Will OpenAI eventually be able to turn a profit?

Recently, OpenAI and Anthropic have both launched IPO plans, with the market's latest valuations for these two AI large-scale model vendors approaching one trillion dollars, reflecting investors' high optimism about their future profit prospects. Given the broad application prospects of AI large-scale models and the huge revenue growth potential of large-scale model vendors, investors' optimism is entirely understandable. However, revenue growth does not necessarily lead to profit growth, and it does not even guarantee corporate profitability. To date, no vendor has achieved independent profitability in its large-scale model business. Theoretically, whether a vendor can achieve sustainable profitability depends on whether it possesses high competitive barriers and stable pricing power, which in turn depends on the industry's market structure and competitive landscape. Research has found that the current large model API call market exhibits a monopolistic competition pattern, with a large number of vendors and very low market concentration. Although market demand is growing exponentially, the low barriers to entry have led to a rapid expansion of the large model supply side, preventing vendors from achieving profitability in line with market demand expansion and instead facing increasingly fierce competition. In this context, some vendors can differentiate their products through technological advantages or scenario adaptation, thereby obtaining short-term excess profits; however, due to limited technological barriers, high price elasticity of demand, and weak user stickiness, even if they achieve excess profits, it is difficult to sustain them. In the long run, those vendors that consistently incur losses will be forced to exit the market, driving the large model API market from monopolistic competition to oligopolistic dominance. However, in an oligopolistic market structure, the profitability of manufacturers remains uncertain, depending on whether they engage in price competition or volume competition. If they cannot coordinate their competitive strategies or establish effective differentiation barriers, oligopolistic manufacturers may not achieve sustainable profitability, and their huge upfront R&D investments may not be recouped. In short, while the technological value and demand growth of large-scale models are undeniable, large-scale model manufacturers that simply "sell tokens" are not necessarily profitable. Therefore, investors need to carefully examine the valuations of large-scale model manufacturers like OpenAI, while manufacturers need to carefully choose their business models and market segments. Regardless of the business model adopted, if manufacturers can establish differentiation barriers in areas such as model capabilities, industry adaptation, enterprise workflows, or application ecosystems, they can reduce users' price sensitivity, gain pricing power in niche markets, and achieve sustainable profitability. Given that the "AI+" model embeds AI functionality into existing products or services to enhance their value to users, strengthen existing differentiation barriers and customer loyalty, it is likely to be a sustainable and profitable business model. Recently, OpenAI and Anthropic have both launched IPO plans, with the market's latest valuation of these two AI large-scale model vendors approaching one trillion US dollars. Their price-to-sales ratios (P/S ratios) have reached 34 and 21 times respectively, reflecting investors' high optimism about their future profit prospects. Given the broad application prospects of AI large-scale models and the huge revenue growth potential of large-scale model vendors, investors' optimism is entirely understandable. However, as is well known, high revenue growth does not necessarily lead to high profit growth, and it does not even guarantee corporate profitability. To date, no manufacturer has achieved independent profitability (net profit) in its large-scale model business. Take OpenAI as an example. Its annualized revenue grew from $2 billion in 2023 to over $20 billion in 2025, a tenfold increase in three years, but the company is still not profitable. However, if the high cost of equity incentives is taken into account, its net profit may still be negative. Moreover, considering the pressure of rapid iteration for large models, its future model training and various R&D costs will remain high, so the sustainability of its operating profit remains to be seen. This means that even for the most leading model vendors, rapid revenue growth cannot guarantee profitability. According to microeconomic theory, a firm's sustainable profitability does not depend on the size of the market demand it participates in, but rather on the market structure and competitive landscape. In a perfectly competitive market, regardless of the size of market demand, in equilibrium, firms can only obtain zero profit (referring to economic profit, not accounting profit), or "normal profit," and cannot obtain excess profits. Conversely, in a monopolistic market, even with limited market demand, firms can still obtain excess profits. Therefore, to assess the long-term profitability prospects of large-scale firms, it is first necessary to analyze the market structure and competitive landscape of the large-scale market. This analysis not only helps investors determine whether the capital market valuation of large-scale firms is reasonable, but also helps firms