Power Is the New Pricing Lever

India runs about 1.3 to 1.53 GW of operational data-center IT load today.

By end of 2026, the industry expects to be at 1.7 to 2.0 GW.

By 2030, 4 to 5 GW, backed by close to $30 billion in cumulative investment.

The capacity story is the easy headline. The harder story is underneath.

Power demand is moving from about 1.5 GW in 2025 to somewhere between 8 and 13.5 GW by 2030 to 2032. That is 2 to 8 percent of India's electricity grid going to data centers within seven years.

1.5 GW becomes 13 GW.

For the BFSI, manufacturing, pharma, and textile CIO sitting on a colo contract that renews in 2026 or 2027, this is not a market-research statistic. It is the cost curve you are about to inherit.

Capacity is not the constraint anymore.

Power is.

But first, some catch-up on infra this week.

The math of Indian data-center build-out for the next five years splits cleanly into two numbers that move at different speeds.

Capacity grew about 260 to 387 MW in 2025 alone. The 2026 trajectory adds another 30 percent of supply on top of that. That part is solvable. Investors are queuing up. Land is being acquired. Cement is pouring.

Power is not solvable on the same calendar.

Power is the binding constraint.

A new 100 MW data center campus needs roughly 100 MW of firm grid power, transformer capacity, water for cooling, and a redundant fuel supply or BESS backup. Each of those moves on a different regulator's timeline. The shortest path from approved-plan to commissioned-load is rarely under 18 months. The longest paths are still being measured in years.

Three Indian hubs are absorbing most of the demand.

Chennai is the subsea-cable proximity hub. Five international landing stations within a 300-km arc, plus the eastern seaboard's better power reliability. The hyperscaler-and-AI buildout has been disproportionately Chennai-weighted for the last two years.

Delhi-NCR is the government, public-sector, and large-enterprise hub. The customer concentration is broad, but the grid is the operating risk.

Mumbai is the BFSI hub. Same-rack latency to the major exchanges and clearing houses is the value the Mumbai-Navi Mumbai DC corridor sells. If your trading-and-settlement workloads run in seconds, you have a Mumbai problem whether you like the power profile or not.

Per BFSI end-user analysis, BFSI accounts for 32.8 percent of Indian data-center end-user demand. The single largest vertical, and the one with the tightest non-negotiables on uptime and latency.

For the Indian CIO planning a 2026-2028 colo refresh, the practical implication is uncomfortable.

You are not negotiating against last year's price-per-rack-U.

You are negotiating against a 2030 cost curve that has not been written down yet.

What to ask, before signing.

✔ What is the contracted price-per-kW today, and what is the escalator clause for the next renewal cycle?

✔ What is the provider's committed PUE for the years 2027 and 2028 of the contract term? PUE drift directly transfers cost to you.

✔ What is the firm-power commitment versus best-effort? Especially for high-density AI inference racks.

✔ What is the BESS (battery energy storage) or generator runtime, in hours, at full IT load?

✔ Which substation feeds the campus, and what is its forward-looking load projection from the state DISCOM?

Capacity is what your provider sells you. Power is what they pay for.

The contract you sign sets which one of those two cost curves you ride.

The fastest-growing slice of Indian DC demand is Tier-4-or-equivalent capacity, driven by BFSI and healthcare buyers demanding 99.995 percent uptime.

99.995% uptime is contractual.

Let us name what that number actually means in operational language.

99.995 percent over a year is roughly 26 minutes of allowed downtime. For a single facility hitting that mark, you need concurrently-maintainable power and cooling. Not just redundant. Concurrently-maintainable. Any single element can be taken out of service for maintenance without dropping the IT load. Two independent utility feeds is the table-stakes spec. Two independent cooling paths. Independent fuel supply. UPS architecture that survives the loss of any one string.

The Tier-4 certification (or the operator's own equivalent) is what lets the procurement contract carry the 99.995 percent uptime line in the SLA.

Below that, the Tier-3 facility offers 99.982 percent (about 95 minutes per year of allowed downtime). For a non-BFSI workload, often acceptable. For a real-time clearing-and-settlement workload, often not.

