OpenAI's Commercial Dilemma: Subscription Model Is Not the Answer for Large Models

In an interview in 2024, Sam Altman said he 'hated' advertising and that mixing ads into AI conversations made him feel 'extremely uneasy.' On January 16, 2026, OpenAI officially announced the testing of advertisements in the free version and Go version of ChatGPT. The complete 180-degree turn in just a year and a half speaks volumes about the situation.

As a core pillar of AI commercialization today, the subscription model may not be as stable as many people imagine.

In China, subscription revenue is often seen as a key measure of success for AI commercialization. A frequently cited argument when discussing the gap between Chinese and American AI industries is this: Look at the US, where an AI subscription costs dozens of dollars per month; then look at China, where price wars have driven API usage fees to near zero. The conclusion seems obvious — a market without willingness to pay cannot foster great AI companies, and China has already lost in terms of business models.

This judgment has its merits. Subscription models do offer advantages: predictable revenue, quantifiable user retention, and Wall Street loves this kind of certainty. After transitioning from selling CDs to the Creative Cloud subscription model, Adobe’s market value increased tenfold. Spotify reshaped the music industry with its subscription model, and Netflix proved that content subscriptions could support a market cap of hundreds of billions of dollars.

However, the true financial picture of OpenAI, often regarded as the benchmark for AI subscription models, is far more complex than these predecessors.

By the end of 2025, OpenAI’s annualized revenue will approach $20 billion. Meanwhile, according to internal documents obtained by The Wall Street Journal, the company expects a net loss of about $9 billion for 2025 — spending $1.69 for every dollar earned. Analysts at Deutsche Bank bluntly stated, 'No startup in history has operated at such a scale of losses. We are completely in uncharted territory.'

These figures paint a strange picture: the world’s most successful subscription-based AI product is also one of the tech companies with the largest losses. Adobe, Spotify, and Netflix all made fortunes through subscriptions, so why does the same model fail in the realm of AI?

More notably, those investors who are putting real money into OpenAI don’t seem to care much about subscription revenue itself. According to the latest news, OpenAI is seeking a valuation between $750 billion and $830 billion in its latest funding round. This figure clearly isn’t based on the logic of 'annual revenue multiplied by some factor' typical of subscription models — even using the most aggressive SaaS valuation multiples, $20 billion in revenue can't justify an $800 billion valuation.

What supports it is a different narrative:OpenAI will become the foundational platform of the AI era, the infrastructure for all future intelligent applications.Investors are buying into this story, not the $20 monthly membership fee.

Here arises a peculiar double standard: When evaluating OpenAI, the logic of being 'the foundation of everything' is applied; but when assessing other AI companies, the yardstick shifts to 'how much subscription revenue you can generate now.' If even OpenAI's investors don’t believe the subscription model can support an $800 billion valuation, shouldn’t we re-examine the ruler itself?

01 The Structural Dilemma of the Subscription Model

If subscription revenue truly were the correct measure of an AI company's value, OpenAI should be the most relaxed venture in the world. $2 billion in 2023, nearly $4 billion in 2024, and projected to reach $20 billion by the end of 2025, with astonishing growth. No SaaS product has ever grown faster.

However, evidence shows that while this company is enjoying the steepest subscription growth curve in history, it is still urgently exploring every possibility beyond subscriptions.

A report from Deutsche Bank in October 2025 showed that consumer spending on ChatGPT in the European market had almost stagnated since May of that year. During the same period, ChatGPT had reached 800 million weekly active users, but only around 40 million were paying subscribers, resulting in a conversion rate of less than 5%. This means that for every 20 users served, only one person pays, while the other 19 continue to use the service and consume computing power just the same.

Traditional software also has a large number of free users, but the difference lies in marginal cost. One additional user of Photoshop imposes a negligible burden on Adobe’s servers. However, one more user of ChatGPT means GPUs have to run a little longer. In January 2025, Altman admitted on social media that even the Pro plan, priced at $200 per month, was still losing money for the company. The higher the charge, the more intensively users utilize the service; the more they use it, the higher the costs.

