OpenAI's preliminary financial data for 2025 exhibits the classic pattern of 'hypergrowth coupled with hyper cash burn.'
According to Zhitong Finance APP, citing reports from media outlets referencing informed sources, as OpenAI, the developer of ChatGPT, prepares for a public listing and discloses a series of financial metrics, the company’s total expenditures for 2025 are estimated at approximately USD 34 billion. More granular operational data indicates that OpenAI reported a net loss of approximately USD 38.5 billion, nearly eight times higher than the USD 5.1 billion loss recorded in 2024. These latest figures have inevitably raised investor concerns regarding the financial sustainability and fundamental growth prospects of the world’s most highly valued artificial intelligence application company.
The surge in expenditures reflects the company’s continued large-scale investments in AI computing infrastructure, including training of large AI models, data center construction, talent acquisition, and expansion into the AI application market. According to informed sources, of the total spending in 2025, approximately USD 19 billion was allocated to research and development, while around USD 6 billion went toward sales, marketing, and other costs.
Despite the high cost base associated with AI computing, OpenAI still achieved robust revenue growth of approximately USD 13 billion during the same period—a substantial increase compared to prior years—primarily driven by a rise in subscription-paying users for ChatGPT’s premium large-model AI services. By the end of 2025, its monthly revenue run rate reached USD 2 billion, significantly higher than the quarterly USD 1 billion level recorded at the end of 2024.
Nevertheless, massive expenditures continue to drive a sharp increase in losses, with OpenAI’s net loss widening from USD 5 billion in 2024 to approximately USD 39 billion in 2025. The report added that a significant portion of this increased loss stems from non-cash accounting charges related to the company’s previous corporate structure, rather than core operating expenses. On an adjusted basis, the overall loss is closer to approximately USD 8 billion.
The reported net loss of USD 38.5 billion should not be simplistically interpreted as pure operational cash burn spiraling out of control, as the vast majority stems from non-cash accounting charges triggered by the company’s structural transformation. During OpenAI’s transition from a nonprofit structure to a Public Benefit Corporation (PBC), convertible equity previously held by investors was typically classified as debt under accounting rules and subsequently revalued upward alongside rising valuations, resulting in approximately USD 30 billion in book expenses. Losses tied to changes in the fair value of convertible equity and warrant liabilities totaled approximately USD 41.5 billion. After excluding these non-cash items, stock-based compensation, and offsets from Microsoft’s computing resources, the actual operating loss amounts to roughly USD 8 billion. Thus, OpenAI did not truly burn through USD 38.5 billion in cash over the year; however, even after adjusting for accounting write-downs, its business model remains in a phase of high-intensity loss-driven expansion.
Multiple media outlets, citing informed sources, reported that OpenAI aims to conduct an initial public offering (IPO) on the U.S. stock market as early as late 2026.
Last week, OpenAI agreed to acquire Ona, a prominent startup providing cloud computing services that enable enterprises to deploy AI agents. This move represents a critical step for OpenAI—the world’s leading AI developer behind ChatGPT—in its effort to enhance enterprise operational efficiency through cutting-edge AI technologies and to facilitate easier deployment of AI agents for businesses.
The core rationale behind OpenAI’s acquisition of Ona is not merely to 'buy a cloud services startup,' but to fill a critical gap in enterprise-grade AI agent infrastructure ahead of its IPO window: a secure, persistent, and governable cloud execution environment.
From a product engineering perspective, this acquisition addresses the 'last mile' challenge of agent commercialization: it is insufficient for models merely to reason or generate code. What enterprises truly need is for agents to execute real-world tasks within environments that enforce controlled permissions, credential isolation, auditable logging, process rollback capabilities, and verifiable outcomes.
OpenAI explicitly stated that Ona’s customer-controlled execution model will allow agents to operate within enterprises’ own cloud environments, while OpenAI provides intelligence and orchestration capabilities—enabling businesses to deploy Codex without relinquishing control over their infrastructure, data, or security boundaries.
Overall, OpenAI’s preliminary 2025 financial figures reflect a classic pattern of 'hyper-growth coexisting with hyper cash consumption': annual revenue of approximately USD 13 billion, more than 2.5 times the USD 3.7 billion recorded in 2024, and a monthly revenue run rate of USD 2 billion by year-end—highlighting the emergence of genuine commercial demand for premium ChatGPT subscriptions, enterprise APIs, agent tools, and the developer ecosystem.
However, total expenditures during the same period reached as high as $34 billion—compared to approximately $12 billion in 2024. In 2025, total expenditures are projected to be about 2.6 times revenue, comprising roughly $19.2 billion in R&D expenses, $5.8 billion in sales and marketing expenses, and $7.5 billion in cost of revenue. This underscores that OpenAI’s growth is not characterized by the asset-light expansion typical of internet platforms, but rather by a capital-intensive AI platform expansion heavily reliant on model training, inference computing resources, data center construction progress, top-tier talent, sales networks, and enterprise deployment ecosystems.
Therefore, for capital markets, OpenAI’s IPO will become one of the most critical valuation benchmarks of the AI super bull market. On the positive side, OpenAI has already demonstrated that generative AI can achieve an exceptionally steep revenue growth trajectory; with annual revenue of $13 billion and a monthly recurring revenue run rate of $2 billion, it is sufficiently positioned to be regarded as the next-generation enterprise AI software platform, AI developer ecosystem, and intelligent agent gateway. On the downside, its path to profitability remains highly dependent on declining computing costs, improved inference efficiency, enterprise adoption rates, the commercial viability of intelligent agents, and returns on data center capital expenditures.
If public markets embrace the narrative of 'accepting high losses in exchange for securing AI inference computing infrastructure and monopolistic access to AI application platforms,' OpenAI could reinforce the valuation ceiling for AI platform companies. Conversely, if investors focus instead on its $34 billion in expenditures, $8 billion in adjusted losses, and dependence on Microsoft’s computing resources, OpenAI’s listing could also mark a turning point for the market to reevaluate arguments about an 'AI bubble,' the cost curve of computing capacity leasing, and the commercialization efficiency of large AI models.
Editor/Deng