This is where the Indian DC buyer's procurement framework gets misread.

The conversation usually starts with "what is the price per kW?" The right opening question is different. It is closer to: "what is the contractual uptime, expressed as minutes-of-downtime-per-year, and what is the credit clause when it is breached?"

A facility that offers 99.99 percent in a brochure but writes 99.95 percent in the contract has just told you what the SLA actually is. The brochure is marketing. The contract clause is the operational reality your audit committee will see.

For the BFSI CIO with a clearing-system requirement, the questions are sharper.

👉 What is the actual measured (not designed) uptime over the last 24 months at this specific facility?

👉 Which incidents triggered the SLA-credit clause, and what was the credit?

👉 What is the largest single point of failure that has been remediated since the facility opened?

👉 If a transformer fails at 14:00 IST on a settlement day, what is the documented switchover time?

Tier-4 is also where the Indian observability gap shows up sharpest. The DC operator's NOC reports facility-level uptime. The customer's NOC reports application-level uptime. The 26-minute facility-budget can be entirely consumed by an event that the customer's monitoring did not see, simply because the facility kept running but a specific power feed to a specific rack-row did not. That is the gap where most "99.995 percent SLA, but my application was down for 40 minutes" disputes happen.

The other half of the Indian DC story this year is policy.

Budget 2026 extended tax exemptions for data-center services to 2047, mandated local incorporation structures for foreign DC investors, introduced a 15 percent safe harbour rule, allocated ₹10,000 crore for container manufacturing, and granted duty exemption for BESS (battery energy storage systems) manufacturing equipment.

Read those provisions together. The signal is not "we like data centers." The signal is "we are pricing data-center investment as a national-infrastructure asset class with a 20-year horizon."

The tax-exemption extension to 2047 takes the policy uncertainty out of the next two procurement cycles. Any DC operator running a 7-year contract with a renewal in 2033 now has policy visibility through that renewal.

The 15 percent safe harbour matters for foreign investors planning Indian DC builds. It is the line below which the transfer-pricing argument with the tax authority does not happen. Lower friction, lower risk premium, lower capex.

The BESS support is the quiet one but the strategically loudest. India's DC sector currently leans heavily on diesel generators for grid backup. BESS lets that flip to lithium-iron-phosphate or sodium-ion arrays as the firm-power backstop. The duty exemption on BESS manufacturing equipment is the supply-side push that makes the operator's economics work.

For the Indian CIO, two procurement implications.

✔ The cost-of-capital that your DC provider is pricing into the contract just dropped (because policy risk just dropped). If the renewal quote does not reflect that, ask why.

✔ The diesel-generator runtime line in your SLA may stop being the right backup-power metric over the next contract cycle. The right question becomes: "what is the BESS-plus-grid hybrid runtime, in hours, at design IT load?" That is the new spec for any 2027-2028 commissioning.

The DC sector got policy visibility through the 2030s.

The CIO's procurement decision got correspondingly longer-horizon.

The decision in front of a BFSI or manufacturing CIO in India for the next 18 months is rarely framed correctly in the vendor pitch.

The pitch frames it as colo vs cloud.

The actual decision is closer to a three-way split.

Option one: pure colo. You own the racks, the OS, the patching. The DC provider owns the power, cooling, physical security, and the facility-level uptime. Best for BFSI workloads with hard latency and uptime requirements, regulated data, and a customer audit footprint that lives well in a Tier-4 facility.

Option two: hybrid. Some workloads in colo, some in a hyperscaler India region, some in a sovereign Indian cloud. The complexity moves into the network underlay and into the operations team. Most Indian enterprises are landing here, often by accident rather than design.

Option three: sovereign or India-resident cloud. The IBM-Yotta Shakti Cloud announcement covered in last week's issue is one example. Yotta on its own, CtrlS Cloud, NxtGen, AWS / Azure / GCP Indian regions running with explicit data-residency. The DPDP Act compliance is easier here. The cost-per-CPU-hour is sometimes higher.

For most BFSI, manufacturing, pharma, and textile buyers, the right answer is some form of hybrid. The trick is being deliberate about which workload lands where, and why.

A workable framework for the procurement decision.

