Firmus, Nvidia partner to build 170,000-GPU AI facility in Indonesia
1 min readAnalysis by AlgoThesis Editorial Desk
The story
Firmus, an Indonesian data center operator, is partnering with Nvidia to construct a 170,000-GPU AI facility in Indonesia — one of the larger announced GPU deployments in Southeast Asia to date. The scale of the buildout signals aggressive regional AI infrastructure investment and cements Nvidia's position as the default supplier for large-scale AI compute globally.
For Nvidia, this is another data point in a string of sovereign and enterprise AI deals across Asia-Pacific. With FY2026 revenue at $215.9B (+65.5% YoY), gross margins at 71.1%, and net margins at 55.6%, Nvidia's financials already reflect dominance in GPU-based AI compute. EPS of $4.90 diluted shows the profitability engine is firing.
The incremental read-through here is positive but modest at the company level — a single facility, even at 170,000 GPUs, moves the needle less when you're already doing $215B in revenue. The more meaningful signal is demand-side: Southeast Asian governments and operators are committing capital to large GPU clusters, supporting Nvidia's forward backlog narrative.
The bull tension centers on whether this deal, and others like it, keeps Nvidia's data center revenue compounding above consensus estimates into FY2027. The bear tension is that at these revenue and margin levels, valuation is already discounting continued hyper-growth, leaving little room for supply chain disruptions, export controls, or demand normalization.
Watchers should monitor whether Nvidia's H100/H200/Blackwell mix in deals like this is disclosed, and whether Indonesian regulatory or export-control dynamics (U.S. AI chip export rules) create any friction in delivery timelines.
The case — both sides
With 65.5% YoY revenue growth and a 71.1% gross margin, NVDA's data center segment keeps attracting sovereign and enterprise commitments — the Firmus 170,000-GPU deal reinforces that the global backlog remains deep and that pricing power is intact across geographies.
At $215.9B in annualized revenue and the current premium multiple, NVDA is already priced for near-perfect execution, meaning a deal of this size is unlikely to meaningfully beat consensus estimates — and export control risk to Southeast Asia (per existing U.S. chip rules) adds real delivery uncertainty that the market may eventually reprice.
The house read
Leans bullNVDA's Firmus deal adds another large-scale international GPU deployment to its pipeline — the question is whether incremental sovereign/enterprise deals still move a $215B-revenue stock that already prices in dominance.
Wrong ifU.S. AI chip export controls to Indonesia or Southeast Asia could delay or block delivery of H-series/Blackwell GPUs, directly undermining the deal; additionally, any macro demand slowdown or hyperscaler capex pullback would compress NVDA's premium multiple quickly given how much growth is already priced in.
Published read · research, not advice