On November 30, 2022, OpenAI released ChatGPT. In the three years since then, the whole world has been changed.
Currently, over 800 million people worldwide chat with ChatGPT every week, seeking advice or completing tasks. It is the fastest-growing app in human history, bar none. OpenAI, which recently underwent restructuring, is now a behemoth valued at over $500 billion, although it remains heavily indebted. Global tech companies are betting heavily on the future of artificial intelligence, pouring hundreds of billions of dollars into model training, computing infrastructure, and even operating system-level reconstruction.
On November 30, 2022, ChatGPT released...
OpenAI is the benchmark that all companies are aiming for, but its position is not secure. DeepSeek's breakthrough at the beginning of the year proved that even with limited computing power, Chinese tech companies remain competitive and maintain a visible lead in key technologies. Even traditional tech giants that have long been overshadowed by ChatGPT's disruptive impact are now slowly catching up.
Google is leading the pack. On November 18th, Google officially launched the Gemini 3 series of models, claiming it to be their most intelligent and fact-accurate model to date, trained entirely on their self-developed TPU chip. The Gemini 3 Pro immediately topped various model benchmark charts upon release, even surpassing OpenAI's newly released GPT-5.1 model. Over the past month, Google's stock price has risen by more than 11%, and its market capitalization has briefly approached $4 trillion.
Gemini 3 Pro tops various model testing charts
OpenAI founder and CEO Sam Altman responded to the launch of Gemini 3 with a tweet: "Congrats to Google on Gemini 3! Looks like a great model." Google CEO Sundar Pichai and Logan Kilpatrick, former head of developer relations at OpenAI, both interacted in the comments section.
This was Altman's public statement; internally, he was much more anxious. On Monday (December 1st), local time, Altman released an internal memo announcing that the company had entered "Code Red" status, requiring the mobilization of all company capabilities to focus on improving ChatGPT, and postponing the development of other projects such as advertising, agents, and even personal assistants.
"Red alert" is a common expression used by Silicon Valley companies when facing a crisis. Depending on the urgency of the event, there are different colored alerts such as yellow and orange, with red representing the highest level of threat.
Three years ago, when ChatGPT was first released, Pichai issued a "red alert" within Google, urging the entire company to bet on AI research and development, and even recalling founders Larry Page and Sergey Brin. Now, the tables have turned, and OpenAI is feeling the pressure.
However, there's no need to overemphasize the "red alert." Compared to real external threats, it's more like a tool for management to convey a sense of urgency internally, used to motivate employees' fighting spirit. If we apply this to the context of Chinese internet companies, Alibaba, which has been announcing "strategic projects" almost every month for the past six months, has already sounded the red alert countless times—and that doesn't even include the massive outage of Alibaba Cloud in the middle of the year.
In an internal memo, Altman revealed that they will release an inference model next week that outperforms the Gemini 3. According to The Information, OpenAI is also accelerating the development of a new model codenamed Garlic. The Garlic model reportedly achieves a breakthrough in pre-training, enabling the packaging of knowledge from large models into smaller architectures. The Garlic model is expected to debut as early as the beginning of next year.
Code Red and Product Company
Before issuing the "red alert," OpenAI's pace of productization and monetization was significantly more intensive than its iteration of underlying models. This made it increasingly resemble a traditional internet product company rather than a cutting-edge research company in the field of artificial intelligence.
In just two months, OpenAI has released a dazzling array of application-layer updates, including the Atlas browser, the conversational application ChatGPT, group chat and shopping research features, and the Agentkit intelligent agent building tool. You can now buy Walmart products, create Spotify playlists, and even book flights and hotels in ChatGPT, covering almost every aspect of life.
These numerous product feature updates are largely thanks to the addition of Fidji Simo. This internet executive, who joined OpenAI in May of this year, previously led Facebook apps at Meta and later served as CEO of Instacart, and is particularly skilled at building consumer-grade super apps.
After joining OpenAI, Simo became CEO of the newly formed OpenAI Applications division, leading product and business implementation, and reporting to Altman. Or to put it more directly, Simo was responsible for helping OpenAI monetize faster.
According to The Information, among those who indicate they currently work for OpenAI on LinkedIn, approximately 630 previously worked at Meta, representing about 20% of OpenAI's current workforce. Aside from a few technical staff, the vast majority are concentrated in commercialization areas such as product development, market growth, and advertising monetization.
A few days ago, it was revealed that OpenAI had added new code references related to advertising and commercial content to the Android beta version of ChatGPT. Some users also posted screenshots of shopping ads pushed by ChatGPT during chats, and the ad information was unrelated to the chat content.
Advertising code and chat ads in ChatGPT Android beta version
While OpenAI is accelerating its product monetization, other tech companies are intensively catching up at the model level. Google's Gemini 3 series models are just one example. Anthropic, a startup founded by former OpenAI executives, has just released its flagship model, Claude Opus 4.5, and the Chinese DeepSeek team has also launched the DeepSeek-V3.2 series models.
Unlike its competitors who emphasize performance breakthroughs, OpenAI's recently updated GPT-5.1 model is merely a minor tweak to the GPT-5 model, emphasizing more intelligent and conversational dialogue. The official blog doesn't even include performance comparison tests.
Simply focusing on personalized experiences isn't enough for OpenAI to maintain its lead. The rapid growth of Google Gemini over the past few months proves that model capabilities remain crucial in the AI competition. In March of this year, Gemini had only 350 million monthly active users, but by October, this number had jumped to 650 million. Many believe that Nano Banana was a key driver behind Gemini's surge in monthly active users.
