OpenAI executives have confirmed the company is holding discussions with Australian authorities to explore the possibility of training its artificial intelligence models locally. This development signals the US-based technology firm's interest in shifting a portion of its frontier model training operations to the Asia-Pacific region.

A senior OpenAI representative noted that the conversations remain at an exploratory stage, with no final agreement yet reached. The core topics under discussion include computing capacity, power supply, copyright regulations, and other compliance requirements.

In December 2025, OpenAI signed a memorandum of understanding with Australian data centre operator NEXTDC to build a massive AI computing campus on Sydney's western outskirts, intended to host a GPU supercluster. However, that agreement was initially scoped purely for inference workloads and did not include a commitment to undertake pre-training of large models within Australia.

The Australian government has been actively courting AI heavyweights such as OpenAI, Anthropic, and NVIDIA, aiming to attract model training operations to its shores and foster a sovereign AI computing ecosystem. A significant hurdle, however, lies in Australia's existing copyright framework: current law lacks a clear exemption for AI firms using web-scraped text data for model training, leaving companies exposed to potential infringement disputes.

Electricity availability and renewable energy infrastructure are equally pivotal to the negotiations. Large-model training is extraordinarily energy-intensive, and Canberra has pledged to support projects with long-term green power purchase agreements. OpenAI stressed that even if the talks conclude successfully, the on-ground training programme would be rolled out in phases, beginning with fine-tuning of select models rather than immediately attempting the full-scale pre-training of the largest frontier foundational models.