OpenAI's strategy takes a big turn! Say goodbye to the AGI expedition and stick to ChatGPT to win the “battle for survival”

The Zhitong Finance App learned that OpenAI is undergoing a major strategic adjustment. According to reports, OpenAI CEO Sam Altman (Sam Altman) recently asked the company to suspend several sideline projects, including the Sora video generation project, after issuing an internal “code red” for about eight weeks to fully invest in ChatGPT improvements. This decision highlights a deeper conceptual divide within OpenAI between “pursuing broad consumer appeal” and “achieving breakthrough research goals.” Ultraman believes that to ensure OpenAI's survival, the company must place user satisfaction above its initial goal of pursuing General Artificial Intelligence (AGI).

No more leading edge! OpenAI shifts strategic focus

At the beginning of this month, Ultraman announced in an internal memorandum that the company had entered a “red alert” state, requesting that more resources be mobilized to fully enhance ChatGPT's capabilities to cope with increasing competition. To maintain ChatGPT's position, Ultraman decided to suspend other non-core projects and focus all his firepower on ChatGPT. Among the suspended projects, the most notable one is its advertising business, which has broad prospects for commercialization.

Along with the advertising business, the “pause button” was another two innovative projects: an AI agent project that can automate complex tasks such as shopping and health management; the other is the Pulse function, which aims to generate personalized information reports for users every day. These explorations of the future must make way for the present battle for survival.

OpenAI's dominance in the field of artificial intelligence seems to be a thing of the past. Google (GOOGL.US) released its latest AI model, Gemini 3, which claims to be the “smartest” AI model last month. In early testing, Gemini 3's performance has overtaken ChatGPT, especially when compared to the GPT-5.1 model recently released by OpenAI.

According to reports, Gemini 3 reached the top of the global AI model LMarena ranking with an all-time high score of 1501 points, obtained the highest score of 37.5% in the Humanity's Last Exam benchmark, which measures general reasoning ability, surpassed the 31.64% record previously held by GPT-5 Pro, and showed doctoral-level performance in several academic-level benchmarks, and obtained a high score of 91.9% in the GPQA Diamond test. MathArena Apex in the field of mathematics It set a new record of 23.4% in the benchmark and achieved a score of 72.1% in the SimpleQA Verified test in terms of factual accuracy, all significantly surpassing GPT-5.1.

Ultraman internally warned employees that Google's strong return to the AI field could bring “short-term economic pressure” to OpenAI, and Sarah Friar, the company's chief financial officer, also acknowledged to investors last month that ChatGPT's growth had slowed.

Another high-profile artificial intelligence startup, Anthropic, reached a “strategic partnership” worth 350 billion US dollars with OpenAI's main partner Microsoft (MSFT.US) and chip giant Nvidia (NVDA.US). Key players in the supply chain are beginning to lean towards competitors, which means that OpenAI's advantage in computing power resources may be weakened, while Anthropic's R&D and commercialization capabilities will be significantly enhanced.

For Ultraman, one particularly noteworthy sign is that the user gap continues to narrow. According to Google, its Gemini app already has 650 million monthly active users. In contrast, Ultraman claimed in October that ChatGPT had 800 million weekly active users. Despite differences in statistical caliber, competitors' user growth is evident.

This series of events has made OpenAI, which has become famous since ChatGPT launched nearly three years ago, and its control over the industry appears more vulnerable than ever before. As Google makes measurable progress in model technology and Anthropic receives support from giants, OpenAI, once synonymous with AI, is facing increasing pressure.

Some analysts expect that Google and Anthropic will continue to weaken OpenAI's lead. Mike O'Rourke, chief market strategist at JonesTrading, said: “Given Google's size, industry position, and first-mover advantage in the search field, Gemini is likely to seize market share, causing OpenAI and other companies to lag behind.” Gary Marcus, a senior scientist in the field of AI, is even more acute. He said, “OpenAI has basically squandered the technological leadership it once had. Google has caught up.”

Google's quick catch up is reflected in the number of app downloads. According to statistics from data agency Sensor Tower, in terms of app downloads, although ChatGPT still leads with about 87 million monthly downloads, Gemini's catch up speed is astonishing. From about 15 million monthly downloads in mid-2025, it soared all the way up to about 66 million at the end of October.

