Originally published at aicloudstrategist.com/blog/savings-plans-vs-reserved-instances-mumbai.html. This is a cross-post for the dev.to community.
Savings Plans vs Reserved Instances for Indian Startups: Mumbai Region ROI Math (2026)
By Anushka B, Founder · 2026-04-22 · 11 min read
Most Indian startups pick Savings Plans by default because the AWS console nudges them that way. For ~85% of workloads that's correct, but the 15% where Reserved Instances still win are worth ₹5-20 lakh/year — and the Mumbai-specific INR math decides which bucket you're in. Here is the detailed break-even analysis we run for Series A-C stacks.
If you want the high-level overview first, start with our existing Savings Plans vs RI primer for India. This post goes deeper: Mumbai-specific list prices, discount-band tables, break-even curves, and the five workload archetypes where the choice actually changes the answer.
The three commitment products, 2026 state
Compute Savings Plan — commits $X/hour of compute spend. Applies to EC2 any family/size/OS/tenancy/region, Fargate, Lambda. Max flexibility, slightly smaller discount.
EC2 Instance Savings Plan — commits $X/hour to a specific instance family in a specific region. Bigger discount than Compute SP, but family-locked.
Standard Reserved Instance — 3-year all-upfront on a specific family/size/AZ/OS. Deepest discount, least flexibility. Convertible RI gives up ~6% discount for family-swap rights.
Mumbai region discount bands (April 2026)
| Product | Term | Payment | Mumbai discount vs On-Demand |
|---|---|---|---|
| Compute Savings Plan | 1 year | No upfront | ~27% |
| Compute Savings Plan | 1 year | All upfront | ~31% |
| Compute Savings Plan | 3 year | No upfront | ~45% |
| Compute Savings Plan | 3 year | All upfront | ~52% |
| EC2 Instance Savings Plan | 1 year | No upfront | ~32% |
| EC2 Instance Savings Plan | 3 year | All upfront | ~58% |
| Standard RI | 1 year | All upfront | ~36% |
| Standard RI | 3 year | All upfront | ~62% |
| Convertible RI | 3 year | All upfront | ~54% |
Reading this table, the naive answer is "3-year Standard RI, always." The real answer is "it depends on your cash runway, growth rate, and how locked-in your instance family is." Below is how we decide.
The 5 workload archetypes and the right commitment for each
Archetype 1: Early-stage multi-service SaaS on m6i/c6i/r6i mix (60% of Series A we audit)
Recommendation: 1-year Compute Savings Plan, no upfront. Cash preservation > extra 4% from all-upfront. Flexibility to change families as the product evolves.
Mumbai math on a ₹10 lakh/month steady-state compute bill: 75% coverage × 27% discount = ₹2.02 lakh/month saved = ₹24.3 lakh/year. Commitment = $8,300/month. No cash upfront.
Archetype 2: Stable workload post product-market fit, 90%+ on one family (late Series B, Series C)
Recommendation: 3-year EC2 Instance Savings Plan, partial upfront. The family lock is real but tolerable, and you gain ~13% extra discount vs Compute SP.
Mumbai math on ₹40 lakh/month spend, all on r6i: 80% coverage × 52% discount = ₹16.6 lakh/month saved = ₹1.99 cr/year.
Archetype 3: Fintech on RBI-constrained Mumbai, known 3-year workload
Recommendation: 3-year Standard RI, all-upfront, on the production DB and API tier + 1-year Compute SP for the rest. RDS RIs are only available as RIs (no Savings Plan), so this is forced for the DB layer. The 62% discount is too big to pass on regulated, mature workloads.
Caveat: only if the fintech has the cash. An early-stage fintech with 18 months' runway should stick with 1-year no-upfront even if the math is worse — optionality is worth money.
Archetype 4: AI/GPU startup with bursty training
Recommendation: No Savings Plan on the training fleet; 1-year Compute SP on the always-on inference tier only. See our GPU cost audit guide. Commit only to the steady inference portion — training goes to Spot + Capacity Blocks.
Archetype 5: Early-stage startup under 24 months runway
Recommendation: No upfront commitments of any kind. If you must commit, do 1-year no-upfront on the absolute floor of your compute (50-60% of current). The additional 5% from all-upfront isn't worth the cash-flow risk.
The Mumbai-specific complication: RDS RI is not interchangeable with Savings Plans
Savings Plans do not cover RDS. For Postgres, MySQL, SQL Server, Oracle, and Aurora, RIs are your only commitment option. A 3-year Standard RI on db.r6i.2xlarge in Mumbai gives ~58% off On-Demand; a 1-year no-upfront gives ~34%. For Indian SaaS with a sizable RDS spend (typically 15-25% of the bill), buying RDS RIs is non-optional once you cross ₹3 lakh/month on databases.
Trap: the AWS console buys Regional RDS RIs by default. These do not guarantee capacity, but they do apply the discount across AZs in the region. Zonal RIs guarantee capacity in a specific AZ but lock you harder. For Indian startups we almost always recommend Regional RDS RIs — capacity in Mumbai is rarely a problem.
