NTT Bengaluru 3 carries 12 MW of critical IT load and 20 MVA of total facility load across 5,800 square metres of data centre space. Twenty megavolt-amperes at a 0.9 power factor is 18 MW, which puts the ratio at 1.50 — exactly the house standard, and the second building in this registry to confirm it. Read naively, MVA against MW, it looks like 1.67. The floor figure is where this page earns its place. Twelve megawatts in 5,800 square metres is 5,203 square feet per megawatt. The constant published on record PRJ-164, built from four other NTT buildings, was 13,300. Bengaluru 3 sits 61% below it, and Bengaluru 4 sits 80% below it. Six NTT buildings, one metric, a five-fold spread. That constant has been withdrawn, and this page says why.
1At a glance
2The constant that broke
Record PRJ-164 published a claim worth repeating precisely, because it is now being withdrawn: that four NTT buildings between 2011 and 2024 all land within 4% of 13,300 square feet of floor per megawatt of critical IT load.
| Building | Floor figure | sq ft per MW | As published |
|---|---|---|---|
| Bengaluru 4 | 17,000 m² | 2,722 | data centre space, whole campus |
| Bengaluru 3 | 5,800 m² | 5,203 | data centre server space |
| Chennai 1 | 1,500 m² | 8,073 | data centre server space |
| Noida 1 | about 1,800 m² | 12,917 | data centre server space |
| NAV1A | 400,000 sq ft | 13,333 | building area |
| Vikhroli | 3,000 m² | 13,455 | server space |
Look at the last column. The two buildings at the extremes are described with the same phrase — data centre space — and differ by a factor of nearly two. Vikhroli and Bengaluru 3 both report "server space" and land at 13,455 and 5,203.
So the constant was not a design standard. It was an artefact of four buildings whose published definitions happened to coincide. With six data points the spread is five-fold and the metric is unusable as a predictor.
Record PRJ-164 states that NTT’s buildings share a 13,300 square foot per megawatt constant and uses it to estimate an unbuilt building’s floor area.
That estimate is withdrawn. The underlying figures on PRJ-164 are unchanged and still correct; the generalisation drawn from them is not supported once Bengaluru is included.
This is what the registry is for. A pattern that survives four records can fail on the fifth, and the honest response is to say so on the next page rather than quietly stop using it.
320 MVA against 12 MW
The ratio question again, and this building is a clean illustration of why the unit matters more than the number.
| Assumption | Facility load | Ratio to 12 MW | Comment |
|---|---|---|---|
| Naive, MVA read as MW | 20.0 MW | 1.67 | what most comparisons do |
| Power factor 0.9 | 18.0 MW | 1.50 | the working figure on this page |
| Power factor 0.85 | 17.0 MW | 1.42 | typical for older plant |
| Per-feed reading, dual feed | about 36 MW | 3.00 | implausible for 5,800 m² |
At 0.9 the answer is 1.50, which makes Bengaluru 3 the second independent confirmation of NTT’s house ratio after NAV1A. That is more useful than the building itself: two buildings, two cities, two different published units, same ratio.
It also means the 1.67 on the previous record — Bengaluru 2 — is the outlier rather than the pattern. Bengaluru 2 publishes 4.5 and 7.5 as megawatts, so no power factor is needed and its 1.67 stands. Bengaluru 3 only looks like it.
Read together, the two Bengaluru records bracket the floor on overhead: somewhere between 4.5 MW and 12 MW, a hall stops paying a disproportionate share of its own plant.
4What 5,203 square feet per megawatt means
If the floor figures cannot be compared across buildings, what can they say on their own? For Bengaluru 3, quite a lot.
| Measure | Value | Basis |
|---|---|---|
| Critical IT load | 12 MW | operator-stated |
| Data centre space | 5,800 m² / 62,430 sq ft | operator-stated |
| Power per square metre of data space | 2.07 kW/m² | 12,000 divided by 5,800 |
| Facility load per square metre | 3.10 kW/m² | 18 MW over 5,800 m² |
| Compare NAV1A | 0.81 kW/m² | on a building-area basis |
| Compare Chennai 1 | 1.33 kW/m² | on a server-space basis |
Two point zero seven kilowatts per square metre of data hall is a dense floor. It is consistent with a hall built for higher-density racks than Chennai 1’s 3 to 6 kVA, and it sits well above NAV1A on the same measure.
The registry’s position on all floor figures, stated plainly: they are only usable within a record, not across records, because "data centre space", "server space" and "building area" are used interchangeably by operators and mean different things. Any market report that divides megawatts by square feet across operators is producing a number with no meaning.
5Procurement
6Limits and open questions
- 20 MVA is apparent power. The 1.50 ratio depends on a 0.9 power factor. At 0.85 it is 1.42; read naively it is 1.67.
- The floor metric does not compare across buildings and is used within this record only.
- The 13,300 constant on PRJ-164 is withdrawn. That record’s underlying figures remain correct; the generalisation does not survive Bengaluru.
- No rack count, PUE, occupancy or tenant list.
- No dated commissioning evidence for this facility in the reviewed set.
- What would close it. A rack count and a per-cabinet kW figure. Bengaluru 3 would then be one of the most measurable buildings in India.
7Sources
Sources: NTT Global Data Centers Bengaluru 3 facility page, NTT DATA global data center locations index, Data Center Dynamics reporting on the Bengaluru portfolio and the 5,800 sq m figure, Datacentermap listing for GDC Bengaluru 3, and NTT’s 2022 investor material which lists Bangalore 3 at 12 MW.