India · INR · DPDP-compliant

Rent GPUs.
Train Models.
Ship Faster.

India's GPU cloud built for builders. Spin up H200s, H100s, A100s and L4s with an SSH key in under 15 seconds. Billed in INR, GST invoices included, no surprises.

▸ ₹40/hr from·99.9% uptime·3 IN regions·UPI & cards
H200 SXME5 SUPERCRITICAL
VRAM141 GB
MEM BW4.8 TB/s
▸ Instant access▸ per-minute billing▸ cancel anytime▸ Instant access▸ per-minute billing▸ cancel anytime
🌀
Cooling fan900 RPM
// How it works

From signup to SSH in three keystrokes.

No credit card walls. No GPU quotas. No hidden ingress fees. Just log in, pick a GPU, and start training.

01

Sign up

Create your account with Google, GitHub or email. Verify your phone, drop in an SSH key, you're in.

Time: ~30 seconds
02

Pick a GPU

Choose between H200s, H100s, A100s, RTX 4090s, or L4s. Pick a region (BLR, MUM, HYD) and tier.

Time: ~10 seconds
03

SSH & go

Copy the SSH command, paste in your terminal. CUDA, drivers, PyTorch and JAX are pre-installed.

Time: under 15s
Live metrics
012345678901234567890123456789+Active Developersbuilding on Race this month
<01234567890123456789sSSH Boot Timefrom click to shell prompt
01234567890123456789.0123456789%Uptime SLAenterprise-grade reliability
₹INRNo FX Surprisesfixed rupee pricing, always
★★★★★

“Spun up an H200 cluster in under 15 seconds. No unnecessary complexity, just SSH and train.”

AK
Aryan K.
ML Engineer
★★★★★

“INR billing saved us ₹4L a month in FX fees alone. Absolute no-brainer for Indian teams.”

PM
Priya M.
AI Researcher
★★★★★

“Switched from Lambda Labs. Same H100 hardware, 30% cheaper, and actual support in IST.”

RV
Rajan V.
Startup Founder
★★★★★

“The FLOP rewards stacked up fast. Essentially free H100 hours by month two.”

NS
Neha S.
DevOps Lead
★★★★★

“Pre-installed PyTorch and CUDA saved a full day of setup. Straight into training on day one.”

KT
Karan T.
Research Scientist
★★★★★

“Per-second billing is a game changer. We run hundreds of short jobs and the savings are real.”

DR
Divya R.
MLOps Engineer
★★★★★

“Spun up an H200 cluster in under 15 seconds. No unnecessary complexity, just SSH and train.”

AK
Aryan K.
ML Engineer
★★★★★

“INR billing saved us ₹4L a month in FX fees alone. Absolute no-brainer for Indian teams.”

PM
Priya M.
AI Researcher
★★★★★

“Switched from Lambda Labs. Same H100 hardware, 30% cheaper, and actual support in IST.”

RV
Rajan V.
Startup Founder
★★★★★

“The FLOP rewards stacked up fast. Essentially free H100 hours by month two.”

NS
Neha S.
DevOps Lead
★★★★★

“Pre-installed PyTorch and CUDA saved a full day of setup. Straight into training on day one.”

KT
Karan T.
Research Scientist
★★★★★

“Per-second billing is a game changer. We run hundreds of short jobs and the savings are real.”

DR
Divya R.
MLOps Engineer
// Pricing

Honest INR pricing. GST invoices on every bill.

Per-second billing. Pay for what you use, never for what you reserved. Fixed in rupees so your CFO sleeps easy.

L4E1
₹20/hr
  • ▸24 GB GDDR6 · 30 TFLOPS
  • ▸Inference & small fine-tunes
  • ▸1-click PyTorch / JAX / vLLM
  • ▸Available BLR / MUM / HYD

Excl. 18% GST · billed per second

A100 SXME2
₹65/hr
  • ▸80 GB HBM2e · 312 TFLOPS
  • ▸Multi-GPU NVLink clusters
  • ▸Best price, perfect for fine-tunes

Excl. 18% GST · billed per second

MOST DEMANDED
H100 SXME5
₹110/hr
  • ▸80 GB HBM3 · 3958 TOPS
  • ▸Train 70B models on a single node
  • ▸NVLink4 · 3.35 TB/s bandwidth

Excl. 18% GST · billed per second

// Zero setup

SSH in, CUDA, drivers, frameworks already there.

Every image is pre-warmed with the stack you'd otherwise spend three hours assembling. Type one command, train.

~ race-cli → bash
// pre-installed · race-base:24.10auto-updated weekly
PyTorch2.5.1cu124
JAX0.4.34cuda
CUDA12.4toolkit
vLLM0.6.3fp8
llama.cppb3982gguf
FlashAttn2.7.0v3
NCCL2.23multi-gpu
TensorRT-LLM10.6infer
Triton24.10server
Jupyter Lab4.2port 8888
Transformers4.46HF
ComfyUIlatestSD

+ Docker, conda, uv, tmux, htop, nvtop, nvidia-smi, git-lfs, ffmpeg, rclone

// Coming soon

Smart Scheduler.
Same iron. 40% off.

Submit a non-urgent training run, fine-tune, or batch inference job. We slot it into off-peak windows across the fleet. Same H200s, same A100s, same L4s — billed at up to 40% less.

01Submitrace job submit train.py
02Queuewaiting for off-peak
03Runoff-peak · same hardware
40%avg discount
2–6htypical wait
100%same hardware
EARLY ACCESS · Q3 2026

Get on the list.

First 100 builders get an extra 10% off their first ₹50K of scheduler runs. We email when the gates open — no spam.

ANKS
312 builders already in
FLOP Rewards

Train more, earn more.
The longer you ride,
the more you keep.

Every FLOP you spend earns FLOP credits. Hit higher Epochs, unlock multipliers, and redeem against future invoices.

Spend FLOPs
on training runs
›
Earn Credits
5–10% back
›
Hold to Level
reach next Epoch
›
Higher Epoch
bigger multiplier
›
Redeem
off future invoices
Click any Epoch to explore
E5
Reactor10%
≥ 10,00,000 FLOP·+ ≥ ₹30,000 / mo
E4
Hot8%
≥ 5,00,000 FLOP·+ ≥ ₹15,000 / mo
E3
Active7%
≥ 2,50,000 FLOP
E2
Ember6%
≥ 50,000 FLOP
E1
Spark5%
No minimum
FLOP Credit Calculator
FLOP
Current Epoch
E5
Reactor
10%
cashback
Credit Value
₹100 per 1,000 FLOP
₹—
Progress toMax Epoch
You're already at the top.
( FLOP ÷ 1,000 ) × Rate = Credit Value
// Start building

Stop waiting on the dollar. Start training in rupees.

No FX markups. No hidden egress. No GPU quotas. Just honest INR pricing and the fastest H200s in India.

// FAQ

Questions we get
all the time.

Loading FAQs...