In July 2026 Tata Communications committed US$152 million to add 98 Tbps of subsea capacity between India and Singapore, across the MIST and Project CS networks, explicitly to link its Mumbai and Chennai AI hubs with Singapore. Ninety-eight terabits per second is throughput. A megawatt is power. One describes how fast data moves between two buildings; the other describes how much electricity is consumed inside one of them. No conversion exists between them, and any report that adds subsea capacity to data centre capacity is wrong. This record carries zero megawatts, and the reason it still belongs in a 242-record registry is that the cable, not the concrete, determines which Indian cities can sell AI capacity to the world.
1At a glance
2What was announced
| Item | Published | Missing |
|---|---|---|
| Investment | US$152 million | Currency |
| Capacity added | 98 Tbps | Terabits per second |
| Networks | MIST and Project CS | Named cable systems |
| Route | India to Singapore | Geography |
| Indian endpoints | Mumbai and Chennai AI hubs | Cities |
| Funding source | Internal accruals | Balance sheet |
| Megawatts | — | Not a valid unit for this record |
US$152 million for 98 Tbps is about US$1.55 million per terabit per second. That is the only ratio the announcement supports, and it is a useful one for anyone pricing subsea expansion — it is simply not a data centre metric.
For scale, 98 Tbps is roughly the equivalent of carrying every second of a high-definition video call for about four million people simultaneously. It is a very large amount of bandwidth between two countries and it consumes no power at either end beyond the optical termination equipment.
3Why throughput and power are not interchangeable
A single 3.2 MW data hall at NTT’s Bengaluru 4 density holds roughly 25 liquid-cooled racks at 130 kW each. Those 25 racks can saturate multiple terabits of long-haul capacity, and the buildings around them draw megawatts whether the cable exists or not.
Equally, 98 Tbps could sit entirely idle and draw almost nothing, or be filled by a single customer whose compute consumes hundreds of megawatts somewhere else.
There is no exchange rate. Any figure that converts Tbps to MW, or adds one to the other, is fabricated.
| Unit | What it measures | Can be summed with | Records |
|---|---|---|---|
| MW | Electrical capacity, IT or facility | Other MW on the same basis | 129 counted records |
| Tbps | Network throughput | Nothing in this registry | PRJ-052, PRJ-236 |
| sq ft | Floor area | Other floor area, with a stated density | PRJ-235, PRJ-237, PRJ-240 |
| ₹ crore | Investment | Other investment, never capacity | PRJ-235, PRJ-238, PRJ-239 |
This table is the whole methodology of the registry in four lines. Most published Indian data centre figures mix at least two of these units inside one sentence. The reconciliation of 26 October 2026 exists because this registry did the same thing with 2,420.37 MW of unlike records.
4The connection that does matter: cables decide cities
The record is worth holding for a reason that has nothing to do with megawatts. Subsea landing points determine which Indian cities can serve international AI workloads, and therefore which cities can charge a premium for their capacity.
| Facility or city | Cable position | Why it matters |
|---|---|---|
| Sify Chennai 02, Siruseri | Only Indian data centre with an on-site open cable landing station | The reason its cost is ₹72.6 crore per MW, the highest in the file except two small Kolkata sites |
| Sify Visakhapatnam | Open cable landing station alongside a 50 MW edge facility | The stated regional gateway to Singapore, Malaysia, Thailand and Australia routes |
| Mumbai | India’s primary landing cluster; named by this record as an AI hub endpoint | Where most international capacity already terminates |
| Chennai | Second landing cluster; named by this record as an AI hub endpoint | The corridor Tata Communications is reinforcing with 98 Tbps |
| Bengaluru, Hyderabad, Noida | Landlocked; dependent on terrestrial links to a landing station | The reason NTT’s Devanahalli campus competes on cost rather than connectivity |
The clearest evidence for this in the registry is a cost anomaly. Sify’s Siruseri campus costs ₹72.6 crore per MW against an India band of ₹51.4 to 60.1 crore. Part of that premium is the cable landing station, which no other data centre in the country has on site.
Tata Communications is spending US$152 million to make the Chennai and Mumbai corridors faster. It is not building a data centre, and it is improving the economics of everyone who is.
5What would change this record
| Item | Why it would matter |
|---|---|
| A named landing station or cable landing party | Links the announcement to a physical Indian site, and potentially to a data centre campus |
| A capex split by landing city | Allows Mumbai and Chennai to be credited separately |
| Any associated data centre fit-out | Occasionally a cable project brings a colocation or interconnect build with a stated capacity |
| A capacity figure for the interconnect halls | Interconnect space is measured in megawatts and would be countable, unlike the cable |
None of those is likely. Subsea investments are announced in dollars and terabits because that is what they are. This record will stay at zero megawatts, and its value to a buyer is as a labelled non-capacity fact — the thing that stops a competitor improving a market total by adding a cable to it.
6Sources
Sources: Tata Communications announcement, July 2026, of US$152 million to add 98 Tbps of subsea capacity between India and Singapore across the MIST and Project CS networks, funded from internal accruals and linking Mumbai and Chennai AI hubs; Gridvik record PRJ-052 for the comparable CtrlS-Ciena optical network treatment; Gridvik records PRJ-224 for the Sify Siruseri on-site cable landing station and its cost per megawatt, and PRJ-231 for the Visakhapatnam landing station. No capacity figure is applicable to this record. Record verified 26 October 2026.