OpenAI, the leading generative artificial intelligence company, has significantly improved its profitability, drawing attention to whether concerns about an AI bubble will subside.
The Information reported Monday, citing sources familiar with internal financial conditions, that OpenAI's compute margin for its paid business has risen substantially. The company's paid compute margin reached 68% as of October this year, up 16 percentage points from 52% in December last year. Compared to January last year at 35%, the margin has nearly doubled. This figure also exceeds competitor Anthropic's projected year-end margin of 53%. Compute margin is calculated by dividing revenue minus computing costs for paid enterprise and consumer services by total revenue. A 68% margin in October means that out of every $100 in revenue, $32 went to computing costs for paid services, leaving $68.
Following recent news of raising an additional $100 billion in investment, OpenAI's valuation is estimated at up to $830 billion (1,229.48 trillion won). While OpenAI is considered a top IPO candidate for next year, doubts about its profitability have intensified as most users access its services for free. As of July, ChatGPT had 35 million paid subscribers, accounting for only 5% of weekly active users. The company reportedly posted a $5 billion loss last year after pouring astronomical sums into AI development.
Since launching ChatGPT in November 2022, OpenAI has improved server efficiency, and higher-priced subscription plans have gained popularity in the market, contributing to improved profitability, analysts say. The Information noted, "While overall computing costs remain high relative to revenue, the company is generating more revenue per dollar invested from paid subscribers and enterprise model sales."
However, some point out that OpenAI's compute margin remains low compared to publicly traded software companies, suggesting a long road ahead. While Google, a leading competitor, reduces development costs by running its own Tensor Processing Units (TPUs), OpenAI relies on expensive Nvidia chips, making cost reduction structurally difficult.
