Skip to content
Curiosity AI
Reserve capacity

MumbaiSan Francisco

U.S. demand.
Delivered from India.

The full-stack AI compute platform: dedicated NVIDIA Blackwell clusters for U.S. neoclouds, AI labs and inference companies, under one contract and one SLA.

Reserve capacity How it works
  • Q1 2027First 5 MW of NVIDIA B300 capacity online
  • 20 MW+AI factory capacity in 2027
  • 6AI factory partners across India
  • 10,000GPUs deploying in India, Feb – Jun 2027

Phase 1: our first 5 MW of NVIDIA Blackwell B300 capacity comes online in Q1 2027.See the roadmap

Value proposition

Dedicated compute,
delivered with certainty.

U.S. AI companies are funded and ready, but long lead times, scarce power and split accountability keep dedicated capacity out of reach. We deliver it from India, under one accountable contract.

  • 12–18

    Months to AI-ready capacity

    Partner AI factories in Chennai and Mumbai, already powered and in development.

  • 50–60%

    Lower build cost

    About $4–5M per MW, with renewable power at $0.04–0.06 per kWh.

  • 1 SLA

    One accountable owner

    One contract and one escalation owner. Workloads are accepted before billing starts.

  • 275 MW

    Capacity by 2032

    20 MW+ in 2027 and 50 MW+ in 2028, built against contracted offtake.

Management estimates. Figures vary by site and configuration.

The opportunity

Demand is outrunning supply.

Global neoclouds have signed more than $200B of AI compute contracts. Deployed capacity is not keeping up.

  • CoreWeave
    $104B
  • Nscale
    $51B
  • Nebius
    $40B

Contracted backlog, Q2 2026 company disclosures.

Where deployment stalls

  • Capacity is scarce

    Everyone is chasing the same GPUs, with 12–20 week lead times on new systems.

  • Power is the bottleneck

    GPUs need powered, cooled, ready AI factories before they can earn.

  • Buyers want certainty

    Dedicated capacity on long-term, managed terms, with committed delivery dates.

  • India changes the equation

    Lower build cost, renewable open-access power, AI-ready sites and a deep engineering bench.

How it works

One accountable service
across every layer.

U.S. customers sign one contract with one SLA. Curiosity originates the demand, integrates partner capacity and owns the service end to end.

  1. Qualify before committing

    We validate capacity, site readiness and partner capability before a cluster is promised.

  2. Accept before billing

    Workloads, network, storage and recovery are benchmarked before the meter starts.

  3. Own the incident

    Shared telemetry, named responsibility and a single escalation path.

  4. Provide a credible remedy

    Supplier obligations aligned with tested recovery options.

  • Layer 4

    AI services

    Model serving and optimised model loading.

    Curiosity Tier 3, from 2028

  • Layer 3

    Clusters

    Managed Slurm and Kubernetes, AI object storage.

    Curiosity Tier 2, pilot H2 2027

  • Layer 2

    Hardware

    NVIDIA B300 and GB300 GPUs, networking and storage.

    Tier 1 clusters from Q1 2027

  • Layer 1

    Physical

    Power, cooling and space in Indian AI factories.

    AI factory partners

Each tier advances when the tier below meets its uptime and repair-time targets.

C7Y

The seven Cs,
under one contract.

Most providers own one or two layers and rent the rest. Curiosity AI brings all seven together, so buyers deal with one accountable owner.

  • Compute

    3–5 year dedicated offtake

  • Capital

    Project capital raised against offtake

  • Chips

    Latest NVIDIA Blackwell, Vera Rubin next

  • Clusters

    576-GPU SuperPODs

  • Customers

    U.S. labs, neoclouds and inference

  • Cloud

    One view of every GPU

  • Capacity

    Partner AI factories in India

Deployment roadmap

Blackwell now.
Vera Rubin next.

Capacity comes online in phases at named partner AI factories. Each phase proceeds against contracted offtake and financing.