identify and select business models and competitive strategies with long-term sustainable profitability prospects. **Introduction to the Main Business Models and API Call Market of the Large Model** Currently, the commercialization of the large model mainly takes four forms: Subscription (for individuals or enterprises, charging monthly or annual fees per seat), API Call (for developers and enterprises, charging based on token usage), and Contract (for government and enterprise clients, providing customized adjustments and maintenance services). The four models—the "AI+" model (embedding large model capabilities into existing products or businesses)—have different pricing methods and serve different customer groups (Figure 1), effectively opening up four (or even more) different market segments. Manufacturers choosing different business models (some choosing multiple models) also means they are choosing different market segments. Among the four business models mentioned above, the API call model can be simply referred to as the "selling tokens" business model. Since the subscription model, contract model, and "AI+" model have limited publicly available data and often involve complex product portfolios, customized solutions, or ecosystem strategies, accurate comparison and quantitative analysis are difficult. In contrast, the API call model offers publicly available data, transparent pricing, unified measurement standards, and measurable market share, making it highly suitable for microeconomic analysis. Therefore, we selected this model to analyze the demand characteristics, market structure, and competitive landscape of the large-scale model API market, and thus assess the profitability of large-scale model vendors. In the early stages of large-scale model applications, the API market consisted of only a few vendors such as OpenAI and Anthropic. Each vendor's interface was independent, requiring users to connect separately and pay monthly or per-token usage. The cost of comparing and converting models was high. As the number of market participants increased, model aggregation gateways (AI gateways) emerged. Specifically, a model aggregation gateway is an intermediary service platform located between users and large-scale model vendors. These platforms are standard two-sided market platforms, and their initiators and operators include organizations such as OpenRouter, Lite LLM Proxy, and Cloudflare. The platform connects to multiple model vendors on one side and users on the other, providing users with a unified model API call interface and charging based on the number of token calls. After a user sends a request to the gateway platform, the platform routes the request to the target model according to user-specified rules or preset strategies; after the model returns the result, the gateway then forwards it to the user (Figure 2). In other words, users only need to use one interface to call multiple models, without having to connect to different vendors separately, significantly reducing search costs, comparison costs, and conversion costs. According to data from model aggregation gateways, the large model API market has seen explosive growth in call volume over the past year. Taking OpenRouter as an example, its platform's weekly API usage has increased more than 23 times in less than a year and a half (Figure 3). This is partly due to the transparency and convenience offered by aggregation gateways, and even more so due to the recent rise of AI agents. Before the rise of agents, a user's interaction with a large AI model typically corresponded to a single API call; however, agents, through task decomposition, multi-step planning, and external tool calls, transform a single user intent into multiple rounds of model API requests, thereby significantly amplifying token consumption and API call demands.

The large model API market exhibits characteristics of a monopolistic competition market

As mentioned earlier, growth in market demand does not necessarily lead to profit growth, and it cannot even guarantee that a company will be profitable; a company's profitability depends on the market structure and competitive landscape of related products. ...span>

As mentioned earlier, growth in market demand does not necessarily lead to profit growth, and it cannot even guarantee that a company will be profitable; a company's profitability depends on the market structure and competitive landscape of related products.

（3) Demand price elasticity is relatively large, but not infinite; there are differences between models, but manufacturers' pricing power is limited. On OpenRouter, free models (with usage limits) and low-priced models saw significantly higher usage, indicating that users are highly price-sensitive. However, some high-priced models still garnered substantial usage, resulting in a U-shaped relationship between model usage and price (Figure 6). Because different models vary in overall capabilities, invocation costs, and applicable scenarios, they are not entirely homogeneous. Data shows that higher-priced models often correspond to stronger technical performance (Figure 7), thus confirming that the price differences between large models stem from "quality differences." The large model market is not characterized by homogeneous competition, but rather by differentiated positioning. Based on this, it can be concluded that the large model API market is not a perfectly competitive market, but rather a monopolistic competition market.

