Networks, Automation, Connectivity & Data Centres · Chapter 14 of 18
What infrastructure actually costs — India versus the US, worked
The rent ratio says India is 2.4× cheaper. The all-in ratio says 1.3×. The gap is the whole lesson, and it is not where people expect.
This chapter does one thing: it builds the same ten-rack deployment in Mumbai and in Ashburn, line by line, and shows why the headline comparison is misleading. The arithmetic is more valuable than the numbers, because the numbers will move and the method will not.
A note on sourcing, because it is unusual here. Not one Indian tier-one colocation operator publishes a rack or kilowatt rate — not STT GDC, NTT, Sify, CtrlS, Nxtra, Yotta, Equinix India, AdaniConneX, Web Werks or ESDS. Every ‘₹25,000–60,000 per rack’ figure circulating online traces to content-marketing blogs rather than to a price list.
The only genuinely published Indian colocation rate card is a government one, and it prices at roughly three times the analyst market rate. That gap is itself worth teaching: a published rate is not a market rate.
So the Indian figures below are analyst estimates with the firm named, and the estimates are labelled. The American figures are largely published.
The unit of account is the kilowatt, not the rack
The single published Indian rate card makes this beautifully explicit: its per-rack price and its per-server price work out to exactly the same rupees per kilowatt per month. Space is not what you are buying. If someone quotes you a rack price without a kilowatt figure attached, you have not been quoted anything.
That is the first question to ask about any colocation price: at what kW, and does it include electricity? Most of the confusingly cheap numbers you will see are space-only, or assume 2 to 3 kW racks, or exclude power entirely.
Mumbai: ten racks at 5 kW, one year
Assumptions, stated so you can vary them: 50 kW committed, average draw at 50% of committed (the industry norm, and the single largest swing factor in the whole calculation), PUE 1.6, four cross-connects, two 1 Gbps transit commits.
| Line | Basis | Annual |
|---|---|---|
| Colocation rent | 50 kW × ~$90/kW/month (analyst midpoint for India) | ~$54,000 ≈ ₹47.5 lakh |
| Power at 50% draw, grid tariff | 25 kW IT × 1.6 PUE = 40 kW metered → 350,400 kWh × ~₹9.20 | ~$36,600 ≈ ₹32.2 lakh |
| Cross-connects | 4 × ~₹10,000/month (estimated — not published in India) | ~$5,450 |
| IP transit | 2,000 Mbps × ~$0.35/Mbps/month (estimated) | ~$8,400 |
| Total | ~$104,500 ≈ ₹92 lakh — about ₹76,500 per rack per month | |
| Same, on an open-access renewable PPA at ~₹5.50/kWh | ~$89,750 ≈ ₹79 lakh |
The open-access line is not a footnote — it is the whole Indian data centre business model
Compare the two power lines. Grid tariff costs about $36,600 a year; an open-access renewable power purchase agreement costs about $21,900. That is a 40% reduction on the largest variable line in the bill.
And it reconciles a discrepancy you would otherwise trip over: analyst figures putting Indian electricity opex at $0.7–0.9 million per megawatt per year are only achievable on open access, not on the state grid tariff.
Which is why Maharashtra's regulatory package — no additional surcharge on open access, energy banking permitted outside evening peak, and a 10% wheeling discount for facilities on 100% green energy — matters more to Indian data centre economics than the headline tariff does. If you are asked what drives Indian data centre viability, the answer is regulatory access to cheap renewable power, not cheap grid power.
Ashburn: the same ten racks
Two contract structures are worth showing, because they price the same physics differently.
| Line | Bundled retail | Triple-net (base + metered power) |
|---|---|---|
| Colocation rent | 50 kW × ~$220/kW/month = $132,000 | 50 kW × ~$150/kW/month = $90,000 |
| Power | included | 25 kW IT × 1.4 PUE = 35 kW → 306,600 kWh × ~$0.0935 = ~$28,700 |
| Cross-connects | 4 × ~$200/month = $9,600 | $9,600 |
| IP transit | ~$7,200 | $7,200 |
| Total | ~$148,800 | ~$135,500 |
Note that the two totals converge, and that convergence is the lesson: the two models price the same physics and differ in who carries utilisation risk. If you reliably draw well below your commitment, triple-net wins. If you spike, bundled wins. That is the actual question to ask before choosing, and it is answerable from your own historical draw data.
