Marina Lyubimova
OpenAI's flagship model may be pulling in billions, but the math still doesn't add up to a profit. Revised research into the economics of large language models reveals that GPT-5 looks healthy on a compute basis alone yet slides into the red once the full cost picture comes into frame. The estimates draw on public reporting and financial modeling to examine OpenAI's revenue and cost structure following the GPT-5 launch and subsequent model iterations, including OpenAI GPT-5.4 achieving an 833 score on the ARC-AGI-2 reasoning benchmark.
$6B in Revenue, $4B in Compute: The Margin Reality
GPT-5 products generated approximately $6 billion in total revenue. Inference compute costs consumed around $4 billion of that, leaving roughly $2 billion in gross profit and implying compute-level margins of 25% to 40%. That number reflects the sheer scale of running AI services globally, including GPU clusters, data center infrastructure, and the continuous inference workloads needed to serve millions of users around the clock.
Operating Costs and Revenue Sharing Push Net Loss to $1.5B
The story shifts once broader expenses enter the calculation. About $2 billion in operating costs, covering salaries, stock compensation, general administration, and sales and marketing, push the operating result close to break-even, with a range of -30% to +10% depending on modeling assumptions.
Revenue-sharing obligations of roughly $1 billion then pull the bottom line further down, landing at an estimated net loss of $1.5 billion. These dynamics are unfolding as OpenAI keeps expanding its model lineup, with details emerging from the GPT-5.4 leak signaling a major AI upgrade with a 2M token context window and strong results in OpenAI's GPT-5.3 Codex benchmark scores.
The revised estimates underline a tension that defines the current frontier AI moment: revenue from generative AI is growing fast, but compute infrastructure, operational headcount, and partnership agreements keep compressing margins. As models grow more capable and inference demand rises with them, achieving sustainable unit economics remains the central challenge for any company building at this scale.
Marina Lyubimova
