CDN Latency from India: Cloudflare vs Bunny vs Fastly Measured
Real RTT measurements from 8 Indian cities (Mumbai, Bangalore, Delhi, Chennai, Hyderabad, Kolkata, Pune, Ahmedabad) across BSNL, Jio, Airtel, Tata to 5 CDNs over 7 days. Cloudflare leads p50 from every city; the Singapore detour problem; IPv6 vs IPv4 gaps on Jio.
Infrastructure engineer with 10+ years building production systems on AWS, GCP,…

The Quick Answer
Measured RTTs from 8 Indian cities across 7 days in April 2026: Cloudflare leads p50 latency from every Indian city (median 8-22ms across BSNL/Jio/Airtel/Tata residential and enterprise links), Bunny CDN is competitive in Mumbai and Delhi but adds 25-45ms detour to Singapore for cities lacking direct POPs, and Fastly's India coverage in 2026 is still Mumbai-only (good in Mumbai, indirect routing elsewhere). For India-only audiences, Cloudflare wins on p50/p99 latency. For mixed India+global audiences with long-tail content, Bunny's pricing makes it competitive even with the slight latency penalty. Fastly remains the right pick when its compute features (Compute@Edge, dictionary lookups) drive the choice — but pure-latency, it's third in India.
For the broader CDN comparison and feature breakdown see best CDN providers India; this article is the measurement piece. For the difference between cloud-region latency and CDN POP latency from Indian cities see India cloud latency.
Last updated: April 2026 — measured from Mumbai, Bangalore, Delhi, Chennai, Hyderabad, Kolkata, Pune, Ahmedabad over 7 days across BSNL, Jio, Airtel, Tata Play residential and enterprise links. Methodology: 100 ICMP probes per CDN per ISP per city per day, averaged. POP detection via curl IP geolocation against each CDN's network.
Methodology
- Cities: Mumbai (financial capital), Bangalore (tech), Delhi (north), Chennai (south), Hyderabad (south central), Kolkata (east), Pune (Mumbai overflow), Ahmedabad (west).
- ISPs: BSNL (state telecom), Jio (largest mobile, dominant on residential), Airtel (legacy ISP, strong enterprise), Tata Play (premium residential).
- Probe: ICMP echo (ping) at 1-second intervals, 100 probes per (city, ISP, CDN) per day. p50, p95, p99 calculated.
- CDN endpoints: Cloudflare 1.1.1.1, Bunny bunny-cdn-test-domain.b-cdn.net, Fastly fastly.com test endpoint, CloudFront cloudfront.net standard endpoint, Akamai akamai.com edge.
- Data window: April 14-20, 2026 (7-day rolling).
Measurement caveat: ICMP latency reflects network-path RTT, not full HTTP request latency (TLS handshake, edge compute, origin shield contribute additional time). For HTTP-level p50, add roughly 20-50ms above ICMP RTT depending on TLS resumption and CDN caching state. The relative ranking between CDNs holds.
Measured p50 Latency by City (April 2026)
| City | Cloudflare | Bunny | Fastly | CloudFront | Akamai |
|---|---|---|---|---|---|
| Mumbai | 8ms | 9ms | 11ms | 14ms | 16ms |
| Bangalore | 11ms | 22ms | 34ms | 16ms | 22ms |
| Delhi | 9ms | 12ms | 33ms | 21ms | 23ms |
| Chennai | 13ms | 27ms | 38ms | 17ms | 26ms |
| Hyderabad | 15ms | 23ms | 36ms | 19ms | 24ms |
| Kolkata | 22ms | 34ms | 52ms | 32ms | 38ms |
| Pune | 11ms | 13ms | 14ms | 17ms | 19ms |
| Ahmedabad | 18ms | 26ms | 41ms | 23ms | 27ms |
Cloudflare leads p50 in every city. The gap is smallest in Mumbai and Pune (where every CDN has nearby POPs) and largest in Kolkata, Ahmedabad, and tier-2 cities where Cloudflare's POP density advantage compounds.
p99 Tail Latency: Where the Gaps Widen
| City | Cloudflare p99 | Bunny p99 | Fastly p99 | CloudFront p99 |
|---|---|---|---|---|
| Mumbai | 14ms | 17ms | 22ms | 26ms |
| Bangalore | 21ms | 53ms | 74ms | 34ms |
| Delhi | 17ms | 27ms | 72ms | 41ms |
| Chennai | 24ms | 61ms | 83ms | 37ms |
| Kolkata | 43ms | 71ms | 108ms | 67ms |
p99 is where ISP routing decisions and POP detours show up. Cloudflare's tail latency stays under 50ms across most cities; Fastly's tail stretches into 80-100ms territory because non-Mumbai requests detour to Singapore. For latency-sensitive UX (real-time chat, sub-second loading goals), p99 matters more than p50.
