During a recent Reddit AMA, OpenAI CEO Sam Altman acknowledged that the company is facing significant delays in releasing products due to a shortage of computing capacity

He explained that the complexity of the models is increasing, which complicates their deployment. “We encounter many limitations and must make difficult choices about how we allocate our computational resources across numerous promising ideas,” he noted when asked about the timeline for upcoming AI models.

Reports indicate that OpenAI has had difficulty obtaining enough infrastructure to effectively run and train its generative models. According to Reuters, the company has been collaborating with Broadcom to develop an AI chip aimed at facilitating model operations, with an anticipated release as early as 2026.

As a result of these capacity constraints, Altman mentioned that the advanced conversational feature, Advanced Voice Mode, for ChatGPT will not receive the previously announced visual capabilities anytime soon. During an event in April, OpenAI showcased the ChatGPT app on a smartphone, demonstrating its ability to respond to visual inputs, such as clothing colors, detected by the camera. However, it has since come to light that the demonstration was hastily arranged to divert attention from Google’s I/O developer conference occurring at the same time, and many within OpenAI felt that the GPT-4o model was not ready for unveiling. Furthermore, the voice-only version of Advanced Voice Mode faced prolonged delays.

In the AMA, Altman also revealed that there is currently no timeline for the next version of OpenAI’s image generation tool, DALL-E, stating, “We don’t have a release plan yet.” Meanwhile, the video generation tool Sora has been delayed due to the need for refinement, safety measures, and increased computing power, according to Kevin Weil, OpenAI’s chief product officer, who joined Altman in the AMA.

Sora has reportedly encountered technical difficulties that put it at a disadvantage compared to competitors like Luma and Runway. It was reported that the initial version of Sora required over ten minutes of processing to create a one-minute video clip. In a recent development, one of the project’s co-leads, Tim Brooks, left OpenAI for Google.

Later in the AMA, Altman discussed the possibility of allowing “NSFW” content in ChatGPT in the future, expressing the company’s belief in treating adult users with respect. He also emphasized that the top priority for OpenAI is enhancing its series of “reasoning” models, known as o1. At the recent DevDay conference in London, OpenAI previewed several upcoming features for these models, including image comprehension.

“We have some exciting releases planned for later this year,” Altman added, although he clarified that none of them would be branded as GPT-5.