The headline number is massive: up to 170,000 NVIDIA accelerators. But the real signal is who the capacity is for. Firmus is targeting AI-native companies, enterprises and ISVs that need dependable access to accelerated compute but may not be able to contract or finance infrastructure at hyperscaler scale.

That makes the Batam campus part of a broader AI cloud access shift. The next generation of AI companies may need access to frontier-class compute without owning every rack, negotiating every power contract or building their own global data center footprint.

Firmus says the campus will integrate NVIDIA DSX, NVIDIA’s full-stack AI factory platform, with its proprietary HyperCube liquid-cooled architecture. The point is to design, simulate and operate the facility as one coordinated AI factory rather than a generic data center filled with GPUs.

This distinction matters for AI workloads. Training and inference need power, cooling, networking, storage, scheduling, reliability and tokens-per-watt efficiency to work together. An AI factory is valuable when it lowers the cost per token and increases usable capacity, not simply when it installs more chips.

AI infrastructure is capital intensive. High-growth AI-native companies often need huge compute commitments before their revenue, margins or funding can comfortably support them. Firmus and NVIDIA’s partnership uses a revenue-sharing and credit-support model to make that capacity easier to access.

This could become an important template. If accelerator vendors and infrastructure operators share upside, they can support AI companies that have strong demand but limited upfront capital. In return, infrastructure providers gain usage-linked revenue streams instead of relying only on hardware sales or fixed capacity leases.