Some may argue that inference costs are decreasing every year, which is true. But at least for now, the cost inflection point has not yet arrived. Moreover, even if costs do decrease in the future, the same will apply to competitors, bringing the risk of price wars.

Beyond costs, the subscription model also requires a moat to lock users in. Adobe’s subscription model is stable because users’ workflows, file formats, and operational habits are deeply tied to Photoshop—switching software means relearning everything. Netflix’s subscription is sticky because it offers exclusive content; if you want to watch 'Squid Game,' you have to go through them.

Conversational AI lacks such binding. Whether it’s ChatGPT, Claude, Gemini, or Grok, opening any of these presents a dialogue box where you input questions and receive answers—it’s as if there are four or five versions of Windows on the market. Long-term use does create some沉淀of personal memory and preferences, but for new users, the differences are minimal.And the growth of the subscription model precisely depends on a steady influx of new users.

With limited headroom on the consumer side, OpenAI itself is shifting its focus toward the enterprise segment. In January 2026, Altman said on X (formerly Twitter) that people think we're mainly about ChatGPT, but our API team is doing an excellent job. The numbers indeed look good: by August 2025, the number of paying enterprise users grew from 3 million in June to 5 million, with over 1 million organizations using OpenAI's technology.

But the enterprise market has its own barriers to entry,AI deployments involving core businesses, data security, compliance audits, and private deployments—all are hard requirements.The reason cloud computing can succeed in the enterprise sector is that it sells infrastructure, leaving customers in control of their own data and logic; AI models are different—data must be fed in for them to work, which touches on the most sensitive nerves of enterprises. The rise of open-source models in the enterprise market is largely because they bypass this trust issue.

Altman's changing attitude towards advertising illustrates the issue more clearly. In 2024, he still said that advertising 'made him very uneasy,' but by January 2026, OpenAI officially announced testing ads in the free version. If subscriptions and enterprise services could truly support an $800 billion valuation, there would be no need to touch this area that once made him uneasy.

Moreover, embedding ads into AI may be much harder than embedding them into search engines. Google search ads work because search results themselves are a bunch of links—users have to click through to find answers. Ads mixed within these links, as long as relevant, are willingly clicked by users—Google sells the path to answers, and advertisers buy positions along that path.All conversational AIs, including ChatGPT, provide direct answers. If a user asks, 'Recommend a pair of noise-canceling headphones,' the AI might respond with Sony WH-1000XM5: strong noise cancellation, 30-hour battery life.

At this point,there’s no need to click on an ad anymore.

A deeper issue is that if users suspect recommendations are due to brands paying, the AI’s persona as an objective assistant collapses. OpenAI’s solution is to place ads below the responses, clearly labeled as 'Sponsored,' without affecting the content of the response or selling user data. Whether this will gain user acceptance is too early to tell; at least for now, it seems more like a forced attempt rather than a clear growth path.

02 AI is not SaaS

Over the past decade, the subscription model has been highly successful in the software industry. Companies like Adobe, Salesforce, and Slack have achieved steady growth through this model. The underlying logic is clear: software is an independent product. Photoshop is a design tool, Salesforce is a customer management system, and Slack is a team collaboration platform.Each has well-defined product boundaries, and users know exactly what they are paying for.。

As a venture, OpenAI also adopted this logic: ChatGPT is a conversational product where users subscribe monthly to gain more conversation credits and access to stronger model capabilities.

On the surface, it seems reasonable, but after several years, we have realized thatAI is spilling over the boundaries of a 'product.'。

By 2025, Microsoft has embedded Copilot into the entire Office suite. When writing in Word, AI helps refine your text; when analyzing data in Excel, AI assists with insights. In the same year, Adobe deeply integrated Firefly into Photoshop, allowing users to select an area and input a few words to generate or replace content. Google's Gemini has permeated every corner of Workspace, from Docs to Sheets to Gmail.