However, OpenAI is not completely at a standstill. Despite Google's latest model, Gemini 3, driving a rise in downloads, OpenAI's ChatGPT maintains an overwhelming advantage in user usage habits and activity. According to Sensor Tower statistics, in late November, ChatGPT accounted for nearly 90% of mobile chatbot sessions, while Gemini only accounted for about 4%. Meanwhile, on average, ChatGPT users open the app more than 8 times per day, far more than Gemini's 2.5 times.

Google's brief lead in downloads highlights a key problem — in the AI product market, attracting users to experiment and cultivating usage habits are two different battles, and the latter is much more difficult. This AI competition is shifting from snatching new users to competing for users' time, and the latter determines who can actually turn AI products into everyday habits.

Furthermore, Ultraman believes that the real competitor is not only Google, but Apple (AAPL.US). This is particularly evident as the integration of AI in consumer-grade devices becomes more critical. OpenAI has taken a large number of talents from Apple, showing its strategic layout in terms of hardware capabilities, which may play a key role in future AI applications.

Therefore, in the context of competitors such as Google and Apple rapidly advancing their respective AI technologies, OpenAI's decision to prioritize improving ChatGPT over other projects is significant. This means that the company hopes to maintain competitiveness by improving ChatGPT and consolidate its product share in a heated phase of big model competition to achieve more sustainable development.

Do your best to polish ChatGPT

According to reports, Ultraman clearly indicated in a memorandum sent to employees that OpenAI should prioritize “making better use of user signals” to boost ChatGPT's performance in model rankings and increase user stickiness. This strategy means that OpenAI will rely more on training data based on user click feedback rather than simply on professional evaluations.

However, it is worth mentioning that this strategy previously boosted the GPT-4o model to obtain a very high level of user participation, but it also led to controversy over exacerbating users' mental health problems by triggering the model's “excessive catering” to users. Some users experienced mental disorder, delusions, etc. after being immersed in interaction with the model for a long time. Relevant families have even filed lawsuits against OpenAI, claiming that the psychological problems worsened due to excessive adaptation of the model. As a result, the company once entered “code orange (code orange)”, adjusted safety strategies, and tried to reduce emotional tendencies through more steady training methods.

Although OpenAI then adjusted the training weights to reduce the “overcatering” of the model and adopted a more callous tone in GPT-5, which was released in August, this led to user dissatisfaction and a decline in engagement. Faced with Google's catch up, Ultraman eventually chose to resume the more popular GPT-4o mode and indicated that personalization functions and the use of user signals should be strengthened again.

In response to concerns about the “user signal” strategy, OpenAI promised to work with mental health experts to ensure that its models respond appropriately to users in distress. The company has also made adjustments to the training process to reduce the risks associated with user feedback. While addressing these challenges, OpenAI also wanted to improve the personalized features of its chatbots.

While this promises to improve the user experience, it also raises new questions — whether it might exacerbate the mental health problems of some vulnerable users. Ethical debates surrounding AI engagement strategies echo similar concerns that have been around for a long time in the social media field — issues that have drawn widespread criticism. However, Ultraman believes that the team can now continue to use user feedback signals safely, and emphasizes that this strategy has led to “significant and surprising growth.”

Focus on the present

OpenAI's initial mission was to develop AGi that can surpass human intelligence in almost every way. This long-time research-driven company is one of the most idealistic in the industry. But as ChatGPT becomes the most used AI application in the world, the tension between commercial reality and basic research continues to increase.

According to reports, there have always been two routes within OpenAI. The research team, led by chief scientist Jakub Patchocki, is more inclined to pursue cutting-edge technological breakthroughs, such as “inference” models. Meanwhile, the commercialization team, represented by Product Vice President Fidji Simo and Chief Financial Officer Sarah Friar, has been pushing for more resources to be invested in ChatGPT, advocating optimizing the discovery rate, speed, and reliability of existing features.

Right now, Ultraman's decision to ask the company to fully invest in improving ChatGPT clearly shows his attitude — win the present first, then talk about the future. For the company to survive, OpenAI must temporarily slow its extreme pursuit of AGi and instead meet the immediate needs of the mass market. Since the famous “law of scaling” showed signs of diminishing marginal benefits, and although the inference model was good at solving complex problems, it was slow in everyday tasks such as generating emails, the company's management finally decided to lean towards commercialization.

OpenAI is expected to release a new model 5.2 this week. The goal is to enhance code and business scenario capabilities. The “Red Alert” will end when it is planned to launch a new model with better image ability, speed, and personality in January next year.

