4 billion, while sales and marketing burned through another $2.2 billion. That pushed the operation into a $700 million loss. Add in revenue sharing deals worth an estimated $1.2 billion, and the final net loss landed near $1.9 billion. These numbers don't even include R&D or depreciation costs.
⬤ The breakdown shows what scaling frontier AI actually costs. Yes, serving these models can generate solid margins, but staffing up, marketing aggressively, and splitting revenue with partners eats through profits fast. It's a reality check for the AI industry—growth is expensive, and turning infrastructure-level wins into actual profitability remains a serious challenge.
Peter Smith
