NVIDIA Partners with Firmus on Indonesian Data Center: 170,000 GPUs, $30 Billion Expected Revenue Over 6 Years
0xBroomberg
Nvidia has struck a structural partnership with Asia-Pacific cloud startup Firmus to deploy 170,000 advanced GPUs in Indonesia's Batam Island — for the first time taking a revenue share tied to chip utilization at a third-party data center. This means Nvidia is shifting from selling shovels to taking a cut of every ounce of gold the shovels dig up.
How big is this data center?
Firmus will build a large-scale data center on Batam Island, Indonesia, deploying at least 170,000 Nvidia advanced chips spanning the Grace Blackwell and Vera Rubin GPU and CPU lines.
Compute capacity is scheduled to come online starting Q1 2027.
This means → a single project with 170,000 GPUs rivals the single-region deployments of some hyperscalers. Firmus is a newcomer playing with giant-tier hardware from day one.
Why is Nvidia willing to share revenue instead of just selling chips?
The structural breakthrough: Nvidia is not only selling hardware but also receiving a share of cloud-service revenue based on chip utilization, while providing credit support for the project.
Per The Information, this "hardware sale + revenue share + credit support" triple structure is a first in Nvidia's partnerships with third-party data centers.
In plain terms = Nvidia used to ship GPUs and move on. Now it clocks in alongside every GPU — the more each chip runs, the more Nvidia earns.
This reflects Nvidia's push to convert one-time hardware sales into a recurring revenue stream, locking in long-term returns.
What backs the $30 billion revenue projection?
Per Reuters, Firmus expects the deal to generate up to $30 billion in revenue over the first six years, based on existing customer commitments.
The target customers are not hyperscalers like Microsoft, Amazon, or Google, but "AI-native" companies — startups built from scratch with AI at their core.
Potential client types include life-sciences institutions, robotics firms, AI software companies (e.g. Lovable, Cognition), and AI model developers (e.g. OpenAI, Anthropic).
This means → Firmus is betting that the next wave of AI compute demand comes from companies "born in the AI era," not from giants that already run their own data centers.
Can Firmus actually handle a project this large?
Founded in 2019 and headquartered across Australia and Singapore, Firmus completed a funding round in April with Coatue, Nvidia, and others, reaching a $5.5 billion valuation.
Total equity raised over the past six months: $1.35 billion. In February, the company also closed a $10 billion debt facility led by Blackstone.
Nvidia already holds equity in Firmus. The two have worked together since 2021, initially co-designing Firmus's proprietary cooling technology, HyperCube.
In plain terms = Nvidia is Firmus's supplier, its shareholder, and its credit backer — deeply aligned interests, not a one-off transaction.
Can this model be replicated?
Co-founder Tim Rosenfield said Nvidia actively backed the project partly because Firmus adopted Nvidia's full hardware stack and specification standards.
Per Reuters, citing people familiar with the matter, Firmus has engaged investment banks to explore a potential IPO, though Rosenfield declined to comment.
This reflects a bigger question: if the "hardware + revenue share + credit" model proves out, Nvidia can replicate it across Asia-Pacific — every emerging AI market could spawn its own "Firmus," with Nvidia taking a cut from each one.
Content is for reference only, not financial advice.