Is the Disney-OpenAI Agreement a 'Watershed' Moment for AI?
Adoption and usage of GenAI-created video has big capacity implications.
You're reading 0 of 1 free page.
Register to read more or Unlock Pro — 50% Off Ends Soon
At the risk of being perceived as beating a dead horse on the topic of AI adoption usage today, following our Oracle (ORCL) , Adobe (ADBE) , and Bank of America (BAC) comments, let's discuss the new relationship between Walt Disney (DIS) and OpenAI announced this morning. Some, like The New York Times, are calling it a “watershed” moment for Hollywood and generative AI as it brings more than 200 Disney characters to OpenAI’s short-form video platform, better known as Sora. Starting next year, as part of the three-year deal, Sora users will be able to make content with characters across Disney, Marvel, Pixar, and Star Wars.
We find this very interesting, given its potential to boost consumer use of Sora. After all, content is king, and we’ve seen what folks have already done with messaging tools, emojis, images, gifs, and video. We also know that when it comes to network traffic, video is a much larger consumer compared to text-based messaging, emails, and the like. Per data from Ericsson (ERIC) , smartphone data usage per user grew from around 3.4 gigabytes (GB) per month in 2018 to over 16 GB per month in 2024 for video alone.
When it comes to generative AI, it looks like that is going to have a similar but larger impact on digital infrastructure.
We say this because it appears that an AI-generated video uses roughly 10,000 to 50,000 times more data than a text query and response. That’s based on a simple text query and response using 1-5 kilobytes of data, while a short-form AI-generated video between 15-60 seconds will consume 5-50 megabytes of data depending on the video resolution, ultimate length of the video, compression and frame rate.
Those figures come straight from asking Anthropic’s Claude AI model the following question:
Claude, how much data does a short-form video created by AI use compared to the amount of data used for a simple query with a text answer?
OpenAI’s ChatGPT answered the same question with the difference ranging from ~100× to 1,000× more data for a short video vs. a simple text answer.
Between these two answers, it’s fair to say that AI-generated video is going to consume far more data and drive far more network traffic than what most folks are using these models for today. In other words, subject to adoption and others offering similar short-form AI-generated video content capabilities, it has the potential to drive data center and digital infrastructure capacity utilization levels higher, fostering the need for even more capacity.
How much of that expected increase is already baked into capital spending forecasts from Big Tech companies is unknown. But what we can deduce is that if adoption and utilization rates for generative AI short-form video creation rise meaningfully, we aren’t likely to see those capital spending levels move markedly lower near term.
The need for more capacity means more data centers, servers, chips, and all that connects and powers them.
Now, we'll keep an eye on how folks use this new Disney-OpenAI offering and see the effect on adoption and usage metrics.
And lest we forget, Disney also said it will deploy OpenAI’s ChatGPT to its employees, and help the company build “new products, tools, and experiences, including for Disney+, and deploying ChatGPT for its employees.”
That sure reminds us about how the initial internet went from being nothing more than a basic digital catalog with no purchasing or payment options attached to the digital Swiss Army knife it is today.
Back in (gasp) 1995, total global internet traffic was around 0.18 petabytes (PB) per month. Data from Cisco (CSCO) points to ~9 to 10 PB per minute being consumed, which, multiplied by the 43,380 minutes in the average month, equates to ~ ≈400,000 PB/month. 🤯
At the time of publication, TheStreet Pro Portfolio had no positions in any securities mentioned.