Management plan. Timing depends on contracted offtake and financing.

  1. Q1 2027

    5 MW

    Phase 1: first 2,304 NVIDIA B300 GPUs live in Chennai and Mumbai.

  2. Feb – Jun 2027

    10,000

    NVIDIA B300 GPUs deploying across partner sites.

  3. 2027

    20 MW+

    20,000 NVIDIA B300 GPUs across Chennai, Noida and partner sites.

  4. 2028

    50 MW+

    Liquid-cooled Blackwell Ultra, then NVIDIA Vera Rubin.

  5. 2032

    275 MW

    A national network of AI factory capacity.

AI factory partners

Supply and AI factories, already lined up.

Six AI factory partners with the power, land and institutional capability to scale, starting in Chennai and Mumbai.

AI factory partner cities
Already powered and in developmentPhase 1Q1 202720,000 B300 GPUs2027
ChennaiYesYesYes
MumbaiYesYes
NoidaYes

  • Chennai

    • Already powered and in development
    • Phase 1, Q1 2027
    • 20,000 B300 GPUs, 2027

  • Mumbai

    • Already powered and in development
    • Phase 1, Q1 2027

  • Noida

    • 20,000 B300 GPUs, 2027

Group structure

One group. Two jurisdictions.

Curiosity AI secures capacity and runs clusters in India. Curiosity Cloud Inc. contracts and supports U.S. customers. Customers sign once, with one accountable owner.

India · Platform

Mumbai

Curiosity AI

Curiosity.AI Private Limited

Secures GPUs, AI factory capacity and deployment partners, and runs NVIDIA B300 clusters to SLA across India.

curiosityai.co

U.S. · Customers

San Francisco

Curiosity Cloud Inc.

Delaware corporation, wholly owned

GPU-as-a-Service and compute-as-a-service, contracted and supported locally for U.S. AI companies.

curiositycloud.ai

Now in San Francisco

Curiosity Cloud Inc. is now in San Francisco.

A U.S. home for U.S. customers: a local counterparty for dedicated compute, backed by capacity in India.

  • A U.S. counterparty

    MSAs, SLAs and long-term offtake with a Delaware corporation, in U.S. dollars.

  • On the ground

    Bryant Street, SoMa, with time in New York, alongside neoclouds, labs and inference companies.

  • A familiar structure

    A Delaware entity that U.S. customers, lenders and technology partners know how to work with.

SAN FRANCISCO BAYBAY BRIDGEFINANCIAL DISTRICTSOMAMISSION BAYVAN NESS AVEMARKET STBRYANT STCuriosity Cloud Inc.Bryant Street, SoMa, San Francisco, CAGPU-AS-A-SERVICE · COMPUTE-AS-A-SERVICESAN FRANCISCO, CA
Bryant Street, SoMa, San Francisco

Team

The team.

Energy infrastructure, technology deployment and GPU software expertise.

Punit K Goyal

Founder & Chief Executive Officer

Compute, capital & customers

Leads capital strategy, compute partnerships and global go-to-market across Mumbai and San Francisco.

Aston, LSE, Harvard Business School

Raman Ladda

Co-Founder & Chief Business Officer

Infrastructure & delivery

Leads deployment operations and contracted capacity delivery across partner AI factories.

XLRI, BITS Pilani

Subhajit Ghosh

Co-Founder & Chief Technology Officer

GPU software & reliability

Leads GPU observability, reliability and inference infrastructure for AI training and serving.

NIT Silchar

Reserve capacity

Dedicated NVIDIA Blackwell compute, from Q1 2027.

Tell us your GPU count, configuration and start date. We’ll come back with sites, timing and terms.

We’ll use these details to reply to your enquiry. Privacy policy

Curiosity.AI Private Limited

The Collective, Altimus, Worli, Mumbai, India

punit@curiosityai.co

Curiosity Cloud Inc.

Bryant Street, SoMa, San Francisco, CA

punit@curiositycloud.ai