Firmus Technologies, the Australian AI infrastructure company valued at $5.5 billion, is putting a very specific bet on Indonesia: it will build its first data centre there through an eight-year partnership with Nvidia, targeting $30 billion in offtake deals. The anchor project is a 360-megawatt Nvidia DSX AI Factory campus in Batam, an island just off the coast of Singapore.

The most important part is not just the megawatts. It is the structure: Firmus is developing this first campus in Batam with Singapore-based DayOne, and it is positioned as the first big test of whether committed AI demand can be pulled forward through long-term off-take arrangements. The campus is set to go live in the next phase of the rollout (the source notes it is “set to go live in […]”, with details cut off), but the direction is clear: Firmus wants to translate infrastructure capacity into contracted revenue at a scale that most new entrants would normally struggle to underwrite.

To understand why decision-makers should care, it helps to frame what an “AI data factory” actually means in practice. Traditional data centers sell compute availability. AI-focused campuses, especially ones described as “Nvidia DSX AI Factory,” are typically built for dense GPU workloads and the whole ecosystem around them: power delivery, cooling, networking, and operational workflows that keep training and inference running reliably. In other words, you are not just buying space. You are buying a production line for AI workloads.

Batam matters because it is geographically adjacent to a major AI market, Singapore, while also giving developers a way to scale physical infrastructure in a different jurisdiction. The island’s proximity to Singapore makes it a strategic location for cross-border demand, partnerships, and talent movement. For Firmus, that geography is a lever. For its partners, it is a way to reduce friction when customers want capacity without waiting years for completely new ecosystems to form.

The eight-year partnership with Nvidia is also a signal. Nvidia has spent the past several years cementing itself not only as a chip supplier, but as an architect of AI systems. When that influence extends into how and where customers deploy, it shifts the competitive landscape. It is one thing for AI infrastructure firms to say they will build. It is another to align the campus model to Nvidia’s DSX branding and be backed by an eight-year relationship. For boards and CFOs, that alignment can reduce technology risk in the same way that long-term supply agreements reduce procurement volatility.

Then there is the $30 billion expectation. Offtake deals are essentially commitments from customers to buy services or capacity, typically with some combination of pricing, volume, or delivery terms. For Firmus, expecting $30 billion in offtake deals is not just a headline number. It is meant to reassure investors, lenders, and potential customers that demand exists to match the build, especially when large AI data centers require heavy upfront capital and long time horizons.

This is where the second-order implications show up for executives watching the space. If Firmus can move from “deal expectations” to signed commitments while still delivering a 360-megawatt campus in Batam, it strengthens a playbook for others: use anchor partnerships and standardized AI infrastructure models to de-risk the business case. If it cannot, the risk is also obvious: expectations without execution can create financing strain, timeline pressure, and pricing pressure when capacity comes online.

Finally, the partnership with DayOne, a Singapore-based company, points to how cross-border consortia are likely to keep winning. Large AI builds are rarely solitary projects. They require local development muscle, permitting navigation, and execution partners who can operationalize construction and launch timelines. For executives in similar roles, the lesson is not “copy this exact deal.” It is that the winners are converging around a template: long-duration anchor relationships, clear workload focus, and demand signals strong enough to justify power-hungry capacity.

Firmus’s move is a reminder that AI infrastructure is turning from a speculative theme into a measurable industrial build, with regulators, grid constraints, customer contracting, and hardware ecosystems all pulling at the same time. The strategic stakes are straightforward: the companies that can secure demand and deliver capacity on schedule will shape where AI compute gets built next.