One more thing the American figures show that the Indian ones cannot: the per-kilowatt rate is a function of deal size. Published rates for a 250–500 kW requirement sit around $196/kW/month nationally, while 10 MW+ requirements in Northern Virginia price at $155–185. A ten-rack deployment sits above the small band, not below it. The spread between a 50 kW deal and a 10 MW deal is roughly 30 to 40%, and you cannot buy the wholesale rate at retail size.
The comparison, and the result that surprises people
| Line | Mumbai (grid) | Ashburn (NNN) | Ratio |
|---|---|---|---|
| Colocation rent | $54,000 | $90,000 | 1.7× — India cheaper |
| Power | $36,600 | $28,700 | 0.78× — India more expensive |
| Cross-connects | $5,450 | $9,600 | 1.8× — India cheaper |
| IP transit | $8,400 | $7,200 | 0.86× — India more expensive |
| Total | ~$104,500 | ~$135,500 | 1.30× |
The headline rent ratio is 2.4×. The all-in ratio is 1.3×. The gap closes for three specific reasons, and every one of them cuts against the common assumption.
Indian grid electricity costs more per kilowatt-hour than Virginia's. Roughly $0.105 against $0.094 — about 12% dearer on the commodity itself.
Indian PUE is worse — around 1.6 against 1.4 — so the facility burns about 14% more power per IT-kilowatt. Compounded with the tariff, India's power bill per IT-kilowatt runs roughly 28% higher.
Indian IP transit costs more, because there are fewer competing transit providers and fewer cable landings than at a major American interconnection hub.
India is cheap on land, construction and labour — not on electrons or bits. Development cost in India is genuinely the lowest in the world at roughly $5.5–8.5 million per megawatt, and the operating side gives a good deal of that back. Anyone who says ‘India is cheap because power is cheap’ has it exactly backwards, and saying so with the arithmetic is a very strong signal in an infrastructure interview.
Bandwidth, and the deflation nobody prices in
Wholesale IP transit is the clearest deflation story in infrastructure. At the most competitive hubs, a 100 GigE port now prices around $0.05 per megabit per second per month, with 100 GigE prices having fallen at roughly 12% compounded annually over the preceding three years. Prices in the established hubs are now bottoming out rather than continuing to fall.
Two consequences worth carrying:
- Bandwidth is no longer the expensive part of infrastructure. Power is. If you are reasoning about a facility's cost and you start with bandwidth, you are starting in the wrong place by roughly an order of magnitude.
- The published medians are large-commit numbers. A 1 Gbps commitment prices several times the 100 GigE median. Price per megabit is a function of commit size, and quoting a median at retail scale will make your estimate badly wrong.
Internet exchanges, where the arithmetic is published
Exchange port pricing is one of the few genuinely public price lists in this whole subject. A large European exchange currently prices a 10 GigE port at around €650 a month and a 100 GigE port at around €3,240 — five times the price for ten times the capacity, so €0.065 per megabit against €0.032.
That per-bit halving at each port step is the core economic fact about exchanges, and it explains why content networks build out to them aggressively: you pay for the port, not for the traffic, so the marginal cost of an additional gigabit is zero.
India's exchange layer is worth a specific note because it is genuinely contested rather than monopolised — the competition regulator examined a complaint against the national exchange in 2024 and found it was not dominant. One historical quirk worth knowing: the national exchange has operated a settlement model in which the net receiver of traffic pays per gigabyte on the imbalance, which is unusual internationally and is a useful contrast when explaining what settlement-free peering means.
The honest MPLS versus SD-WAN number
Vendor material claims WAN savings of 50 to 84% from replacing MPLS with SD-WAN. Independent procurement data on a like-for-like 1 Gbps circuit — MPLS against dedicated internet access plus a managed SD-WAN service — puts the saving at around 25%.
The gap between 25% and 84% is not dishonesty exactly; it is the difference between replacing a circuit and replacing a bandwidth tier. The vendor comparison swaps a 50 Mbps MPLS circuit for a 500 Mbps broadband one and counts the whole difference as a saving. Asking which comparison a number represents is the right response to any WAN business case, and it is exactly the kind of question that marks out someone who has bought this before.
Check yourself