← Back to feed

Yotta Data Services is committing 12 billion dollars to a combined order of 80,000 to 90,000 Nvidia GPUs spanning two generations: roughly 40,000 to 50,000 Vera Rubin units and 40,000 to 45,000 GB300 units. Vera Rubin is Nvidia’s first HBM4 platform, and Yotta is among the earliest operators globally to place large-volume orders. CEO Sunil Gupta says the company already controls a majority of India’s installed GPU capacity. The combined order is designed to widen that gap before rivals can close it.

The Hardware Bet

Vera Rubin is Nvidia’s next-generation platform after Blackwell, its first to use HBM4 memory. Yotta’s order of up to 90,000 GPUs represents a direct doubling down on Nvidia-only infrastructure rather than hedging across CUDA, ROCm, and TPU stacks.

That stack choice is deliberate. Running mixed GPU software environments across a single data center adds engineering overhead. Yotta is standardizing on CUDA across two sites — its Navi Mumbai and Greater Noida facilities — mixing Vera Rubin and GB300 hardware generations but keeping the software layer uniform. The company’s prior buildout was 30,000 Blackwell GPUs at roughly $3B. The new combined order is larger at every dimension: more GPUs, more capital, newer silicon.

The IPO Arithmetic

The financing math is uncomfortable. Yotta’s IPO targets a valuation near 6 billion dollars with proceeds up to 900 million dollars from the public listing — a fraction of the 12 billion dollar capital commitment. Total fiscal-year fundraising including pre-IPO private rounds is reportedly targeting around 1.5 billion dollars, still well short of the full commitment.

The gap is expected to be filled through a mix of debt financing, vendor financing arrangements with Nvidia and its partners, and potential additional equity rounds ahead of any public listing. That is a standard playbook for AI infrastructure operators — the asset itself (GPUs with contract-backed revenue) secures the debt. Still, the exposure ratio is notable.

India’s GPU Lead

Yotta operates data center campuses in Navi Mumbai and Greater Noida, with long-term scalable capacity targets of 2GW and 250MW respectively across those sites. By Gupta’s account, its installed GPU capacity already exceeds rivals’ combined base in India. The 80,000 Rubin order, if fully delivered on schedule, would make any catch-up effort by competitors require buying into a generation that is already being superseded.

India has framed AI compute as a sovereignty issue, but the actual bottleneck has always been physical accelerators, not policy. Yotta is occupying that position early — not as a model builder but as the rack. The question the IPO will answer is whether financial markets price that role the same way the GPU markets do.

Key Numbers

• Order size: ~80,000-90,000 GPUs (40,000-50,000 Vera Rubin + 40,000-45,000 GB300)
• Capital commitment: $12B
• Platform: Vera Rubin (Nvidia’s first HBM4 GPU) + GB300
• IPO: targeting $6B valuation, up to $900M in public listing proceeds
• Prior deployment: 30,000 Blackwell GPUs, ~$3B