Jensen Huang has said that OpenAI’s GPT-6 Astra represents the company’s arrival at artificial general intelligence, while highlighting the enormous computing infrastructure reportedly used to train it. In the post, he says Astra was trained on more than 100,000 NVIDIA Grace Blackwell NVLink72 systems and that another 400,000 GPUs are coming online.

Original post on X

Load the post to view it as published on X. X may receive connection data.

View the original post on X ↗

The infrastructure claim

The post offers only a brief description, but its central technical detail is the reported training scale. GPT-6 Astra is presented as the latest step in a progression from ChatGPT to o1 and then Astra over four years. That framing connects the model’s development to a rapid expansion in both model capability and computing capacity.

The reference to more than 100,000 Grace Blackwell NVLink72 systems is a claim about the infrastructure behind Astra, not a detailed account of how the system was trained. The post does not provide information about the model’s architecture, capabilities, evaluation results or the specific role of the additional GPUs.

What the AGI statement establishes

“AGI has arrived” is Huang’s conclusion, not independently demonstrated evidence in the post. Artificial general intelligence is not defined in the announcement, and no benchmark or capability description is provided to show how Astra meets that threshold. Readers should therefore understand the statement as a major declaration about GPT-6 Astra, rather than as a settled technical assessment.

The immediate practical significance of the announcement is its scale: if the reported figures are accurate, Astra is being associated with a substantial expansion of AI computing infrastructure. The post says 400,000 more GPUs are coming online, but does not specify when they will be available, where they will be deployed or what systems they will support.