Given Google's massive user base, the impact of Gemini 3's penetration within the Google ecosystem will be far greater. Unlike its predecessor, Gemini 3 was integrated into Google Search upon release. With comparable model capabilities, this distribution advantage could drive more users from ChatGPT to the Google ecosystem.
Deedy Das, a partner at Silicon Valley investment firm Menlo Ventures, cited Similarweb data, stating that in the two weeks since Gemini 3 launched, ChatGPT's daily unique active users (7-day average) have decreased by 6%. During the same period, Gemini usage saw a significant increase, rising from 22% to 31% of ChatGPT's traffic.
Deedy Das shared a screenshot
In summary, under the dual pressure of model performance and user growth, Altman internally declared a "red alert" state. The Wall Street Journal stated that this notification is the clearest indication of the pressure OpenAI is facing from competitors, "who have narrowed the startup's lead in the AI race."
OpenAI's previously aggressive product feature updates may be temporarily suspended. In an internal memo, Altman announced that the company's resources would be focused on improving the user experience of ChatGPT. Other non-essential projects will be postponed, such as advertising, AI agents for health and shopping, and a personal assistant called Pulse.
Altman's priorities all revolve around ChatGPT, including enhancing personalization, improving image generation capabilities, increasing speed and stability, and reducing model over-rejection. Of course, improving GPT's ranking on public benchmarking platforms like LMARaena is also crucial.
OpenAI has more realistic concerns.
It's difficult to say that OpenAI is currently facing the risk of being disrupted. Even with the aggressive moves of competitors like Gemini 3, ChatGPT remains the AI application with the largest user base globally.
According to Nick Turley, head of ChatGPT, ChatGPT currently accounts for 70% of global AI assistant usage and 10% of search activity. "Our current focus is on continuously improving ChatGPT's capabilities, growing it, and expanding its reach globally."
But all of this is based on OpenAI maintaining its leading model performance. Since the release of GPT-5, the gap between OpenAI and other technology companies has been narrowing, and ChatGPT's user growth has slowed down.
According to data from third-party data company Apptopia, ChatGPT's daily active user growth on mobile devices has slowed significantly since April and has stagnated since September. Meanwhile, ChatGPT's average daily active user time spent on the platform in the US has decreased by 22.5% since July, and the average number of conversations has also decreased by 20.7%.
Apptopia statistics
Deutsche Bank, after tracking financial transaction data in the European market, also found that ChatGPT's monthly growth rate of consumer spending has slowed significantly since May 2025, suggesting that ChatGPT's paid user growth in Europe may have stagnated. This may be one of the reasons why OpenAI's commercialization efforts have gradually accelerated over the past year.
At the same time, OpenAI is maintaining unprecedented spending on AI infrastructure. Following its profitable restructuring, Sam Altman announced that OpenAI would significantly expand its infrastructure investment, with the ultimate goal of investing $1 trillion annually in infrastructure.
To date, OpenAI has committed approximately $1.4 trillion to infrastructure development. Its partner list includes a host of giants such as Nvidia, AMD, Oracle, and Broadcom. These companies often see their stock prices surge as a result of deals with OpenAI.
American tech media are particularly concerned about whether OpenAI's financial situation can support such massive capital expenditures. Unlike traditional giants like Google, which have existing businesses to support them, OpenAI's funding comes almost entirely from venture capital, and it is still operating at a huge loss.
Recently, OpenAI's CFO, Sarah Friar, expressed support for government guarantees for AI infrastructure financing, which sparked considerable controversy in the United States. Criticism arose that OpenAI was attempting to get the government to foot the bill for its massive AI investments. Friar subsequently issued a statement clarifying that OpenAI had not sought government guarantees.
Clarification statement issued by Sarah Fryer
Sam Altman also had to issue a lengthy response, stating, "We do not need and do not want the government to guarantee OpenAI's data centers." Altman said that OpenAI will rely on its own revenue growth, external investment, and future sales of computing power to support its computing power investment.
According to him, OpenAI's revenue is projected to exceed $20 billion this year and grow to hundreds of billions of dollars by 2030. Even so, this revenue will still be insufficient to cover OpenAI's current capital expenditures.
Sam Altman's lengthy clarification statement (the original text is quite long; the screenshot is only a portion).
Altman believes that now is a critical time for OpenAI to expand its investment in AI infrastructure, and that it must invest heavily now rather than developing it gradually. "Large-scale infrastructure construction takes a considerable amount of time to complete, so we must act."
He stated that OpenAI faces a greater and more likely risk of insufficient computing power than of excess computing power. "Everything we are seeing right now indicates that the world will need far more computing power than we have planned."
The creation of ChatGPT is a testament to the power of brute force. Before ChatGPT, most mainstream models had fewer than 300 million parameters. In 2022, OpenAI increased the number of parameters in GPT-3 to 175 billion, finally demonstrating the emergent intelligence capabilities of large models. At the end of the year, OpenAI released ChatGPT based on the GPT-3.5 model, officially ushering in the AI era.
Now, Sam Altman is attempting to maintain OpenAI's lead by building even more massive computing power. Until everything is finalized, it's difficult to judge whether OpenAI's choices are correct. But at least, it has already attracted enough people to join it.
“Our goal is to become an extremely successful company, but if we make mistakes, the responsibility lies with ourselves,” Sam Altman said.
This article is from the WeChat public account "Shanshang" , author: He Jian, editor: Jiang Jiao, and published with authorization from 36Kr.
