OpenAI CEO Sam Altman says lack of compute capacity is delaying the company’s products

In a recent Reddit AMA, Sam Altman, CEO of OpenAI, addressed the challenges the company faces in releasing new products, primarily highlighting the constraints related to compute capacity. He explained that the increasing complexity of AI models has led to difficult decisions regarding how to allocate computational resources among various promising ideas, which has hindered OpenAI from launching products as frequently as intended.

Altman’s remarks were corroborated by reports suggesting that OpenAI has been grappling with limitations in securing sufficient infrastructure to train and run its generative models. Notably, Reuters reported that OpenAI has been collaborating with Broadcom to develop a dedicated AI chip, expected to be ready by 2026, which could alleviate some GPU availability issues.

As a consequence of these compute constraints, Altman disclosed that features for ChatGPT, particularly the much-anticipated Advanced Voice Mode with visual capabilities, will not be introduced in the near future. At an event in April, OpenAI had showcased a demo of ChatGPT interacting with visual cues through smartphone cameras. However, reports indicated that this demonstration was hastily prepared to overshadow Google’s I/O developer conference, and many at OpenAI felt the feature was not yet sufficiently developed, leading to delays in releasing even the voice-only mode.

During the AMA, Altman mentioned that there is no set timeline for the next iteration of OpenAI’s image generator, DALL-E, and that the video-generating tool, Sora, has also been delayed. Kevin Weil, the company’s Chief Product Officer, elaborated that Sora’s development has been hindered by technical challenges, including significant processing time needed to create video content, which has put it at a disadvantage compared to competitors like Luma and Runway. Recent personnel changes within the team, including the departure of co-lead Tim Brooks to Google, may also have contributed to the setbacks in Sora’s development.

Additionally, Altman touched upon OpenAI’s openness to considering the inclusion of “NSFW” content in ChatGPT in the future, emphasizing the company’s belief in respecting adult users. He also highlighted that improving the reasoning capabilities of their series of models remains a top priority for OpenAI. At a recent DevDay conference in London, the company previewed several upcoming features, including advancements in image comprehension, although Altman specified that nothing currently in development would be branded as GPT-5.

Altman’s candor during the AMA has provided insights into the operational hurdles faced by OpenAI and the company’s commitment to navigating these challenges while maintaining a focus on safety, scalability, and quality in its AI offerings.