Ever since generative AI (GAI) and large language models (LLMs) gained rapid popularity, with early sites like ChatGPT, Craiyon, and Dall-E taking the internet by storm, there has been a massive investment in GAI and LLMs. We’ve seen these get rolled out on a very wide scale, with programs like Copilot, Gemini, and Grok being integrated into products that people already use on a daily basis, while others like Claude or Sora were designed for specific uses. But they also have popped up in every corner of the internet, often in the form of navigation chatbots that come up when you open a website. The Illinois Institute of Technology has not been immune to this burst; in addition to Illinois Tech having offered a degree in AI for years now, advertisements for a prospective Claude club started appearing across campus about a month ago, and the College of Architecture held a meeting about the place of GAI in architecture on March 25, to name a few.
However, as this massive investment started to pile into the GAI and LLM development space, many people started to question the long-term viability of these investments. For instance, in 2024, researchers projected that OpenAI was projected to lose far more than it made, saying “[d]educt the potential costs of $8.5 billion from revenue of up to $4.5 billion, and you get losses of $4 billion to $5 billion” (“Why OpenAI Could Lose $5 Billion This Year”, Amir Efrati and Aaron Holmes, The Information, July 24, 2024). Across the same period, Perplexity had an operating cost of around 164% of its revenue (“For Google Challenger Perplexity, Growth Comes at a High Cost”, Sri Muppidi, The Information, May 19, 2025). While there has been some stabilization due to fundraising to make up for operating losses, this has raised concerns that GAI and LLM may be a bubble.
An economic bubble refers to an industry where more money is being invested into its development and rapid growth than it could realistically make. Bubbles pop when people realize the money isn’t being made back, and investments stop. Usually, bubbles are in new industries; recent bubbles include the dot-com bubble that came around the growth of early websites, and the housing market crash around the rapid development of the subprime mortgage lending. It can sometimes be hard to tell what will and won’t be a bubble in advance, because you need to invest money to develop any industry; most will start as an operational loss. Doing so doesn’t mean that an industry is a bubble. This is further complicated by the fact that there are some industries that could eventually become profitable, but it would take so long that investments stop early. These can still bubble, despite the fact that they did have the potential to be successful.
The profitable use case for GAI and LLMs would be increasing productivity enough that companies can afford to lay off employees, paying some money for a subscription to the service, but saving it in the long run by having to pay fewer employees. However, it’s becoming increasingly clear that AI does not actually optimize workflow as much as people imagine. For instance, some studies have found that only 14% of workers found it actually made them more efficient, and estimated that rework due to simple AI errors could add about 1.5 weeks of busy work annually. There can be discrepancies across departments, with executives finding it to be the most effective and HR finding it the least useful, but they raised serious concerns about the long-term viability of AI for efficiency (“These Studies Warn That AI Tools Aren’t As Efficient as You Think”, Kit Eaton, Inc., January 28, 2026).
The other potential profitable use case would be if GAI and LLM images, videos, and text could be licensed and sold similar to stock photos. This definitely wasn’t the primary case, but it was a potential secondary revenue source. However, the Supreme Court recently turned down an appeal in Thaler vs. Perlmutter, upholding a lower court decision that AI creations cannot be protected under copyright law, which requires human authorship. That does not mean people will not try to sell you these images, but legally, they are not the intellectual property of anyone. This ruling dealt a blow to this use case for AI.
As the efficiency issues and copyright issues have become more and more prominent, the concerns that GAI and LLMs may be an industry bubble have been growing. It’s possible this industry could have been profitable if the efficiency was worth paying for, but it’s becoming increasingly doubtful that there is actually a benefit to them. Coupled with how much money these companies are losing on AI, the viability of the industry is becoming more and more called into question. These have been around since basically the industry started, but have become much more prominent since it was announced SoraAI would be shutting down, due to being “a resource black hole” (“OpenAI closes Sora video-making app and cancels $1 Billion Disney deal”, Osmond Chia and Emma Calder, BBC, March 25, 2026).
SoraAI is a short-form video generation program from OpenAI, which also makes services like ChatGPT. According to OpenAI, it will be discontinued on April 26, and they are recommending that people download any previously made videos before it does so. They have said they will try to make downloads available for some time after the shutdown, but as of the writing of this, they haven’t committed to this. SoraAI is not the first AI program to shut down, but it is the largest, and with OpenAI being the company that arguably started the GAI and LLM boom when they released ChatGPT, it is seen as somewhat more significant than other companies doing the same. This shutdown was dealt another blow after Disney cancelled a licensing deal with SoraAI. Disney was the first major studio to license their characters to be created through GAI images and videos, but the loss of SoraAI has ended this deal (Chia and Calder).
Both the shutdown and the cancelled deal seemed to back up that this is not profitable and there is no realistic use case worth paying for. This would support the idea that the GAI and LLM industry is a bubble. Which leads back to the essential question: Does the SoraAI shutdown indicate the start of the AI bubble burst? And the honest answer is that it is probably too soon to tell. It’s a definite blow to the industry and could cause a broader turn, but this could also just be one failed project. However, I would say that it’s probably good to keep an eye on things for the future. This definitely could be the start of popping bubbles.