With the Sora 2 video generator, OpenAI seems to have ignited a financial bonfire with its creation tool. Projections revealed by Forbes journalist Phoebe Liu indicate that the company could be spending up to $15 million every day to power this machine that produces animated clips, often described as whimsical or useless.
On an annual scale, this would represent more than $5 billion, a colossal sum that raises questions about the economic viability of this technology. The notion of an AI bubble is being mentioned more and more, and even Sam Altman is warning about excessive spending, often without a return on investment. At Yiaho, we are also fully aware of the challenges related to costs and free AI, so let’s examine this together.
Sora 2: Unsustainable spending
The head of Sora at OpenAI, Bill Peebles, admitted it himself in late October: current costs are simply unsustainable.
These estimates are based on a series of assumptions, as the company jealously guards its precise data on tool usage. They include shifting variables such as the price of graphics processors (GPUs), calculation efficiency, user volume, and video generation frequency.
To understand the scale of the phenomenon, we need to dive into the technical inner workings.
Generating videos with AI is expensive
Unlike text models, which are simpler to manage, video generation systems like Sora process data in four dimensions to ensure smooth consistency across dozens of frames per second.
Cantor Fitzgerald analyst Deepak Mathivanan estimates the internal cost for OpenAI to produce a ten-second video—the standard duration for one credit unit—at around $1.30.
This evaluation is based on about forty minutes of GPU time per clip, spread across several parallel units, with a rental rate of around two dollars per hour. An expert from SemiAnalysis confirms the plausibility of this figure, while noting variations depending on the model versions used.
Currently, OpenAI charges one dollar for a basic video via its programming interface, and three dollars for a more sophisticated version, with no apparent profit margin for now.
The equation becomes explosive when you factor in the volume of activity.
With an estimated 4.5 million users on the Sora app, and assuming that a quarter of them produce an average of ten videos daily (a hypothesis put forward by industry observers), that results in more than 11 million clips per day.
Multiplied by the unit cost, we are nearing $15 million daily.
This projection excludes failures: videos abandoned for non-compliance or drafts that consume resources without being completed.
Read more on this topic: How to create images for free with AI?
Free access to attract users
Offering free access to this feature is a bold strategy, typical of tech giants in a growth phase. At Yiaho, we are convinced that free access is not only great for internet users, but also, and above all, the future of artificial intelligence!
OpenAI is also betting on massive adoption to dominate the market, accepting initial losses in the hope of reducing costs in the long term and monetizing later. OpenAI is not profitable, but the company has been adopted by millions of users with extreme speed. This is a proven tactic on the internet: prioritize audience and engagement before profitability, as Meta or Google have done in the past.
Sam Altman, CEO of OpenAI, has for now ruled out an ad-based model for GPT, which is considered insufficient given the computing power needs. At the same time, data collected from users who do not opt out is used to refine algorithms, satisfying a voracious appetite for annotated video content.
This could provide a lasting competitive advantage. OpenAI is thus betting on a future where the abundance of free videos will have forged a profitable ecosystem, turning a financial sinkhole into a potential gold mine.
Source: Forbes
