The $15-a-Minute Problem: How AI Video's Flagship Product Rose and Fell in a Single Year

In the frenzy of the artificial-intelligence boom, it is easy to assume that every impressive demo is destined to become a lasting product. The strange, swift story of OpenAI's Sora is a bracing reminder that this is not always true. In roughly a year, generated video went from jaw-dropping showcase to discontinued service, and the reason has almost nothing to do with the quality of the technology and everything to do with money.

Sora was, by any measure, a technical marvel. Yet it is now being wound down, and its brief life offers one of the clearest case studies yet in the unforgiving economics that lurk beneath the AI industry's soaring ambitions. The lesson is that in this business, being astonishing is not the same as being sustainable.

A dazzling leap forward

The second generation of the model, Sora 2, was released to select users in the United States and Canada in late 2025 and represented a genuine step change. It produced video that was noticeably more physically accurate than earlier efforts, and crucially, it could generate synchronized dialogue and sound effects, moving beyond silent clips toward something resembling complete scenes.

For a moment, it felt like the future of media had arrived. The ability to conjure realistic, sound-complete video from a simple text prompt hinted at a wholesale transformation of advertising, filmmaking and social content, and the technology was showered with the kind of attention usually reserved for the launch of a major consumer gadget.

OpenAI clearly believed in the commercial potential, signing a landmark partnership reported to be worth around a billion dollars with Disney. The deal was designed to unlock the legal, licensed use of beloved characters in custom scenarios, pairing cutting-edge generation with the intellectual property and guardrails of one of the world's most powerful entertainment empires.

The bill that broke the model

So why is a product this advanced, backed by a deal this large, being shut down? The answer lies in a single, brutal number. Rendering high-quality video is astonishingly compute-intensive, and estimates suggested that each minute of Sora-generated content cost OpenAI roughly fifteen dollars in cloud infrastructure alone.

That figure is catastrophic for a consumer product. When users pay a flat monthly subscription and then generate minute after minute of video, the costs balloon far beyond what those subscriptions bring in. Every enthusiastic user effectively became a source of mounting losses, turning popularity itself into a financial liability rather than an asset.

The practical consequences were swift. The consumer Sora app was shut down in the spring of 2026, and the underlying interface for developers is slated to be discontinued in the autumn, effectively retiring the Sora brand as a whole. A product that had defined the frontier of AI video was quietly switched off, not because it failed to work, but because it worked too expensively.

What Sora's fall teaches us

The story punctures a comforting myth about this technological moment. It is tempting to assume that once an AI capability is demonstrated, its widespread adoption is only a matter of time. Sora shows that the gap between a working demo and a profitable product can be a chasm, and that the cost of the underlying computing power is the bridge many products will never manage to cross.

This is the quiet counter-narrative running beneath the AI gold rush. For all the talk of limitless potential, the industry is bound by very physical constraints: chips, data centres, and the enormous electricity bills that power them. The most advanced models can be so expensive to run that no consumer price makes them viable, at least not yet.

None of this means AI video is dead. The capability will almost certainly return, most likely in cheaper, more efficient forms or aimed at professional customers willing to pay for what they use rather than casual subscribers. The technology has proven itself; what remains unproven is the business model that can support it.

Sora's rise and fall, then, is not a story of failure so much as a preview of the reckoning coming for much of the industry. As the initial excitement fades, the companies that endure will be the ones that solve not just the hardest technical problems, but the hardest economic ones. In AI, as everywhere else, the bill always comes due.