These scenarios share one common characteristic:Users don't come to 'use AI' directly; AI is simply a capability they invoke while doing other tasks.Polish a document while writing, fill in details while editing images, draft responses while replying to emails—use it and move on, sometimes without even realizing they're 'using AI.'

The premise of the subscription model is that users recognize the value of an independent product and are willing to pay for its continued use. However, when AI becomes a foundational capability embedded everywhere, this premise disappears. You wouldn't subscribe separately for spell-checking in Word or pay extra for automatic object removal in Photoshop. AI is becoming something omnipresent but no longer constituting a standalone product.

This may be the fundamental reason why the subscription model has failed in the AI field. It's not that OpenAI's costs are too high, or that its moat is too weak, but rather that the AI category itself is sliding from being a product towards becoming infrastructure—and the way to make money from infrastructure is different.

DeepSeek’s choice, to some extent, follows this logic to its conclusion.

DeepSeek has completely abandoned subscriptions, with all code and weights fully open-sourced, and API calls nearly free. Within seven days of the release of the R1 model in January 2025, DeepSeek gained over 100 million new users without any advertising, and for two consecutive quarters topped the list of domestic AI application monthly active users, nearing 190 million. From the three major telecom operators to China Southern Power Grid and PetroChina, from Lenovo and Huawei to various cloud providers, integrating with DeepSeek has become an industry standard.

DeepSeek is betting on this:When you become the foundation for enough enterprises, commercial value will naturally emerge elsewhere—whether from enterprise services, custom development, or some form that we can't yet clearly see.。

Alibaba’s Qwen is taking a different path. In January 2026, Qwen was fully integrated into Taobao, Alipay, Fliggy, and AutoNavi, allowing users to complete a closed loop from decision-making to payment with just one sentence. The AI itself does not charge, but Alibaba takes a cut from every transaction facilitated by AI. This treats AI as part of its own ecosystem’s transaction chain, rather than as an independent product.

ByteDance’s explorations are also worth considering. In December 2025, DouBao collaborated with Nubia to launch an AI-powered phone. When a user says, 'Help me order the cheapest bubble tea,' the AI can compare prices across platforms, place the order, and make the payment—all without opening any app. Although this feature faced a collective ban from various apps, it directly highlighted AI’s revolutionary potential within the existing interest chains of mobile internet.

No one knows whether these attempts will succeed, but their very existence indicates something: the subscription model, once considered the 'default answer' for AI commercialization, is now being questioned. If the essence of AI is infrastructure rather than a product, then measuring it by SaaS standards might have been wrong from the start.

03 Conclusion

History provides a reference point.

In 2007, Google announced that Android would be open-sourced. Wall Street was baffled: without licensing fees for the operating system, how would this project make money? At the time, Nokia's Symbian and Microsoft's Windows Mobile were both charging fees. Google’s response was: an operating system is not a destination, but an entry point; keep the entry point low-cost and profit from the ecosystem.

A decade later, Android holds over 80% of the global smartphone market share, and Google has built a commercial empire through search, advertising, and the Play Store, far larger than any licensing model could have achieved. Companies that insisted on selling their operating systems as products have become footnotes in history.

Today, what DeepSeek, Alibaba, and others are doing follows the same underlying logic—AI is not a destination but infrastructure; keep the infrastructure low-cost or free, and make money from the transactions and services it supports.

Of course, Android's success had its own specific historical context: the explosion of mobile internet, the wave of smartphone upgrades, and the application ecosystem evolving from scratch. Whether today's AI market is at a similar structural inflection point remains uncertain. Can DeepSeek's open-source strategy truly translate into commercial revenue? Can Alibaba's closed-loop transaction model change user habits? These are unknowns.

But one thing may already be clear:What remains uncertain is which infrastructure model will succeed. Subscription models alone are unlikely to define the future of large-scale AI models. $Alibaba (BABA.US)$$BABA-W (09988.HK)$$TENCENT (00700.HK)$$Microsoft (MSFT.US)$$Alphabet-C (GOOG.US)$

Disclaimer: This article is intended for learning and communication purposes only and does not constitute investment advice.

51K Views