Yotta targets 80,000 Nvidia GPUs across two India campuses in a $12 billion build
The AMW Read
Independent Indian GPU operator announces hyperscaler-scale 80k-chip/$12B build plus $1.5B IPO, updating the infra player map and compute/capital baselines.
Yotta targets 80,000 Nvidia GPUs across two India campuses in a $12 billion build
Yotta Data Services, the Hiranandani Group-backed operator that already controls an estimated 60–70% of India’s installed GPU capacity, said at GFF Mumbai on September 11, 2026 that it will deploy 40,000 Nvidia Vera Rubin chips at its 120 MW D4 facility under construction in Greater Noida and 40,000 GB300 (Blackwell Ultra) chips at the 80 MW NM2 campus in Navi Mumbai. The $12 billion commitment arrives weeks after Yotta filed plans to raise up to $1.5 billion in a public offering, extending a July 2026 raise of $150M at a $4.5B valuation that the AI Market Watch index records as its tracked funding total (index covers ~5,000 companies, not a census).
For AI infrastructure markets, this is the first hyperscaler-scale GPU cluster claim from an Indian operator that is neither a US cloud subsidiary nor sovereign-wealth backed—pushing an independent neo-cloud into the same capacity conversation as dedicated AI factories while New Delhi treats advanced GPU stock as a strategic asset. The disclosed 200 MW combined envelope sits at the lower bound of what ~1,111 NVL72 racks would draw at 190–230 kW each (211–255 MW before cooling), so full 80,000-GPU density implies phased commissioning rather than a single cutover; D4 go-live is framed for May–August 2027.
Builders and investors should read the IPO as a credibility check on Nvidia allocation, liquid-cooling readiness, and contracted utilization—not headline chip counts. Who actually rents the Vera Rubin and GB300 capacity, and at what utilization, will decide whether this becomes sovereign training/inference supply or leased overflow for global clouds.
