At the beginning of December 2025, Sam Altman declared a "red alert" at OpenAI in San Francisco: focus all efforts on ChatGPT, as Google pushes Gemini 3.0 (including the Pro version) and energy and financial costs skyrocket. Projects put on hold, strategy disrupted, space ambitions lurking. What does this refocusing mean for users, digital employment, ecological footprint, and AI governance?
What the "red alert" memo reveals
At the beginning of December 2025, an internal memo attributed to Sam Altman, CEO of OpenAI, circulated in San Francisco. It mentions a red alert situation centered on ChatGPT, with a directive: refocus efforts on the company’s most visible consumer product. Teams are encouraged to prioritize speed, reliability, personalization, and topic coverage, at the expense of peripheral projects.
This document, reported by several media outlets, is set against a backdrop deemed "turbulent." In the background: the upscale move of Gemini 3.0 at Google/Alphabet, perceived internally as an immediate risk to ChatGPT’s attractiveness. Management argues that every delay, every unpolished feature, can shift cohorts of users to the competition.
Why Google worries OpenAI
According to this internal literature and press cross-references, Gemini 3.0, especially in its Pro version, has exceeded certain expectations and benchmarks. The ecosystem is expanding, including on the video side with the Nano Banana tool. As a result, some advanced users are increasingly testing Google’s offering to the detriment of ChatGPT.
In terms of volumes, Google claims about 650 million monthly users for Gemini, while OpenAI mentions more than 800 million weekly users for ChatGPT. Two different metrics, but a catch-up dynamic is emerging. In a battle of adoption and habits, the slightest performance differential can weigh heavily.
OpenAI: projects suspended – Pulse, advertising, and agents
The "red alert" comes with postponements. The research assistant Pulse (OpenAI) is put on hold. The advertising formats envisioned for ChatGPT and for GPTs are postponed. Other vertical agents (shopping, education, health, professional niches) see their schedule slip. The directive is clear: focus engineering on the core of the chatbot.
This choice delays the financial thesis that attributed about 20% of future revenues to advertising and secondary agents. The company emphasizes immediate usage quality. Consequently, it aims to curb attrition and maximize retention. However, it does not hesitate to defer more experimental monetization levers.
OpenAI: finances 2025, trajectory under tension
According to estimates relayed by the specialized press, OpenAI would have posted record losses in the first six months of 2025. Furthermore, these losses would be around $13.5 billion for $4.3 billion in revenues. This is despite a valuation nearing $500 billion. Other analyses estimate OpenAI’s profitability pushed to 2029.
Even more ambitious, plans for data centers valued at over $1.4 trillion by 2030 are mentioned, involving cumulative fundraising exceeding $200 billion. These amounts are due to infrastructure costs: cloud, chips (Nvidia, AMD), energy, and cooling. The assumed strategy is one of hypergrowth financed by strategic partners. Among these partners are Microsoft and SoftBank as well as the markets.
Sam Altman and Stoke Space: the space track
The other thread of the story is akin to applied science fiction. In the summer of 2025, Sam Altman reportedly approached Stoke Space, a Seattle startup designing reusable rockets. The hypothesis being studied: a majority stake, worth several billion dollars, to build a competitor to SpaceX. The discussions have stalled, with no agreement at this stage.
Beyond the publicized rivalry with Elon Musk (co-founder then opponent of OpenAI, head of SpaceX and xAI), the strategic calculation aims at a down-to-earth goal: securing, in the long term, new computing capacities. The data centers could follow the rockets.
Data centers in orbit: solution or mirage?
The idea of space data centers has been recurring for years in tech literature. In orbit, cooling would benefit from a more stable environment. Additionally, solar production would be abundant. Furthermore, certain latencies could be optimized for specific use cases. Conversely, there are launch costs to consider. Similarly, maintenance constraints and space debris would pose a problem. Finally, questions of sovereignty and space law are added.
On the ecological front, the argument of a potentially better water and energy balance deserves independent proof. Transporting heavy equipment into orbit at a significant initial carbon cost. The debate remains open: a future solution for sustainable computing or a narrative intended to reassure investors?
Risks, responsibilities, and procedures
On November 6, 2025, seven plaintiffs filed a lawsuit in California for involuntary manslaughter. Additionally, the accusation of assisted suicide was brought after the death of four users. This death is attributed to the use of GPT-4o. These are allegations made in the context of an ongoing procedure. At this stage, no court decision confirms these claims.
In the internal memo as reported, no new explicit measure is mentioned for vulnerable audiences. This point will likely raise political questions and a debate on the safety of systems. Moreover, the prevention of suicidal risk will be discussed as well as the duty of care of platforms.
What this changes for the public
For users, the immediate effect could be positive: a faster, more reliable, more customizable, and accessible ChatGPT for the greatest number. But the ecosystem is concentrating around a few players with an unmatched scale advantage. Access to high-performance AI assistants becomes a consumer good. Moreover, it is also a public service issue in fact.
On digital employment, the battle of assistants accelerates the demand for orchestration skills (advanced prompting, integration, quality control) and monitoring of drifts (hallucinations, biases, security). On the energy side, the trajectory of data centers and chips promises increasing consumption, hence the pressure for real efficiencies here, orbit may not be the miracle solution.
Act 1: the alert in the corridors
In the corridors of OpenAI, one imagines the dashboards turning orange: tickets pending, latencies stretching, feedbacks piling up. The "red alert" sets a tempo. Cut what distracts, polish the core of the experience. In this endurance race, the first meter counts as much as the last.
Act 2: the billionaire who dreamed of rockets
In Seattle, Stoke Space dreams of circular engines and second stages that come back to land. Sam Altman sees further: computing in orbit, network above the clouds, cooling without fresh water. The plan has the madness of pioneers and the arithmetic of Excel sheets. Between vision and dilution of focus, the line is thin.
Act 3: what remains?
The question that obsesses investors and regulators remains: can OpenAI hold on until 2029 without profitability? Do the massive fundraising bet, the idea of data centers valued in trillions of dollars, and the product pauses make a credible path? Or are we witnessing the story of a bubble seeking its balance?
In a democracy, the answer cannot be left to the markets alone. It calls for safeguards: transparency on energy costs, independent risk assessment, interoperability and portability to avoid platform lock-in, active protection of vulnerable users.
Open questions
- Can OpenAI finance its trajectory by 2030 without breaking its main product?
- Are space data centers a matter of serious engineering or storytelling for sovereign funds and major investors?
- Does this assistant war benefit the public (quality, price, rights) or reinforce our dependence on a few platforms?
- Where are the additional protections for fragile audiences in this permanent urgency?
Methodological note: this article is based on internal documents reported by the press and on analyst estimates. The complaints mentioned concern ongoing procedures. Thus, the alleged facts cannot be considered established. Moreover, this remains true as long as no final court decision is rendered.