OpenAI CEO Sam Altman: No IPO in 2026
OpenAI CEO Sam Altman confirmed in an interview with Fortune that the company will not go public in 2026, stating that given the current AI safety situation, now is not the right time for an IPO. He clearly responded "not in 2026" and indicated that there is no rush to enter the public market.
Altman mentioned that going public should wait until the business is ready and society can adapt to AI models of varying capabilities. He emphasized that the company needs to prioritize safety and alignment, and may slow down the development of cutting-edge models as capabilities enter a new phase. In the interview, he also discussed the "absolute" risk of losing control and the commitment to halt training if safety thresholds are not met. When asked if a trillion-dollar IPO is still a priority, he reiterated, "We are not in a hurry to IPO," stating that going public under the current safety circumstances is not wise.
Previously, OpenAI had submitted a confidential S-1 to the U.S. SEC around June 2026, with internal discussions about a window in the third or fourth quarter of that year. Advisors provided two paths: going public in 2026 at a lower valuation or holding out for a valuation of about $1 trillion until 2027. Multiple reports indicate that Altman considers anything below $1 trillion unacceptable. The company's most recent private valuation was approximately $852 billion, supported by around $122 billion in funding from investors like SoftBank, Amazon, and Nvidia. CFO Sarah Friar had previously told employees that the target is set for 2027, but it could be moved up if business improves significantly.
In the same week, Anthropic CEO Dario Amodei published an article titled "We Must Pace the Frontier," calling for a slowdown in the escalation of frontier capabilities and unilaterally committing to grant third-party evaluators permanent, near-employee-level access to systems to verify safety measures, report incidents, and assess alignment during training. Altman subsequently agreed on the need to control frontier progress, stating that this has been a major topic for OpenAI in recent weeks, and promised that independent evaluators would receive access similar to employees, with details to be announced soon. Elon Musk also publicly stated, "Dario is right."
Amodei's plan consists of three steps: embedded evaluators, coordination of safety benchmarks among democratic nations, and government requirements for other frontier companies to match. He cited organizations like METR and proposed providing external teams with workspace, access, and company laptops, with permissions close to internal risk control. OpenAI has yet to disclose the list of evaluation agencies, starting times, and permission comparison charts. In the past two weeks, members of Anthropic's safety team have resigned, warning that the competition could lead to systems that humans cannot control.
In market mechanisms, this is an event-driven re-pricing of both the funding window and safety narrative. The buyers are strategic capital and computing power suppliers who can still increase their stakes in private markets and are not in a hurry for liquidity; the sellers are those who view a 2026 IPO as an exit point and discount employee equity. The beneficiaries are OpenAI and Anthropic, which can continue to sign long-term computing contracts as private companies and avoid quarterly disclosures of training interruptions; the pressured parties are options priced for a 2026 IPO, shareholders like SoftBank who have already booked high valuations, and AI companies that need to benchmark publicly after going public. Funds will not immediately withdraw from training clusters but will shift from "pricing issuance this year" to "opening books after safety governance is verifiable."
ABAB AI Insight
OpenAI's journey from a non-profit lab to Microsoft's investment and the widespread release of ChatGPT, followed by submitting an S-1 in 2026, has always been about "first scaling capabilities, then absorbing computing bills through private markets." Altman stated at the end of 2025 that his excitement about becoming a public company CEO was "0%", while acknowledging that the limit on the number of shareholders would eventually force publicization. Removing the IPO from 2026 effectively ties the exit clock to the safety narrative rather than the revenue curve. The company is still losing money and locking in huge computing commitments, with the private market valuing it at about $852 billion, while the public market requires it to prove its trillion-dollar valuation.
The capital path is to delay securitization and continue to fund through private placements and cloud contracts. Microsoft, Nvidia, Amazon, and SoftBank are sending chips, cloud services, and cash into training clusters in exchange for ecosystem binding and future equity. Adding a layer of "employee-level external evaluators" means giving some R&D knowledge rights to third parties in exchange for regulation and social license. The motivation is dual: to avoid every training interruption becoming a stock price event after going public, and to hedge against competitors like Anthropic defining safety standards first. Resource mobilization shifts from investment bank roadshows back to safety teams, evaluation agencies, and policy communication.
A similar structure occurred in the social media sector around 2018 after Cambridge Analytica, where content moderation was enhanced, and in the nuclear power industry where on-site supervision was exchanged for operating licenses. The volatility of SpaceX post-IPO was used by advisors to argue that the "super-large tech IPO window is narrowing." The industry phase is shifting from expansion to control: model capabilities are still increasing, but pricing power is beginning to depend on who can allow outsiders into the training pipeline. Anthropic is positioning itself with unilateral commitments, while OpenAI is using reciprocal commitments to prevent standards from being monopolized by competitors.
The structural judgment is that regulatory changes are compounded by capital concentration. The mechanism is that as frontier models begin to participate in the improvement of next-generation models, if private companies are immediately exposed to quarterly reports and short-selling research, any alignment failure will directly discount their trillion-dollar valuation. Thus, leading labs choose to postpone securitization and advance oversight rights to the training process. Whoever controls the access to "seeing training like an employee" will define the next generation of licenses, rather than who rings the bell first.
ABAB News · Cognitive Laws
- Safety narratives can delay securitization but cannot eliminate bills.
- Let outsiders into the machine room before discussing trillion-dollar pricing.
- Slowing down in a competition is often to seize definitional power.