Adelaide-based Nebula Data is expanding its GPU-computing business across Australia, the US and Europe, while preparing a pay-by-usage service for smaller companies. Its next move reflects a practical constraint in the AI boom: finding power and suitable data-centre space can be harder than finding customers for the chips.
Watch Desk analysis
What happened
Nebula Data began as an IT services business and now supplies dedicated GPU computing for companies running AI models, with a focus on inference: the work models do when answering queries or processing documents. Founder and chief executive Xiamuxirifu Daoran told Business News Australia that the company has expanded its sourcing network overseas after finding Australian data-centre space and power infrastructure difficult to secure. Customers can use capacity in a different country from their own, depending on performance, cost and data-compliance needs.
The company’s existing NEB Compute service reserves bare-metal GPU capacity for customers under fixed-term agreements, typically lasting three or five years. Daoran says Nebula is also preparing shared GPU clusters for smaller businesses, billed by tokens used rather than through a dedicated lease. He expects a cluster to go live in mid-January 2027, with a South Australian site and a Luxembourg site also targeted for early 2027. The company says the new service will be a major focus from the second quarter.
Why it matters
This is a concrete example of AI infrastructure moving beyond the handful of familiar cloud giants. Dedicated capacity may suit established businesses that want predictable access, while usage-based billing could lower the upfront commitment for smaller firms. The trade-off is that Nebula’s timetable and plans are company projections, not evidence that every proposed site or service is already operating.
The account also puts a useful wrinkle in the data-centre boom: more demand does not automatically mean more usable capacity. Daoran says GPU orders can take up to 36 weeks, and argues that power approvals and suitable facilities are major constraints in Australia. The business is responding by sourcing compute across borders, rather than waiting for every customer and server to share a postcode.
Our read
The interesting development is not simply another GPU farm. It is the attempt to sell AI computing in two different ways: reserved capacity for customers that need guarantees, and token-metered clusters for those that would rather not sign a multi-year lease before they know what their workload needs. Businesses should start with the job they want AI to do, not the shiniest GPU on the brochure, as Daoran advises. Sensible advice, and refreshingly less expensive than buying first and asking questions later.
What to watch
- Whether Nebula’s planned mid-January cluster and early-2027 sites go live on schedule.
- Whether smaller businesses adopt token-based billing for real workloads.
- How the company explains its energy use as its international GPU capacity grows.
Discussion spark: For a smaller business testing AI, is pay-per-token access a better route than reserving dedicated GPU capacity, or does predictable access matter more than flexibility?
Sources and evidence
- How Nebula Data went global in hunt for AI compute (12 November 2026, 00:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.