1. Inventory workloads by three axes: regulated-data sensitivity, latency requirement, predictability of resource demand.

2. Highly regulated, low-latency, predictable → colo, with the BFSI Tier-4 spec from section two above.

3. Highly regulated, latency-tolerant, predictable → sovereign or India-resident cloud, with the data-residency line written into the contract.

4. Less regulated, variable demand → hyperscaler India region with the standard governance harness.

5. Anything experimental → hyperscaler with auto-scale and a hard cost cap. Move it to colo or sovereign only after it has a stable demand profile.

The mistake most Indian CIOs make is doing this workload-assignment decision once, at the start of a DC contract, and not revisiting it. The right cadence is annual, ideally aligned with the budget cycle.

The mistake the vendor side encourages is treating the decision as binary. Either "all in on colo with us" or "all in on cloud with us." Neither serves the Indian buyer's actual mix.

India to add 2-3 GW data centre capacity in 5-7 years, market seen at $22 billion by 2030
ANI / Industry analyst forecast, 2026-04.
The capacity forecast from a slightly more conservative analyst view than the 4-5 GW number used above. Worth reading the assumptions before quoting either number in a procurement memo.

Wi-Fi Alliance brings Wi-Fi 7 features to 20 MHz-only devices
RCR Wireless, 2026-01.
Useful operational note for the Indian campus refresh buyer with installed-base of 20 MHz endpoints. Wi-Fi 7's backward-compatibility story matters more than the marketing usually admits.

Sterlite Technologies invests $100M in US AI connectivity
Voice & Data India, 2026-05.
Indian connectivity infrastructure money going outbound into the US AI build-out. The supply-side mirror of this issue's Indian-DC story.

Wireless Broadband Alliance: Road to Wi-Fi 7 reference architecture
WBA, ongoing.
Multi-vendor reference deployments across manufacturing, transport, stadium, and campus segments. If your 2027 Wi-Fi refresh is in evaluation, this is the multi-vendor view your vendor's deck will not show.

India DC capacity to reach 1.7 GW by end of 2026, driven by government support
Angel One, 2026.
The capacity-side primary source for this issue. Read the breakdown by state and the comment on power as the binding constraint.

Every infrastructure procurement decision is a bet on which cost curve you ride for the next seven years.

For most Indian enterprises, the bet has historically been about price-per-rack-U or price-per-CPU-hour. Those are the units the vendor's commercial team is trained to defend.

The cost curve that actually compounds across a seven-year contract is power.

Capacity is solvable on a 12-18 month timeline. Investors will fund it. Operators will build it.

Power is not solvable on the same timeline. A new transformer is 24 months of utility process. A new substation is 4 to 6 years. A reliable BESS-plus-grid hybrid is the difference between a Tier-4 SLA you can defend and one your auditor will challenge.

The Indian CIO who reads the Budget 2026 tax provisions as just "good for the DC sector" is reading them at the wrong altitude. They are also "good for your renewal negotiating posture for the next 20 years," because they reduce the operator's cost-of-capital and the operator should be passing some of that through. If your renewal quote is not reflecting it, ask why.

Buy power. Lease compute.

The seven-year procurement frame inverts. The thing you want to lock in long is the power contract, the substation feed, the BESS-plus-grid runtime, the firm-power commitment. The thing you want to keep flexible is the compute layer, the workload mix between colo, hybrid, and sovereign, the rack density of any given row, the choice of CPU generation.

VEMIO™ exists because the operational reality of that inversion needs an observability plane underneath it. Not just "is the rack up." Closer to "is the rack on the feed I am contractually paying for, is the PUE matching the SLA, and what is the rolling 24-month measured-versus-designed uptime trend." Indian BFSI auditors are starting to ask exactly that question. The CIO who has the data ready answers in 20 minutes. The CIO who does not, answers six weeks late, after the third vendor escalation.

The procurement decision is not pricing. It is posture.

Reply to this email with the one DC or colo procurement decision you are revisiting this quarter, and we will feature the most operationally interesting reply (anonymised, with consent) next issue.

Until next time,

Ajay Salvi & the Vinay Enterprises team.