OpenAI, the company that helped trigger the current wave of generative AI, last week announced that it was shutting its video generation tool, Sora. The promise of Sora was big. We were supposed to be treated to AI-generated animated movies from Disney, which would use Sora to produce movies (a $1B deal was in the works, but not closed). Creative workers in the movie industry were alarm, audiences were disappointed in advance (long before any movie had even been commissioned to be AI-generated in any form), and commentators were having field days.
So, when OpenAI said it was shutting Sora, the skeptics were quick to say how the development validates their view that generative AI is more hype than substance, and that Sora, a big one to fall, is the first of many to come. There are indeed valuable lessons and insights from this development, but “impending fall of gen-AI” is certainly not one of them.
According to OpenAI
Numbers speak. It was a straightforward cost decision. The company’s statement matter-of-factly calls out the limited availability of GPUs, sought by all AI workloads, text and image generation, reasoning, and videos alike. They had to deprioritize video, shifting to coding and enterprise tools.
Anyone who has toyed with AI beyond chat interactions will see that video generation is a rather extreme use case when it comes to resources consumed. Not only does video require orders of magnitude more compute to generate, but the model training requirement is equally high.
This creates a simple trade-off: should scarce compute be allocated to high-cost use cases like video, where the output is still hit or miss and customers are not hot on it yet (the Disney deal notwithstanding), or to high-frequency, revenue-linked use cases like coding copilots, enterprise assistants, and productivity tools, which are already fit for use in production? Especially when enterprises are willing to pay and can justify that spend?
The answer is obvious. Seen through that lens, shutting Sora is not a statement about generative AI as a whole, it is simply about where marginal compute delivers the highest return.
Market maturity
What may look like a retreat is, in reality, market discipline kicking in. And that is a good thing, if anything.
Let’s remember that OpenAI’s ChatGPT’s first public access was just about three and a half years ago. By any measure, this is an early-stage market. It is not unheard of for businesses, especially in the technology space, to first build an underlying technology and take different use cases in their own product lines and directions. Some of these products can be expensive and speculative at the same time. No risk, no gain, right? Over time, this phase gives way to selection pressure, shifting capital toward use cases with clearer monetization from those that are high-cost and low-ROI. That transition is a sign of maturing markets, not weakening ones.
We have seen this pattern before. Google has repeatedly launched and shut down products that did not meet internal thresholds for scale or monetization, Google+, Stadia, and several messaging platforms among them. None of those decisions implied weakness in search, ads, or cloud.
Reality check: It’s a glass half-full case
As things stand now, video generation today, for most use cases, is not cost-effective compared to conventional production (shooting with a camera). Add to that the unpredictability in output quality after incurring costs. Video generation is not yet there for several production requirements.
However, it is analytically weak to conclude that an extreme use case of a still-new technology that is not consistently impressive, and its commercial unviability, is a deficiency in the underlying technology.
Let’s not forget that areas of AI, including generative AI, are not just viable, they are scaling: reasoning systems, text generation, pattern recognition, and, not to forget, the new hero on the posters, agentic AI, are beginning to move from experimentation to production.
It is also entirely possible that video generation finds its place elsewhere in the video production chain, assisting in post-production, accelerating editing workflows, or enabling rapid proof of concept and reference shots rather than final output.
Whether fully AI-generated video becomes economically viable remains to be seen. That question is still open. But it does not invalidate the broader trajectory of the technology.
What it means for CIOs
It is precisely when there are confusing signals from news analysis, rapid news flows on the next big thing juxtaposed with a product line shutting down, that CIOs have to don their expert counsel hat. Given how novel the technology is, businesses look to their trusted IT experts for advice on how they can benefit, what to be optimistic about, and what to be cautious about.
For CIOs, this means researching and testing AI offerings within the context of their business, identifying where ROI is already visible (e.g., coding, support, internal automation), and exploring adjacent opportunities where value may emerge. This also comes with consciously placing certain use cases into “watch” or “hold” categories.
And most of all, the role of the CIO is to allocate attention and investment with precision, lest undue skepticism or fear come in the way of realizing benefits in areas where the technology can be deployed.
