A 360‑megawatt AI Factory campus will be built in Batam, Indonesia, after Firmus Technologies and NVIDIA announced a strategic partnership to create one of Southeast Asia’s largest dedicated AI compute hubs.

The Batam AI Factory will deploy roughly 170,000 NVIDIA AI accelerators across Grace‑Blackwell, Vera and Rubin platforms, and — based on current customer commitments — is expected to generate $25 billion to $30 billion in offtake agreements during its first six years of operation. The campus is targeted to begin commercial operations in the first quarter of 2027 and is being developed with Singapore‑based digital infrastructure platform DayOne.

Firmus plans to integrate NVIDIA’s DSX AI factory platform with its proprietary HyperCube liquid‑cooling architecture. The combined stack aims to speed deployments, improve energy efficiency and reduce customers’ cost per token — a key metric for generative AI workloads that track compute cost against model output.

“AI‑native companies need access to scalable, energy‑ and cost‑efficient compute infrastructure to compete globally,” said Tim Rosenfield, co‑CEO of Firmus Technologies. “This partnership with NVIDIA provides unprecedented access to the most advanced AI accelerators in the world, with the certainty, scale and flexibility that best fits a high‑growth trajectory.”

Why Batam

Batam’s location just across the water from Singapore gives the campus proximity to one of the world’s most important financial and cloud hubs while offering lower land and operating costs. For international AI firms seeking latency advantages without Singapore’s higher real‑estate and power prices, Batam presents a compelling alternative. The island’s existing industrial infrastructure and favorable business arrangements with Singapore make it a logical site for large‑scale digital infrastructure.

Market and technical impact

Concentrating approximately 170,000 accelerators at a single site can meaningfully expand regional GPU capacity at a time when global demand for AI chips remains high. The mix of Grace‑class and Rubin platforms suggests the campus will support both large‑scale model training and inference workloads, catering to AI‑native startups, enterprise customers and independent software vendors.

Firmus’ liquid‑cooling HyperCube design paired with NVIDIA’s DSX factory approach targets lower PUE (power usage effectiveness) and reduced operational cost per compute unit versus air‑cooled designs. Liquid cooling is increasingly standard for high‑density GPU deployments, but it requires specialized infrastructure, supply chains and maintenance expertise.

Various media outlets have reported the Batam AI Factory as a US$5 billion project, but the developer has not directly disclosed the cost.

Risks and unknowns

Delivering thousands of high‑end accelerators depends on continued NVIDIA supply and resilient global logistics; any chip or component delays could push the timeline and revenue projections. Securing 360 MW of stable, long‑term power and the necessary local permits on the proposed schedule is another practical challenge. The headline $25–30 billion offtake figure is based on current commitments and could change if customers scale back or contractual terms evolve.

Regional competition is also intensifying. Hyperscalers and regional data‑center builders are rapidly adding GPU capacity across Asia Pacific, which may compress pricing or extend sales cycles for large committed blocks of capacity.

Additionally, regional hyperscale projects such as AirTrunk’s JHB2 campus in Johor — a 270+MW, liquid‑cooled facility aimed at cloud and AI workloads — show how APAC markets are racing to add large‑scale GPU capacity.

What to watch next

• Delivery schedule for NVIDIA accelerators and dataset‑ready systems.
• Firmus’ commercial model details: minimum committed capacity, flexible burst options, and pricing per token or hour.
• Power sourcing and grid agreements for the 360 MW IT load.
• Interconnection and peering options with Singapore, Jakarta and other APAC nodes.

Factsheet: Batam AI Factory

• Location: Batam, Indonesia
• Capacity: 360 MW (IT load)
• GPUs: ~170,000 NVIDIA AI accelerators (Grace‑Blackwell, Vera, Rubin)
• Expected offtake: $25–30 billion (first six years, per current commitments)
• Partners: Firmus Technologies, NVIDIA, DayOne (Singapore)
• Target online date: Q1 2027

The Batam AI Factory, if delivered on schedule, could shift APAC’s GPU capacity dynamics by offering a high‑density, lower‑cost alternative to Singapore and other regional hubs. Much will depend on supply chains, power arrangements and whether committed customers convert those early agreements into long‑term usage — factors that will determine whether Batam becomes a model for future AI‑scale campuses across the region.