CITIC Construction Investment: OpenAI's advertising monetization implementation accelerates large model commercialization

CITIC Construction Investment points out that the first batch of ChatGPT advertisements will go live in early February 2026, initially targeting free and Go subscription tier users, with charges based on CPM (impressions). CITIC Construction Investment believes that OpenAI’s advertising monetization is relatively restrained, and responses do not prioritize advertising weight, which better balances user experience; unlike the mobile internet era, the marginal cost of large models will not decrease with user growth. Instead, costs such as computing power will increase as user numbers and dialogue volume grow. Exploring the commercial value of free users is an essential step to open up the business closed loop, and diversified monetization methods like advertising are a necessary choice for large model companies with strong B2C attributes.

Full Text Below

CITIC Construction Investment: OpenAI’s Commercialization of Advertising Accelerates

The first batch of ChatGPT ads will be launched in early February 2026, initially charging based on CPM, with a restrained approach. According to The Information on January 21, OpenAI has begun selling ad space for ChatGPT to dozens of brands, initially charging per CPM, with the possibility of introducing a pay-per-click model in the future. The restrained approach to advertising monetization mainly reflects two points:

1. Prioritizing user experience. ChatGPT’s answers are based on user intent and needs, without giving advertising weight; ads will be clearly marked.

2. Limited open ad slots. During the trial phase, each advertiser’s budget cap is $1 million, and a self-service system is not yet available; all collaborations require manual coordination.

Why Promote Advertising Monetization?

1. The low marginal cost economic model of the mobile internet era is no longer applicable. The marginal cost of large models will not decrease to extremely low levels with user growth. ChatGPT consumes reasoning costs per dialogue, and the dialogue volume growth curve is steeper than WAU (weekly active users). Not all users pay for tokens consumed. From 2023 to 2025, OpenAI’s computing power consumption and revenue both grow at about 3 times per year, with no signs of the marginal diminishing returns typical of internet products.

Over 90% of ChatGPT users are free, and unlocking the commercial value of these users is essential for large model companies to complete the business closed loop. ChatGPT’s ads will primarily target free users and the newly launched $8/month ChatGPT Go subscribers, while Plus, Pro, Business, and Enterprise users will not see ads. Last July, its WAU exceeded 700 million, but paying users were only 35 million, with a payment rate of about 5%. The company projects WAU will reach 2.6 billion and paying users 220 million by 2030, with a payment rate of around 8.5%, still relatively low. As model capabilities improve and user volume and deep dialogue increase, the token consumption of free users will grow larger. Not exploiting this user base’s commercial value will lead to continued losses. Representing B2C large models like OpenAI will push for diversified monetization. According to The Information, by 2029, 40% of OpenAI’s revenue will come from ChatGPT subscriptions, 23% from Agent products, 19% from new products, and 18% from API calls.

2. Continuous large-scale investment in computing power makes commercialization urgent, and enhancing “self-sustaining” capabilities is critical. According to The Information, in the first half of 2025, OpenAI’s revenue was $4.3 billion, with a loss of $13.5 billion. It is expected that OpenAI’s computing costs will reach hundreds of billions of dollars by 2028. Last November, multiple media outlets reported that OpenAI plans to invest about $1.4 trillion over the next 8 years in data center construction. Based on cash flow breakeven forecasts, OpenAI and Anthropic will still need to bear net outflows for more than 2 years. OpenAI and Anthropic are expected to achieve positive cash flow around 2030 and 2028, respectively, with Anthropic’s breakeven delayed by a year compared to last year’s predictions. From 2025 to 2028, OpenAI’s cash outflows will increase year by year, totaling over $100 billion.

Large model commercialization will accelerate, benefiting service providers and other companies. By 2026, larger AI revenue realization is expected. As overseas leading model companies implement advertising monetization, domestic giants are expected to follow quickly, focusing on advertising and marketing sectors, while continuing to monitor the commercialization of top large models.

1. When and in what form will OpenAI’s advertising be implemented?

OpenAI’s first ads will go live in ChatGPT in early February, initially charging based on CPM, with a restrained approach. According to The Information on January 21, OpenAI has begun selling ad space to dozens of brands, initially charging per CPM, with the possibility of introducing a pay-per-click model in the future. The restrained approach mainly involves two points:

1. Prioritizing user experience. ChatGPT’s answers are based on user intent and needs, without giving advertising weight; ads will be clearly marked.