The entry barriers in the large model market are lower than expected, mainly due to the following reasons: (1) Based on the expectation of high returns in the future of large models, investors are vying to raise funds for large model R&D institutions through various means such as PE, VC, CVC (VC within large enterprises), and IPO, which has greatly reduced the capital threshold. (2) The existence of open source models and the "distillation" behavior have reduced the learning cost for latecomers, enabling them to absorb and replicate verified technological achievements at a lower cost, compressing the technological gap between leaders and followers, and greatly reducing the technological threshold. (3) A highly open and mobile labor market allows high-end AI talent to move relatively freely between companies, which lowers the talent threshold for companies and accelerates the diffusion of cutting-edge large-scale model technologies among companies. In summary, based on the above analysis and data from OpenRouter, Epoch AI, and other institutions, the current large-scale model API market possesses the basic characteristics of a monopolistic competition market. Profitability prospects of the large-scale model API market: Generally speaking, in a monopolistic competition market, companies can obtain limited pricing power in the short term by relying on product differentiation, thereby obtaining excess profits (Figure 9, middle figure). However, excess profits attract new firms, diverting market demand from existing firms and causing their demand curves to gradually shift downwards, thus narrowing the excess profit margin until it approaches zero, at which point the market reaches long-term equilibrium (Figure 9, right). In other words, in a monopolistic competition market, although firms may obtain excess profits in the short term, these excess profits will eventually disappear in the long run at equilibrium. Given that the current large-scale model API market exhibits characteristics of a monopolistic competition market, the above mechanism also applies to it. However, due to the large upfront investment costs of large-scale models, despite rapid market demand growth, the demand curves of most manufacturers (D) have so far failed to exceed the average cost curve (ATC), and therefore they are all operating at a loss (Figure 9, left). Of course, given the exponential growth in demand for large models (the demand curve will shift upwards) and the rapid decline in training costs for large models (the average cost curve will shift downwards), at some point in the future, the demand curve (D) may exceed the average cost curve (ATC). This allows for profitability (referring to excess profit, as shown in Figure 9). Anthropic's recent performance validates this dynamic process. However, as mentioned earlier, in a monopolistic competition market, excess profits will attract more firms to enter that market segment (or manifest as other firms striving to narrow the technological gap and product quality differences with the leading firm), thereby diverting market demand from existing firms or the leading firm. This leads to a downward shift in the demand curve for individual firms, causing excess profits to gradually disappear (Figure 9, right). Clearly, in a monopolistic competition environment, it is not easy for model firms to achieve sustainable economic profits or excess profits. Due to the high training costs and rapid iteration speed of large models, coupled with fierce price competition, many manufacturers are forced to launch new generations of models before the previous generation has even recouped its costs, resulting in prolonged losses. Over time, those with insufficient financial resources and weak commercialization capabilities may be forced to exit the market, and market share is likely to gradually concentrate on a few leading manufacturers with advantages in capital, technology, brand, and ecosystem, driving the market structure from monopolistic competition to oligopolistic dominance. However, even in the event of oligopolistic dominance, whether large model manufacturers can achieve sustainable profits still depends on the competitive strategies adopted by these oligopolistic manufacturers. According to firm theory, typical oligopolistic competition includes price competition (Bertrand Competition) and quantity competition (such as Cournot Competition or Stackelberg Competition). In the Bertrand Competition model, due to price competition, the market equilibrium price will approach marginal cost, and firms cannot obtain excess profits. In the Cournot or Stackelberg Competition models, the market equilibrium price can be higher than the firm's marginal cost, thus generating positive unit profits. However, given the extremely high fixed costs of large-scale model development (such as R&D, training, and computing infrastructure), if unit profits are insufficient to cover initial investments, even if oligopolistic firms adopt quantity competition, it is difficult to say whether they can ultimately achieve overall profitability. In reality, in many oligopolistic industries (such as telecommunications, aviation, automobiles, oil, and food delivery platforms), oligopolistic manufacturers do not necessarily enjoy high profits, but only obtain average or low profits (or even often operate at a loss), which confirms the above theory. In conclusion, the current large-model API call market exhibits a monopolistic competitive landscape, with numerous vendors and very low market concentration. Almost all large-model vendors are operating at a loss. Although API market demand is growing exponentially, the low barriers to entry have led to a rapid expansion of the large-model supply side, preventing vendors from achieving profitability in line with market demand expansion. Instead, they face increasingly