Expenditure soared to 34 billion US dollars! OpenAI continues to burn money wildly before going public, AI arms race enters a stage of "surging income and coexisting with loss black hole"

Some media reports citing informed sources revealed that as ChatGPT developer OpenAI prepares for its public listing and discloses a series of performance data, the company's overall expenditure in 2025 is estimated to be around $34 billion, according to reports. In terms of more detailed operational data, sources said that OpenAI's net loss is approximately $38.5 billion, which is almost 8 times higher than the $5.1 billion loss in 2024. These latest operational data undoubtedly raise questions among investors about the financial sustainability and fundamental growth prospects of the world's highest-valued artificial intelligence application company. It is understood that the surge in expenditure reflects the company's continued large-scale investment in artificial intelligence computing infrastructure, including AI large model training, data center construction, talent recruitment, and the expansion of the AI application market. Sources said that out of the total expenditure in 2025, around $19 billion will be used for research and development, while about $6 billion will be used for sales and marketing and other costs. Despite the high costs associated with AI computing power, OpenAI achieved strong revenue growth of approximately $13 billion during the same period, compared to previous years. This increase is mainly attributed to the growth in subscription-paid users of ChatGPT's high-end AI large model services. By the end of 2025, the monthly revenue run rate had reached $2 billion, significantly higher than the $1 billion per quarter level at the end of 2024. However, despite the large-scale expenditure, losses have increased significantly, with OpenAI's net loss expanding from $5 billion in 2024 to approximately $39 billion in 2025. The report added that a large part of the increase in losses is attributed to non-cash accounting charges related to the company's previous corporate structure, rather than core operating costs. Calculated on an adjusted performance data basis, the overall loss is close to $8 billion. The $38.5 billion net loss cannot be simply understood as a result of uncontrollable operational burn, as the majority of it comes from non-cash accounting charges resulting from the company's transition from a non-profit structure to a Public Benefit Corporation (PBC). During this transformation, investors' convertible rights were usually treated as liabilities in accounting and revalued as the valuation increased, resulting in approximately $30 billion in balance sheet costs; with a total loss of about $41.5 billion related to the change in fair value of convertible rights and stock warrants. After excluding these non-cash items, equity incentives, Microsoft computing resource deductions, and other factors, the actual operating loss is approximately $8 billion. Therefore, OpenAI did not actually burn $38.5 billion in cash in one year, but even after accounting for the cancellations, its operating model is still in a phase of high-intensity expansion with losses. According to sources cited by multiple media outlets, OpenAI aims to potentially conduct an IPO on the US stock market by the end of 2026. Last week, OpenAI agreed to acquire Ona, a well-known startup that provides cloud computing services to support the deployment of artificial intelligence agents for enterprises. This move is an important part of the efforts by the global top AI developer with ChatGPT to make its cutting-edge AI technology more effective in improving business efficiency and helping enterprises deploy AI agents more easily. The core logic behind OpenAI's acquisition of Ona is not simply to "buy a cloud service startup," but rather to fill in the crucial infrastructure gap of enterprise-level AI agents before the IPO window: a secure, persistent, and governable cloud execution environment. From a product engineering perspective, this acquisition addresses the "last mile" of commercializing intelligent agents: it is not enough for models to reason, write code, enterprises truly need agents to perform real tasks in an environment with controlled permissions, isolated credentials, auditable logs, reversible processes, and auditable results. OpenAI explicitly stated that Ona's controllable execution models will allow agents to run in an enterprise's own cloud environment while OpenAI provides intelligence and orchestration capabilities, allowing enterprises to deploy Codex without sacrificing infrastructure, data, or control over security boundaries. Overall, OpenAI's preliminary financial data for 2025 shows the typical characteristics of "super growth and super cash consumption coexisting": annual revenue of approximately $13 billion, more than 2.5 times the $3.7 billion in 2024, with a monthly revenue run rate of $2 billion by the end of the year, highlighting the emerging real business demand for high-end ChatGPT service subscriptions, enterprise APIs, agent tools, and developer ecosystem. However, total expenditure reaching $34 billion during the same period - compared to approximately $12 billion in 2024, with 2025 expenditure about 2.6 times revenue - includes research and development costs of around $19.2 billion, sales and marketing costs of about $5.8 billion, and revenue costs of about $7.5 billion. This also indicates that OpenAI's growth is not a light asset internet platform-style expansion, but a "capital-intensive AI platform expansion" that relies heavily on model training, inference computing resources, data center construction processes, top talent, sales network, and enterprise deployment systems. Therefore, for the capital market, OpenAI's IPO will be one of the most critical valuation touchstones for the AI super bull market. On the positive side, OpenAI has proven that generative AI has rare revenue growth climbing speed, with $13 billion in annual revenue and a monthly revenue run rate of $2 billion, enough to support its position as the next generation enterprise AI software, AI developer ecosystem platform, and agent entrance; while the negative aspect is that its path to profitability still relies heavily on decreasing computing costs, improving inference efficiency, increasing enterprise payment penetration, closing the loop on agent business, and data center capital spending returns. If the public market accepts the narrative of "high losses in exchange for AI inference computing power infrastructure resources and AI application platform monopoly entry," OpenAI will strengthen the valuation ceiling for AI platform companies; if investors focus more on the $34 billion expenditure, $8 billion adjusted loss, and Microsoft computing dependence, then OpenAI's listing may become a turning point for the market to reexamine the "AI bubble" argument, the cost curve of computing power leasing, and the commercial efficiency of AI large models.