⚡ INDIA POWER

India GPU Electricity Cost Calculator

A GPU's sticker $5/hr never shows the power bill. Here's what an H100, A100, B200 or RTX 5090 actually draws in kWh, what that costs in ₹/kWh by your state, and the hidden power cost it stacks onto every rental hour — live.

⚡ Power draw modeled at real utilization · 🧮 Tariff = state-wise commercial estimate (May 2026), fully adjustable

1. Your scenario

Tariff & usage

Tariff slider auto-ranges to your state's commercial band (shown at right). Real draw = rated W × (0.45 + 0.55 × util%) — an idle GPU still burns ~45% of rated power. Utilization is the honest lever most cost estimates ignore.

Your power bill

₹ figures live-computed from your tariff + GPU power draw + utilization. Rental $/hr defaults from the gpuprice.in feed (). Power model is an engineering estimate, not a utility bill.

All GPUs at this tariff

GPU Real draw Power ₹/hr ₹/month Sticker rent Power vs rent

"Power vs rent" = power ₹/hr as a % of the live sticker rental ₹/hr (INR = ₹83/USD). When it climbs past ~5%, electricity is a real component of your true GPU cost — exactly the hidden line item this tool exists to surface.

Which state has the cheapest power for this GPU?

Priced for your current GPU + utilization + hours/day, at every state's tariff midpoint — cheapest first. See where it's worth colocating.

State (₹/kWh band) Power ₹/hr ₹/month/GPU % of rent Vs cheapest

Ranks use each state's commercial-tariff midpoint (May-2026 estimate): Gujarat & UP start lower (₹3.60–7.30 / ₹3.35–8.00), Maharashtra / Karnataka ride higher, especially in metros with private DISCOMs. Real rates depend on slab, category & DISCOM — this is a directional colocation planner, not a bill.

Why this matters (honest methodology)

This is a directional power-cost estimator, not a utility bill or financial advice. Always verify against your DISCOM's published tariff and your provider's meter. INR↔USD uses ₹83 from the live feed.

Frequently asked questions

How is real power draw calculated?

Rated power (H100=700W, A100=400W, RTX 5090=575W, H200≈700W, B200≈1000W) scales with utilization: real draw = rated × (0.45 + 0.55 × util%). At 70% an H100 draws ≈ 600W. Multiply by hours/day × days/month × ₹/kWh tariff for the bill. It's an estimate — real draw varies with workload (training vs inference), cooling, and PSU efficiency.

Why does the tariff slider range differ per state?

Each state's commercial tariff spans from a subsidized low slab to a high upper slab. We load your state's band (e.g. Gujarat ₹3.60–7.30, Maharashtra ₹4.43–12.80) when you pick it, so the slider only offers realistic ₹/kWh for that state. Midpoint is the default; slide to your exact DISCOM rate or slab.

Is electricity included in cloud GPU rental prices?

Most on-demand cloud GPU rentals (Vast, RunPod, Lambda, Spheron) bundle power into the hourly rate, so your power bill is already covered. But dedicated/colocation or "power measured separately" deals bill it on top — that's where this tool matters most. If you own GPUs, power is always a separate permanent cost.

Which GPU is cheapest to run per hour in India?

The RTX 5090 (≈575W) draws less than an H100/B200 and costs much less to buy. But "cheapest to run" also depends on what you need: for heavy training you need the compute, not just the lowest wattage. Power ₹/hr here = tariff × real kW, so an H100 at ~0.6 kW ≈ ₹3.6/hr at ₹6/kWh — compared to a rental sticker of $2.50/hr (~₹210/hr), a fraction.