OpenAI Confronts Compute Ceiling, Reallocates Resources Amidst Expansion Drive OpenAI stops its Sor
openai sora
| Source: Mastodon | Original article
OpenAI has halted development of its Sora video‑generation app, citing a shortage of compute capacity needed to keep its core AI services on track. The decision, announced in a brief internal memo that leaked to the press, redirects GPU clusters earmarked for Sora to the training and inference pipelines behind ChatGPT‑4o, the company’s flagship conversational model, and the upcoming multimodal suite slated for release later this year.
The move underscores a growing tension between OpenAI’s ambition to launch consumer‑facing products and the massive hardware demands of next‑generation large language models. Earlier this month the firm told investors it aims to spend roughly $600 billion on compute by 2030, a figure that forces it to prioritize projects with the highest revenue potential. By pausing Sora, OpenAI can preserve the bandwidth required to meet its aggressive rollout schedule while avoiding a costly overextension of its infrastructure.
OpenAI’s compute strategy is already being reshaped by a series of multi‑cloud deals. A multi‑year, $38 billion partnership with Amazon Web Services will supply the bulk of the raw GPU power for future model training, while a joint venture with Oracle promises 4.5 GW of dedicated AI data‑centre capacity. These agreements give the company flexibility to shift workloads between providers, but they also highlight the sheer scale of resources required to stay ahead in the AI arms race.
What to watch next: analysts will be looking for signals on whether OpenAI will revive Sora once its primary models are stable, or if the pause signals a longer‑term shift toward a tighter, revenue‑driven product pipeline. The next quarterly earnings call should reveal how the compute reallocation is affecting margins, and whether the AWS‑Oracle infrastructure rollout is on schedule to support the company’s $600 billion compute target.
Sources
Back to AIPULSEN