2. Limited open ad slots. During the trial phase, each advertiser’s budget cap is $1 million, and a self-service system is not yet available; all collaborations require manual coordination.

2. Why promote advertising monetization?

1. The low marginal cost economic model of the mobile internet era is no longer suitable. The marginal cost of large models will not decrease to extremely low levels with user growth. Ads are a way to indirectly have free users pay for the consumed computing power and other costs. ChatGPT consumes reasoning costs per dialogue, and as user numbers grow and models better meet user needs, the dialogue volume growth curve is steeper than WAU. Not all users pay for tokens consumed. From 2023 to 2025, OpenAI’s computing power consumption and revenue both grow at about 3 times annually, with no signs of the marginal diminishing returns typical of internet products.

Over 90% of ChatGPT users are free, and unlocking the commercial value of these users is necessary for large model companies to complete the business closed loop. ChatGPT ads will primarily target free users and the newly launched $8/month ChatGPT Go subscribers, while Plus, Pro, Business, and Enterprise users will not see ads. Last July, its WAU exceeded 700 million, but paying users were only 35 million, with a payment rate of about 5%. The company projects WAU will reach 2.6 billion and paying users 220 million by 2030, with a payment rate of around 8.5%, still relatively low. As model capabilities further improve and user engagement deepens, free user token consumption will increase. Not exploiting this user base’s commercial value will lead to ongoing losses. Representing B2C large models like OpenAI will push for diversified monetization. According to The Information, by 2029, 40% of OpenAI’s revenue will come from ChatGPT subscriptions, 23% from Agent products, 19% from new products, and 18% from API calls, possibly including advertising, AI + e-commerce/medical scenarios.

2. Continuous large-scale investment in computing power makes commercialization urgent, and improving “self-sustaining” capabilities is critical.

According to The Information, in the first half of 2025, OpenAI’s revenue was $4.3 billion, with a loss of $13.5 billion. It is expected that OpenAI’s computing costs will reach hundreds of billions of dollars by 2028. Last November, multiple media outlets reported that OpenAI plans to invest about $1.4 trillion over the next 8 years in data center construction. In the first three quarters of 2025, MiniMax’s revenue was $53.44 million, with an adjusted net loss of $186 million; Zhipu’s H1 2025 revenue was 190 million RMB, with a net loss of 2.35 billion RMB.

From cash flow breakeven forecasts, overseas top large model companies OpenAI and Anthropic will still need to sustain net outflows for over 2 more years. They are expected to achieve positive cash flow around 2030 and 2028, respectively, with Anthropic’s breakeven delayed by a year compared to last year’s predictions. Between 2025 and 2028, OpenAI’s cash outflows will increase year by year, totaling over $100 billion.

3. Core viewpoints

Large model commercialization will accelerate, benefiting service providers and other companies. By 2026, larger-scale AI revenue realization is expected. The low marginal cost unit economic model of the mobile internet era is no longer applicable. In the large model era, as user numbers grow, costs such as computing power consumption also increase, and marginal costs will not drop to extremely low levels. Over 90% of OpenAI’s users are free, and it is projected that about 91.5% will remain free by 2030. Unlocking the commercial value of these users is essential for completing the business closed loop. Coupled with continuous large-scale investments by companies like OpenAI domestically and abroad, we expect large model companies to accelerate commercialization. As overseas leading model companies implement advertising monetization, domestic giants are likely to follow quickly.

Focus on advertising and marketing sectors, while continuing to monitor the commercialization of top large models.

Copyright protection efforts are below expectations, with risks including unclear intellectual property rights, user privacy data leaks, discontinuation of collaborations with IP or celebrities, shifts in public aesthetic preferences, intensified competition, low user willingness to pay, difficulty changing consumption habits, corporate governance issues, subpar content launch performance, slower-than-expected development of generative AI technology, high product R&D difficulty, delays in product launch, rising marketing costs, talent attrition, increasing human costs, regulatory policy risks, and underwhelming commercialization capabilities.

(Source: People’s Financial News)