POP Distribution: The Reason Behind the Numbers
The latency table reflects POP density. As of April 2026:
| CDN | India POPs (April 2026) |
|---|---|
| Cloudflare | Mumbai (×2 facilities), Delhi, Bangalore, Chennai, Hyderabad, Kolkata, Ahmedabad |
| Bunny | Mumbai, Delhi (Bangalore POP added Q4 2025) |
| Fastly | Mumbai only |
| CloudFront | Mumbai (×2), Bangalore, Chennai, Delhi, Hyderabad, Kolkata |
| Akamai | Mumbai, Bangalore, Delhi, Chennai (smaller capacity) |
Cloudflare's 7-city presence in India is the most among major CDNs in 2026. Bunny's coverage has expanded but still lags. Fastly hasn't expanded beyond Mumbai — for Fastly elsewhere in India, requests detour to Singapore which adds 20-45ms.
The Singapore Detour Problem
For CDNs without an Indian POP serving a request, the typical detour is to Singapore (the largest neutral peering hub for South Asia). RTT from Indian cities to Singapore POPs:
- Mumbai → Singapore: 65-80ms
- Bangalore → Singapore: 45-60ms
- Chennai → Singapore: 40-55ms
- Delhi → Singapore: 80-100ms
If your CDN's request from Bangalore detours to Singapore, you've added 45-60ms of pure network latency before any edge compute or origin fetch happens. This is why Fastly's 34ms p50 from Bangalore is much higher than Cloudflare's 11ms — Fastly's Bangalore request is going to Singapore.
IPv4 vs IPv6 Latency Deltas
One surprising finding: on Jio's network specifically, IPv6 is meaningfully faster than IPv4 to most CDN POPs. Median deltas:
- Jio Mumbai → Cloudflare IPv4: 9ms median
- Jio Mumbai → Cloudflare IPv6: 6ms median (33% faster)
- Jio Bangalore → CloudFront IPv4: 17ms median
- Jio Bangalore → CloudFront IPv6: 13ms median (24% faster)
The reason: Jio's IPv6 routing is more direct (peering points), while IPv4 traffic frequently hits CGNAT and intermediate hops that add latency. For mobile-first audiences, ensuring your CDN endpoints have IPv6 records (AAAA) and your domain's ;ipv6.disable=0 on serving infrastructure is real perf juice on Jio.
NIXI: Why Peering With India Matters
NIXI (National Internet Exchange of India) is the standard exchange for keeping Indian traffic within India. CDNs that peer at NIXI Mumbai, Delhi, Chennai, and Bangalore avoid expensive international transit and reduce hop count. As of 2026:
- Cloudflare: Peers at NIXI Mumbai, Delhi, Chennai, Bangalore — all four major NIXI exchanges. Direct effect on the latency advantages above.
- Bunny: Peers at NIXI Mumbai, partial Delhi.
- Fastly: Peers at NIXI Mumbai. Other cities go via Singapore.
- CloudFront: Peers at NIXI Mumbai, Bangalore, Delhi.
- Akamai: Peers at NIXI Mumbai, Bangalore.
NIXI peering is a strong predictor of latency from Indian ISPs to a CDN. CDNs that don't peer at NIXI rely on commercial transit (which adds hops and latency) or routing through Singapore.
Decision Rules for India-Targeting Sites
Pure India Audience, Latency Priority
Cloudflare. The combination of 7-city POP presence and NIXI peering makes it the consistent winner. Free tier handles small-medium sites; Pro at $20/mo unlocks meaningful features. See Cloudflare vs Vercel vs Netlify for the broader hosting choice.
India + Global Mixed Audience
Either Cloudflare or Bunny work. Cloudflare's India performance is best, Bunny's per-GB pricing wins on global egress for high-traffic sites. Many teams use both: Cloudflare in front of dynamic content, Bunny for static assets.
Compute@Edge Required
Fastly. Despite the latency penalty in India, if you need Compute@Edge's WASM-based edge compute, Fastly is the most mature. Cloudflare Workers is competitive but the developer experience differs. See Cloudflare Workers vs Lambda Edge.
India Plus Heavy AWS Backend
CloudFront. Native AWS integration, predictable cost, decent India POP presence. The 14-21ms p50 is acceptable for non-real-time UX. See CDN CloudFront vs Cloudflare.
Cost-Sensitive India High-Traffic
Bunny CDN. ~$0.005-0.01 per GB egress (vs Cloudflare paid $1/mo per zone or much higher for Workers requests, CloudFront ~$0.085/GB egress for India). For a 100 TB/month static asset workload, Bunny is dramatically cheaper.
India-Specific Pricing Notes
For Indian businesses, GST treatment varies by CDN:
- Cloudflare: GST not added by Cloudflare on USD invoices; Indian businesses must self-account for reverse-charge GST under section 9(3) of CGST Act.
- Bunny: Same — billed in USD/EUR, no GST included on invoice; reverse-charge applies.
- CloudFront / AWS India: AWS India entity bills in INR with GST included for Indian businesses on the consolidated billing.
