Sora AI would need over 750k Nvidia H100 GPUs to rival TikTok & YouTube

Image Source: Open AI

Sora AI blew everyone’s minds with its demo videos and we can’t wait for Open AI to launch it later this year. However, it would be a struggle for Sora AI to work with all the demands from all the excited users.

Right now, even the GPT-6 training is demanding a lot of resources and computing power. There is a huge need for raw GPU power and right now, NVIDIA is running the show in AI. However, Microsoft and Open AI already drew the plans for a $100 billion supercomputer to cope with the huge demand.

But that’s not all, for ChatGPT’s over 10 million daily queries, Microsoft is consuming an outrageous quantity of water just for AI generation.

AI needs a lot more GPUs for text to video generation

I’m telling you all this because the demands for text to video generation, the demands grow exponentially. According to Factorial Funds, Sora AI needs an estimated 4,200-10,500 Nvidia H100 GPUs for 1 month just to train the model.

They calculated that a single NVIDIA H100 GPU can generate approximately five minutes of video per hour. Based on this estimation, and the fact that on TikTok there are uploaded 17 million minutes of video and on YouTube 43 million minutes of video, Sora AI would need to generate between 15.3 million and 38.1 million minutes of video to be on par with the two platforms.

Going further with this estimation, Sora AI would need approximately 750k NVIDIA H100 GPUs to generate the same amount of video footage as TikTok and YouTube.

Of course, Open AI never expressed a desire to rival the above-mentioned platforms, but it’s fun to put things into perspective.

I’ve already reported about the possible implications of Sora AI for the video industry but we’re still a long way to go before that.

What do you think about Sora AI text to video generation? Let us know in the comments section below.