OpenAI Revenue Up 18%, Losses Widen Further
According to The Wall Street Journal, OpenAI reported an 18% quarter-over-quarter revenue increase to $6.7 billion for Q2, but operating losses have widened, and operating profit margins continue to decline, disappointing some shareholders.
$6.7 billion is the quarterly revenue, which annualizes to approximately $26.8 billion; this figure is still below the recently reported annualized revenue run rate of over $40 billion, indicating that the latter may include accelerated revenue from later months, contracted revenue, or different internal accounting metrics, and cannot be directly compared to the year-over-year revenue for Q2. OpenAI has not publicly released a complete quarterly financial statement.
OpenAI achieved $13.07 billion in revenue in 2025, but net losses reached $38.5 billion. The latest Q2 data indicates that commercial revenues from ChatGPT subscriptions, enterprise APIs, coding tools, and advertising are still growing, but the revenue growth rate has not yet covered the expanding costs of model training, inference, talent, GPU leasing, and data centers.
The further negative operating profit margin means that an additional $1 in revenue has not brought enough gross profit improvement to cover the incremental computing power and operating costs. For companies relying on cutting-edge models, an expanding user base may simultaneously increase inference token consumption; if the unit inference cost decreases more slowly than the growth in usage, faster revenue growth may also amplify losses.
Recently, OpenAI stated that the annualized revenue run rate in July grew by over 20% quarter-over-quarter, driven by core consumer business, subscriptions, AI coding software, and early advertising business; however, the company has not disclosed the profit margin, cash consumption, or capital expenditure associated with this run rate.
In market mechanisms, paid ChatGPT users, enterprise API customers, and developers are the demand side for OpenAI's revenue; Nvidia, AMD, cloud vendors, data centers, GPU clouds, and power suppliers gain supply-side revenue from OpenAI's training and inference expenditures. If model fees, enterprise seats, and agent application revenues do not grow faster than the decline in inference costs and the expansion of user usage, capital will continue to flow from equity investors to computing power and cloud infrastructure suppliers; AI companies that can lock in low-cost computing power, increase model unit token revenue, or enhance customer willingness to pay through software workflows will benefit.
Source: Public Information
ABAB AI Insight
OpenAI's commercialization path has undergone a continuous transformation from a research institution, API services to consumer subscriptions and enterprise software. The launch of ChatGPT in 2022 brought a user base, while GPT-4 and subsequent models drove enterprise APIs and ChatGPT Plus subscriptions; recently, coding tools, enterprise seats, and advertising have been reported as new revenue drivers. However, the $13.07 billion revenue in 2025 corresponds to a net loss of $38.5 billion, indicating that the challenge for cutting-edge model companies is not a lack of demand, but that the cost curve of training and inference infrastructure has not yet been covered by the revenue curve.
A capital path is forming a structure of "model company losses, infrastructure profits first." OpenAI obtains capital and computing power from investors, partnerships with Microsoft, and the cloud and chip supply chain, then sells subscriptions, tokens, and software services to consumers and enterprises; a significant portion of cash flow returns to GPU, network, data center, power, and cloud service providers. Nvidia's chip sales, cloud vendors' computing power leasing, and contract revenues from GPU clouds like CoreWeave can be realized during the construction period, while OpenAI must wait for customers to continue paying, inference costs to decrease, and product pricing to increase before it can achieve operational leverage.
A historical analogy is the early growth model of Uber and the early infrastructure curve of AWS. Uber once subsidized to quickly acquire orders and supply but had to prove the unit economics of each order in the long term; AWS made heavy capital investments first and then improved profit margins through shared resources, scale utilization, and long-term contracts with enterprises. OpenAI is currently closer to a combination of both: it bears the upfront investment in model R&D and computing power, as well as customer acquisition, free user services, and price experimentation aimed at consumers, and has not yet fully established a stable pricing and utilization structure similar to cloud services.
Essentially, it belongs to capital concentration. Cutting-edge AI requires ultra-large-scale GPUs, data centers, power, data, model engineering, and security teams, with fixed costs far exceeding traditional SaaS; therefore, laboratories with strong financing capabilities can maintain losses and expand model scale, but short-term cash flow and bargaining power in the value chain are more easily concentrated at the chip, cloud, and power asset ends. Only when OpenAI converts model capabilities into indispensable enterprise workflows, increases revenue per unit of computing power, and reduces inference costs can capital returns potentially flow back from the infrastructure layer to the model layer.