fierce competition. Theoretically, some vendors could differentiate their products through technological advantages or scenario adaptation, gaining a degree of pricing power in relevant market segments and thus obtaining short-term excess profits. However, due to limited technological barriers, high price elasticity of demand, and weak user stickiness, even if these vendors achieve excess profits, it will be difficult to sustain them. In the long run, those manufacturers that consistently incur losses will be forced to exit the market, and market share is likely to gradually concentrate on a few leading manufacturers, driving the market structure from monopolistic competition to oligopolistic dominance. However, under an oligopolistic structure, a manufacturer's profitability remains uncertain, depending on whether they engage in price competition or quantity competition. If they cannot coordinate competitive strategies with competitors or establish effective differentiation barriers, oligopolistic manufacturers may not achieve sustainable profitability, and their huge upfront R&D investments may not be fully recovered. In short, while the technological value and demand growth of large-scale models are undeniable, large-scale model manufacturers that simply "sell tokens" may not necessarily achieve long-term profitability. Therefore, investors need to calmly examine the valuations of large-scale model manufacturers like OpenAI, and large-scale model manufacturers also need to carefully choose their business models and market segments. For investors, there are three points to consider. First, given the significant growth potential in market demand for large-scale models, it's difficult to confirm or disprove investors' assessments of the profitability and valuation rationality of large-scale model manufacturers in the short term. Therefore, even if the market valuation of large-scale model manufacturers is unreasonable, valuation correction is likely to be a lengthy process, and it's possible that irrational pricing in the market will persist for a considerable period. Second, this article only discusses the API call model ("selling tokens"), and its conclusions are not applicable to the other three business models (subscription, contract, or "AI+" models). Therefore, for large-scale manufacturers employing multiple business models, the conclusions of this article alone cannot be used to determine the reasonableness of their valuations; rather, the long-term value of other business models must be considered simultaneously. Third, even for large model vendors that primarily use the API call model, it cannot be ruled out that they may adjust their business strategies in the future, adopting multiple business models, providing new products or services, exploring new application scenarios, or innovating their business models, thereby gaining new development opportunities. Therefore, their valuation needs to be viewed dynamically, with continuous tracking and updates. For large model vendors, it is important to note that the four business models for large model applications each correspond to different market segments, and their profit logics differ. For example, the "AI+" model embeds more AI functions into existing products or services, which helps to enhance the value of existing products or services to users, strengthen existing differentiation barriers and customer stickiness, and help vendors achieve broader and more sustainable profitability. Contractual models are often deeply integrated with users' private data, workflows, and business systems. Users may face higher migration costs, higher stickiness, and lower price transparency (comparability). Larger model vendors have greater pricing power and are therefore more likely to generate and maintain excess profits. Compared to the "AI+" model and contractual models, subscription models are closer to (but different from) API call models in terms of market structure, user characteristics, user stickiness, price transparency, and price elasticity of demand. Therefore, the findings of this study have some reference value for large model vendors adopting subscription-based business models. Of course, further in-depth and detailed research is needed on these three business models. However, regardless of the business model adopted, if manufacturers can establish differentiated barriers in areas such as model capabilities, industry adaptation, enterprise workflows, and application ecosystems, they can reduce user price sensitivity, enhance customer stickiness, and gain more stable pricing power in niche markets, ultimately achieving sustainable profitability. Finally, it is worth noting that compared to large model manufacturers facing fierce competition and difficulty in achieving sustainable profitability on aggregation platforms, large model API aggregation platforms (such as OpenRouter) may be able to form sustainable business barriers through "network effects." As the distribution entry point for API call requests, aggregation platforms connect model manufacturers on one end and developers and enterprise users on the other, exhibiting typical two-sided market characteristics and potentially forming a "two-sided network effect": the more models integrated, the richer the user choices, and the stronger the platform's attractiveness; the larger the platform's user base, the more concentrated the call demands, and the stronger the model manufacturers' willingness to integrate. If a platform can leverage various technological means and customized services to further enhance customer loyalty beyond the "two-sided network effect," it may be able to effectively prevent price competition from rival platforms, maintain its first-mover advantage, and ultimately achieve a "winner-takes-all" outcome. Further in-depth research is warranted in this area.