The "RI coverage report" you should be running monthly
AWS gives you a Coverage Report under Cost Explorer -> Reservations. Most startups look at it once, commit, and forget. We run it monthly for our retainer clients and check three numbers:
Coverage % — compute hours covered by SP+RI vs total eligible hours. Target 75-85%.
Utilisation % — SP/RI hours actually used vs committed. Target 98%+. Under 95% means you over-committed.
Effective Savings Rate (ESR) — actual discount realised vs list price, net of unused commitments. Target 28-32% for 1-year SP, 48-55% for 3-year.
See RI coverage governance for Indian startups for the full monthly ritual we run.
Break-even analysis: 1-year no-upfront vs 3-year all-upfront in Mumbai
Scenario: ₹12 lakh/month committed compute (~$14,500/month at 82 INR/USD).
| Option | Cash day 1 | Monthly discount saved | 3-year total saved |
|---|---|---|---|
| 1-year no-upfront Compute SP, renewed yearly | ₹0 | ₹3.24 lakh | ₹1.17 cr |
| 3-year no-upfront Compute SP | ₹0 | ₹5.40 lakh | ₹1.94 cr |
| 3-year all-upfront Compute SP | ₹2.08 cr | ₹6.24 lakh (effective) | ₹2.24 cr |
| 3-year all-upfront EC2 Instance SP | ₹2.32 cr | ₹6.96 lakh (effective) | ₹2.50 cr |
The 3-year all-upfront EC2 Instance SP saves ₹1.33 cr more than 1-year rolling — but requires ₹2.32 cr cash upfront and locks your instance family for 36 months. For a cash-rich Series C with a mature product, it's the right call. For a Series A it is almost never.
What we see in audits: 4 recurring commitment mistakes
Committing before rightsizing. Buys SP against inflated baseline. Saves 27% of a wrong number.
Over-committing to "hedge growth." Founder anticipates 3x growth, commits to 2x current. Growth doesn't arrive; utilisation drops to 70%; ESR collapses.
RI for DynamoDB or ElastiCache ignored. Both have their own Reserved Capacity products. Indian SaaS with ₹2 lakh+/month on DynamoDB routinely ignores DynamoDB Reserved Capacity (~42% off).
Savings Plan in the wrong payer account. In AWS Organizations, SPs bought in member accounts don't share across the org. Always buy in the management account with RI sharing on.
The founder's decision tree, one paragraph
If runway < 18 months: no commitments, optimise On-Demand instead. If 18-30 months: 1-year no-upfront Compute SP on 60-70% of steady state. If 30+ months and product-market fit confirmed: 1-year all-upfront Compute SP or 3-year no-upfront Compute SP on 75-80%. If 36+ months of cash, stable family, regulated workload: 3-year all-upfront EC2 Instance SP + 3-year RDS RI. Never exceed 85% coverage — the last 15% is where you absorb spikes and new services.
Frequently asked questions
Q: Can I buy Savings Plans in INR?
No. AWS bills in USD globally. Indian entities pay via INR-to-USD conversion (typically by HDFC/ICICI corporate cards or RBI LRS route for smaller amounts). Budget a 0.5-1% forex spread in your ROI math.
Q: What happens to my Savings Plan if I move from ap-south-1 to ap-south-2?
Compute Savings Plans apply across regions automatically — your ap-south-2 spend gets the same discount. EC2 Instance Savings Plans and RIs are region-locked; you'd need to modify or sell the RI on the marketplace.
Q: Can I sell unused RIs on the AWS Marketplace from India?
Yes, if your AWS account is US-based. From an Indian-incorporated account, the RI Marketplace sell-side is restricted. Convertible RIs can always be exchanged regardless of account geography.
Q: How does Graviton (c7g, m7g) affect this math?
Graviton is typically 15-20% cheaper than x86 equivalents in Mumbai. If your workload can run on ARM, switch first, then commit. Committing to x86 when ARM would work is a common mistake; the commitment doesn't translate.
Q: What's the ESR I should target?
Effective Savings Rate of 30-35% on the total compute bill (including uncommitted On-Demand and Spot) is healthy for a post-audit Series A-B Indian SaaS. Above 40% usually means over-commitment risk; below 22% means under-commitment.
Q: Should we stack SP with Spot?
Savings Plans don't apply to Spot (Spot is already discounted separately). Your commitment covers only On-Demand equivalent hours. A healthy mix: 70% SP-covered On-Demand + 25% Spot + 5% uncommitted On-Demand for bursts.
Q: How do you audit commitment posture without console access?
We need the Cost and Usage Report (CUR) with resource-level granularity and RI/SP utilisation data. From that we can reconstruct coverage %, utilisation %, and ESR offline. No IAM access required.
Related reading: Savings Plans vs RI for India (primer) · RI coverage governance · AWS cost audit for India · FinOps assessment · AWS cost calculator
We run written 24-hour AWS cost audits. Founder-led. Free. No sales call. Send your last bill, get a PDF back. Request yours →
Top comments (0)