The videos created with Sora, Open AI’s text-to-video generator, are as spectacular as the secrecy surrounding the service is disconcerting. It takes such an exorbitant amount of computing power to generate these images that it even raises suspicions about whether Sora may exist. At least in a way similar to ChatGPT. And it is also not clear that there are enough chips to process all that data. At least from Nvidia, the main company that sells them and that cannot supply them because the waiting list is large.
The success of the Lumiere brothers’ cinematography had a lot to do with how economical it was from the beginning. So much so that it even became a fair show. But Nvidia’s H100 GPUs, which are at the heart of most artificial intelligence, are neither cheap nor plentiful.
It is very different to generate low-resolution photos or illustrations using artificial intelligence, like those of Dall-E, from creating videos with Full HD resolution of up to 60 seconds. That is the technical capability of Sora announced by Open AI. Many server farms are needed to generate this immense stream of images.
The company has not said a word about when Sora will be available or at what price. Launching it on a global level surely requires an astronomical investment. Every day we see users who have access to the trial version of Sora publish incredible videos on social networks, but only OpenAI knows the real cost of generating them.
Everything indicates that Sora is similar to nuclear fusion energy: something fascinating that we can create in a laboratory, but that cannot be used on a practical level. Sora could be released at any time, but first you have to know if it makes sense to generate videos with artificial intelligence. Could it be even more expensive to create footage with Sora than to film it with a camera or 3D mold it manually?
It is also unclear whether Sora is viable on an energy and environmental level. It’s scary to think about the amount of water its massive deployment would require. The only thing that Open AI seems to show us with this tool are the physical limits of artificial intelligence. Yes, physical.
Shortly after the videos created with Sora were released, it was leaked in the Wall Street Journal that Sam Altman is seeking authorization from the United States Government to create a huge company dedicated to the manufacture of chips. And equally important: he is seeking multimillion-dollar financing to achieve it.
It is still paradoxical that the main economic benefits of artificial intelligence are not obtained by software developers like Open AI, but rather by those who produce the tangible part of the technology. The big business of AI is in the manufacture of the hardware that allows ChatGPT or Gemini to answer our questions.
It seems that Sam Altman wants to enter that market and even lead it. But creating chip factories is slower and more expensive than manufacturing algorithms. Nvidia, the company that sells much of the hardware that powers artificial intelligence, has seen its stock price skyrocket. It is on its way to becoming the largest technology company in the world.
What happened recently on the Wall Street stock market is a milestone in the history of artificial intelligence. Nvidia’s stock price skyrocketed on February 22. This happened after several days of doubts. Some analysts thought that the artificial intelligence market was overvalued and a bubble could be creating. But the financial results report released by Nvidia after closing Wall Street on Wednesday was surprising.
Its sales were expected to be significant. But if those numbers weren’t as spectacular as Nvidia’s stock growth, that could reflect that artificial intelligence was overrated. What happened on the 21st when the New York Stock Exchange closed is that Nvidia’s sales figures exceeded all expectations. And that sparked a rally in stocks the next day.
Nvidia’s incredible sales growth shows that investing in artificial intelligence is serious. Its hardware sales, among which the H100 GPU stands out, have made the CEO of Open AI react. But Sam Altman’s plans to make chips hit major obstacles.
The main one is the enormous energy cost and ecological impact of multiplying the computing power of artificial intelligence. It is estimated that cooling ChatGPT servers to maintain a simple conversation requires half a liter of water.
Either OpenAI has an amazing technology that we don’t know about, and it is not ruled out that it has it as it demonstrated with ChatGPT, or the cost of generating videos like Sora’s is very difficult to assume. There are also reasonable doubts as to whether it is worth putting so much effort into creating artificial intelligence capable of creating videos. Sora seems like a wonderful tool for those who are dedicated to audiovisual creation or art, but it is not clear that the world needs to fill TikTok with videos created with artificial intelligence.
