Hosting Prices, Bill Drivers and 2026 Rules
On-demand vs reserved instances vs savings plans vs spot: cloud pricing models for Indian businesses
On-demand, reserved instances, savings plans and spot capacity explained: how each is priced and when each fits an Indian small business.
Public clouds sell the same server under several pricing models. On-demand is the default: no commitment, pay per second or hour, the highest unit price. Reserved instances, savings plans or committed-use discounts lower the price in exchange for a one- or three-year commitment, which you pay whether or not you use it. Spot or preemptible capacity is spare capacity at a deep discount that the provider can take back at short notice. The right mix depends on how steady each workload is. This guide explains each model for Indian MSMEs. It belongs to the hosting pricing overview, and the break-even example works through the sums.
The models side by side
| Model | What you commit to | Typical discount vs on-demand | Flexibility | Fits |
|---|---|---|---|---|
| On-demand | Nothing | None (the reference price) | Start and stop any time | New, variable or short workloads |
| Reserved instance (1 or 3 years) | A specific instance type or family in a region | Meaningful; deeper for 3 years and upfront payment | Limited; some can be exchanged or resold | Steady servers you will keep |
| Savings plan / committed spend | A fixed spend per hour for 1 or 3 years | Similar range; slightly less for flexible plans | Applies across sizes, sometimes across families or services | Steady spend, changing server mix |
| Committed-use discount (resource-based) | vCPU and memory in a region | Similar range | Tied to region and resource type | Steady compute |
| Sustained-use style discounts | Nothing; applied automatically on some clouds | Smaller | Automatic | Servers running most of the month |
| Spot / preemptible | Nothing; capacity can be reclaimed | Often the deepest | Interrupted with short notice | Batch jobs, testing, fault-tolerant workers |
Names and exact terms differ by provider. Discounts are published on each provider's price page per region and change over time, so this table does not state percentages.
On-demand: pay for flexibility
On-demand is right while you are still learning what you need. Use it for the first one to three months of any new workload, for development and test servers that can be switched off at night, and for seasonal peaks.
Illustrative example. A test server at an illustrative ₹6,000 a month on-demand runs only 10 hours a day on weekdays: about 220 of 730 hours. If it is stopped the rest of the time, the compute cost falls to about ₹6,000 × 220 ÷ 730 = ₹1,808. A commitment would cost more than this.
Reserved instances and savings plans: pay for steadiness
A commitment lowers the effective hourly price, but you are paying for capacity for the full term. Payment options usually include all upfront, partial upfront and no upfront (monthly); the deepest discount generally goes with all upfront and three years.
Illustrative example. A production database server costs an illustrative ₹10,000 a month on-demand. A one-year commitment at an illustrative 35% discount costs ₹6,500 a month, saving ₹3,500 a month or ₹42,000 a year before GST. A three-year commitment at an illustrative 55% discount costs ₹4,500 a month, but locks in the size for three years.
Risks of committing
- Paying for idle capacity if you downsize, move to a managed service or migrate away.
- Region lock: a commitment in Mumbai does not cover servers in Hyderabad.
- Technology change: a newer, cheaper instance generation may arrive during the term.
- Currency: an upfront dollar payment fixes the dollar amount, but the rupee cost depends on the exchange rate on the payment date. See USD vs INR billing.
- Business change: a commitment outlives a project that is cancelled.
A practical rule
Commit only the baseline you have measured for at least two to three months, not the peak. Cover the peaks on-demand. Start with one year unless you are confident about three.
Spot and preemptible capacity
Spot capacity can be very cheap but can be reclaimed with a short warning (often seconds to a couple of minutes). It is suitable for:
- batch processing, report generation and media conversion,
- build and test pipelines,
- stateless web workers behind a load balancer, mixed with on-demand servers.
It is not suitable for a single database or anything that cannot be interrupted.
Commitments on VPS and dedicated hosting
Outside the big clouds, "commitment" usually means prepaying months or years for a discount. A three-year prepaid VPS at a low monthly equivalent may look cheap, but check the renewal price, refund terms and whether you can upgrade mid-term. See renewal price jumps.
A simple mix for a small production setup
| Workload | Suggested model (illustrative) |
|---|---|
| Production web and database servers running 24×7 for 12+ months | 1-year commitment on the measured baseline |
| Extra web servers for festive peaks | On-demand, autoscaled |
| Nightly reports and image processing | Spot |
| Development and test | On-demand, switched off out of hours |
Accounting and tax points
An upfront commitment is a prepayment. Discuss with your CA how it is expensed over the term and how GST or reverse charge applies when the provider invoices. If the provider is outside India and you are GST-registered, reverse charge applies to the supply. See GST, reverse charge and TDS.
FAQ
Can I cancel a reserved instance?
Usually not for a refund. Some providers let you exchange or sell certain reservations; check the terms before buying.
Do commitments cover managed databases?
Some providers offer separate reservations for managed databases and other services. Check which services each plan covers.
Is spot safe for my website?
Only if the site runs on several servers and losing one does not take it down. Do not run a single-server site on spot.
How long should I measure before committing?
At least two to three months including a busy period, such as a quarter-end or a festive season.
Sources
- AWS EC2 pricing — on-demand, savings plans and spot pages per region
- Google Cloud regions — regions in Mumbai and Delhi
- Azure regions — regions in Pune, Chennai, Mumbai and Hyderabad
- CBIC GST — reverse charge on imported services
