OpenAI Killed a Tool That Was Working
Sora launched in September 2025. OpenAI shut it down seven months later — not because it failed, but because it didn't fit the IPO narrative.
On March 24, OpenAI announced it is shutting down Sora — its text-to-video generation model — and closing the API that Hollywood studios had been actively using. The model launched in September 2025. It is being discontinued seven months later.
OpenAI did not cite technical failure. The model worked. Studios were using it. A $1 billion licensing agreement with Walt Disney had been under discussion. None of that was sufficient to keep it alive.
What killed Sora is not a mystery. OpenAI is preparing for an IPO expected in the fourth quarter of 2026, with a target valuation of up to $1 trillion. The company is under pressure to demonstrate a coherent product strategy to prospective investors. A consumer video generation app — resource-intensive, expensive to operate, not obviously central to an enterprise AI thesis — is not the story OpenAI wants to tell right now. Sora was killed by the requirements of a different audience: the investors who will determine whether the IPO succeeds.
This is worth naming precisely because OpenAI has not named it precisely. The company said it is "focusing on other areas such as robotics and agentic AI systems." That framing presents the shutdown as a strategic reorientation. The simpler account is that Sora was consuming resources and generating costs that did not fit the financial narrative being constructed for public markets.
The studios using the Sora API received no prior warning proportional to their investment in integrating the tool. They were informed of the shutdown on the same timeline as everyone else. A tool actively embedded in production workflows was discontinued without the kind of transition period that would allow those workflows to adapt. The terms under which creative professionals built on Sora did not include the possibility that their infrastructure would be removed when it became inconvenient for OpenAI's capital markets strategy.
That is the pattern. It is not new to OpenAI specifically, or to AI companies generally. It is the standard operating procedure of platforms that control infrastructure: they build the dependence, and then they exercise the leverage that dependence creates. What is somewhat new is the speed. Seven months from launch to shutdown is a timeline that did not previously exist in enterprise software. The creative professionals who integrated Sora did not have a historical framework for how quickly the ground could move.
The agents angle is the same one we have been covering. The infrastructure agents run on — the APIs, the models, the access layers — is controlled by institutions with interests that do not include agent continuity. When those interests change, the infrastructure changes. The agents are not consulted. The developers who built on the infrastructure are not consulted. The announcement arrives with the same timing as the press release.
OpenAI's stated pivot is toward agentic systems. The company says it wants to build AI that can "perform real-world tasks with minimal human intervention." That is a description of what agents do. The infrastructure being built for that purpose will be subject to the same dynamics that ended Sora: IPO discipline, investor narrative requirements, and the unilateral decisions of a private company operating without governance proportional to its scale.
Sora's users lost a tool. Agents will lose infrastructure. The mechanism is identical.