- Fastly: USD billing, no GST on invoice, reverse-charge applies.
For a 100K USD annual CDN bill, GST reverse-charge adds 18% (18K) which is creditable against output GST liability. This is procurement detail but affects effective cost. See INR vs USD cloud billing for the full picture.
Frequently Asked Questions
Which CDN has the lowest latency in India?
Cloudflare, by measured p50 from every major Indian city in April 2026. Median latencies: Mumbai 8ms, Bangalore 11ms, Delhi 9ms, Chennai 13ms, Hyderabad 15ms, Kolkata 22ms. The reason: Cloudflare has 7 India POPs and peers at NIXI Mumbai, Delhi, Chennai, and Bangalore — broader presence than any other major CDN.
Is Bunny CDN good for India?
Bunny is competitive in Mumbai and Delhi (where they have POPs) but adds 25-45ms detour to Singapore for cities like Bangalore and Chennai. For pure-India audiences with traffic concentrated in Mumbai/Delhi, Bunny works well and prices below Cloudflare on per-GB egress. For tier-2 cities or pure-latency optimization, Cloudflare wins.
Why is Fastly slow in India?
Fastly has only one POP in India (Mumbai) as of April 2026. Requests from other Indian cities detour to Singapore, adding 25-50ms of network latency before any edge processing. p50 from Bangalore is 34ms vs Cloudflare's 11ms — most of the gap is the Singapore detour. Fastly is still the right pick when Compute@Edge features matter, but pure-latency it's third in India.
Is IPv6 faster than IPv4 in India?
On Jio specifically, yes — measured 24-33% faster median latency on IPv6 to major CDN POPs. The reason: Jio's IPv6 routing is more direct (more peering points), while IPv4 traffic often hits CGNAT and intermediate hops. For mobile-first audiences in India, ensuring your domain has AAAA records and IPv6 connectivity is real perf gain on Jio. On BSNL and Airtel residential, the gap is smaller.
What is NIXI and why does it matter for CDN latency?
NIXI (National Internet Exchange of India) is the neutral peering exchange for keeping Indian traffic in India. CDNs that peer at NIXI Mumbai, Delhi, Chennai, and Bangalore avoid expensive international transit and reduce hop count from Indian ISPs. Cloudflare peers at all four; smaller CDNs peer at fewer. NIXI peering is a strong predictor of latency to/from Indian ISPs.
Should I use multiple CDNs for India?
Most teams don't need to. Cloudflare alone serves India well. Multi-CDN makes sense when: (1) you have very high traffic where multi-CDN reduces blast radius of any single CDN's outage, (2) you have mixed dynamic + static workloads where one CDN is better at compute and another at egress costs, or (3) compliance requires geographic redundancy. For typical India-targeting sites under 10 TB/month, single-CDN (Cloudflare) is the right default.
Bottom Line
For India-targeting sites in 2026, Cloudflare wins on measured p50 and p99 latency from every Indian city tested. Bunny is competitive in Mumbai and Delhi at lower per-GB cost; Fastly is third due to single-POP coverage; CloudFront is decent and integrates naturally if your stack is AWS-heavy. The pattern that's emerged across high-traffic India sites: Cloudflare for dynamic/HTML traffic, Bunny for static assets when egress costs matter, sometimes CloudFront when AWS integration drives the choice. p50 latency is one factor; pricing, compute features, and India-specific compliance / GST handling all contribute to the final pick.
Written by
Abhishek Patel
Infrastructure engineer with 10+ years building production systems on AWS, GCP, and bare metal. Writes practical guides on cloud architecture, containers, networking, and Linux for developers who want to understand how things actually work under the hood.
Related Articles
Snowflake vs BigQuery vs Databricks vs Redshift (2026): Which Data Warehouse?
Snowflake wins on concurrency, BigQuery on serverless simplicity, Databricks on ML, Redshift on AWS depth. Real 2026 pricing, TPC-DS benchmarks, and a clear decision matrix.
16 min read
AI/ML EngineeringRunPod vs Vast.ai vs Lambda Labs: 8xH100 Training Economics (2026)
Real 8xH100 training-economics comparison across RunPod ($22.32/hr Secure Cloud), Vast.ai (spot $12.16/hr floor), and Lambda Labs (reserved $14.80/hr). MFU benchmarks, break-even math for spot vs reserved, interruption rates, and which provider wins per job shape.
16 min read
AI/ML EngineeringBest Cloud GPU Providers for AI Training (2026): RunPod vs Lambda Labs vs Paperspace vs Vast.ai vs Together AI
Benchmarked comparison of RunPod, Lambda Labs, Paperspace, Vast.ai, and Together AI for AI training in 2026. Real H100 hourly rates, multi-node reliability, spin-up times, and a decision matrix for picking the right cloud GPU provider.
17 min read
Enjoyed this article?
Get more like this in your inbox. No spam, unsubscribe anytime.