Short answer

• For the very lowest latency in India you need either (a) a trading member with co‑located servers at the exchange (colocation / DMA) or (b) a specialist low‑latency/ institutional API provider. Retail mobile/web apps are much slower than colocated/DMA setups. (Reuters.com)

Who/what to look at (providers known for low‑latency execution)

• Zerodha (Kite Connect) — popular retail/algo API; widely used for automated trading and low latency retail execution (not exchange‑colocation by default; many users run servers close to Mumbai/AWS to reduce latency). (afaqs.com)
• Upstox — fast trading platform and APIs with developer support; used by many algo traders. (Upstox.com)
• Specialist low‑latency providers / DMA / HFT vendors — firms that sell co‑location/DMA, normalized low‑latency feeds and order gateways (examples: Estee Advisors, Pocketful and other institutional connectivity vendors). These are built for very high throughput and millisecond execution. (esteeadvisors.com)
• Full‑service bank brokers (ICICI Direct, Kotak, HDFC, Motilal Oswal) — larger infrastructure and stable execution during volatile markets; some offer algorithmic/DMA services for institutional clients. (afaqs.com)

Why colocation / DMA matters

• Colocation places your matching engine / gateway inside or next to the exchange data center so network hops and milliseconds are removed; exchanges (NSE/BSE) offer co‑location and have been expanding capacity because demand is high. Colocation is the only practical way to get single‑digit millisecond end‑to‑end latencies. (Upstox.com)

Practical tiers (what “fastest” means depending on your needs)

• Retail API / Cloud server near Mumbai (best for most algo/automated retail traders): 10s–100s ms depending on setup and processing. Use brokers’ APIs (Kite, Upstox, etc.) and host in Mumbai/AWS region or colocated VPS. (kite.trade)
• Institutional / DMA with colocated racks: single‑digit to low‑double‑digit milliseconds end‑to‑end; requires exchange approval, membership or a trading member that offers colocated access. Best for high‑frequency or latency‑sensitive execution. (Upstox.com)

How to choose the right provider (brief checklist)

1. Latency guarantees / real measured latency (ask for ping/trace tests from your server location to their gateway). (Pocketful.in)
2. Exchange access type: retail API vs DMA vs colocated rack. Colocation = lowest latency. (Upstox.com)
3. Throughput limits, order throttling and rate limits. (esteeadvisors.com)
4. Market data feed quality (tick‑by‑tick vs aggregated) and timestamp accuracy. (community.Upstox.com)
5. Costs: setup, monthly colocation/DMA fees, per‑order brokerage and data fees. (esteeadvisors.com)
6. Regulatory/approval requirements (SEBI/exchange rules for algos/HFT; some services require member approval). (Upstox.com)

Recommended next steps (if you want to move fast)

1. Decide your target latency and order volume (retail algo vs HFT).
2. If retail algorithmic trading is enough: trial Kite Connect (Zerodha) and Upstox APIs and host your execution server in Mumbai/AWS; measure round‑trip latency. (kite.trade)
3. If you need sub‑20 ms latency or higher throughput: contact DMA/colocation vendors (Estee Advisors, institutional brokers or specialised firms like Pocketful) to discuss colocated racks or a colocated gateway and pricing. Ask for real latency numbers and rack/port availability. (esteeadvisors.com)
4. Validate by doing latency tests (ICMP/TCP pings, order round‑trip measurements) and review real order fill times under load.

If you’d like, I can:

• pull and compare specific latency / API docs and pricing pages for 3–5 brokers (Zerodha, Upstox, Pocketful, Estee/other DMA) and summarize which is best for your exact latency target, or
• draft an email/template you can send to colocated/DMA providers to request latency numbers, fees and trial access.

Which of those two would you prefer?