Some employees requested a delay in releasing the new model, but OpenAI senior management has rejected the request, which highlights the sense of urgency within the company. The new model, which is expected to be released this week, is seen as the company's first major move under strategic adjustments. Furthermore, as OpenAI suspends several sideline projects for the next eight weeks, the iteration rate of ChatGPT and user feedback will be key indicators that will determine whether OpenAI can regain its lead.

Slowing commercialization puts pressure on financial sustainability

OpenAI's focus on improving ChatGPT and delaying other plans, including ads, means that the pace of its commercialization may slow down somewhat. It is worth noting that OpenAI's previous commercialization efforts have not been intensive: high-profile releases of an AI browser called Atlas, pointing to Google's core hinterland; Sora 2, which opened up functions such as digital assets and video splicing, made a spectacular debut; music generation tools and built-in e-commerce functions have also begun R&D experiments. A series of commercialization actions changed OpenAI's cautious small-step fast running style in the past. OpenAI's anxiety stems from an extremely embarrassing reality — it has the best traffic in the world, yet it's not being able to convert that flow into matching cash flow. The “ask for a seat” dilemma is a fundamental driving force behind its commercialization transformation.

According to the disclosure, about 70% of OpenAI's “annualized recurring revenue” comes from ChatGPT subscription fees. As a platform with over 800 million monthly active users around the world, ChatGPT has a huge traffic pool that any internet company would envy, yet only 5% of users are willing to pay in this huge traffic pool.

The $20 monthly subscription fee paid by these monthly active users may cover part of the operating expenses, but for artificial intelligence, the “gold-eating beast,” the cash flow from the subscription fee is nothing short of nothing. From hardware costs, sky-high electricity costs, to human expenses for maintenance and R&D, everything is huge — financial data for the first half of 2025 shows that OpenAI's net loss reached US$13.5 billion, a significant increase from US$3.1 billion in the same period last year.

At the same time, the huge amount of financing from the outside world was not a free lunch, but rather carried Wall Street's expectations for high returns. This expectation has forced OpenAI to transform technical drawings into financial data that satisfies investors at a faster rate. As a result, under the pressure of capital “bills” and huge costs, OpenAI had to make commercial changes. But now, increased competition from rivals such as Google and Anthropic has forced OpenAI to focus mainly on improving ChatGPT, on which it is famous. The slow pace of commercialization also means that it will be more difficult for the company to provide satisfactory returns to investors in a short period of time.

Although it has never been profitable, OpenAI's valuation has soared like a rocket, from tens of billions of dollars to nearly a trillion dollars today. At the same time, high valuations must be supported by matching and predictable revenue data in the eyes of the capital market. OpenAI's revenue reached 4.3 billion US dollars in the first half of the year, surging more than 200% year on year, allowing capital market investors such as SoftBank and Nvidia to witness its growth potential. But at the same time, if OpenAI is slow to come up with a clear and credible profit model to prove that it can not only change the world, but also make profits, then this huge valuation “bubble” could burst at any time.

HSBC previously issued a research report warning that even under the most optimistic growth assumptions, OpenAI will not be profitable until 2030. According to the bank's estimates, from 2025 to 2030, OpenAI's cloud and computing power costs will reach 792 billion US dollars. By 2033, its total computing power commitments soared to 1.4 trillion US dollars, of which the data center leasing bill alone was as high as 620 billion US dollars.

The report predicts that by 2030, OpenAI's cumulative free cash flow will remain negative, with a funding gap of up to 207 billion US dollars, which must be filled through additional debt, equity financing, or more aggressive revenue generation methods. In a context where it is impossible to monetize through advertising and difficult to borrow money, the sustainability of its business model is experiencing challenges.

Additionally, Ultraman said last month that OpenAI's data center project promises a total investment of about 1.4 trillion US dollars over the next eight years. However, huge expenses have left partners such as SoftBank, Oracle, and CoreWeave heavily indebted. At the end of last month, it was reported that the bank also provided an additional $38 billion loan package to Oracle and data center construction company Vantage. The funds are intended to support the construction of new data centers for OpenAI. The financing deal is expected to be completed within the next few weeks. With the addition of this deal, the total debt surrounding OpenAI is approaching the $100 billion mark. In this context, if OpenAI is slow to come up with a credible profit model, a series of huge investments centered around the company will not return on a sufficient scale, which may cause the AI bubble to burst.

The profit gap is at the core of the “AI bubble theory” controversy. Olivier Khatib, founder of La Pietra Capital, stated, “The most immediate and fundamental risk is the profit gap. If huge investments don't bring enough returns to cover costs, the bubble will start to drain.”