Expert guide
Compute Exchange for data centers in India: what procurement teams need to know
Scarce and stranded at the same time
Compute capacity in India is both scarce and stranded. GPU clusters are allocated months ahead and then sit idle between projects; colocation operators carry racks and power that are committed but unused; enterprises hold reserved capacity they no longer need; developers have shells energised ahead of tenants. Demand arrives from AI start-ups and research groups that need accelerator time by the week or month, from enterprises that need burst or project capacity, and from operators that want to fill or offload wholesale blocks. There is no transparent market to match them and pricing is opaque.
MyDCNeeds' Compute Exchange is a private intermediary for that market. Capacity and requirements are listed confidentially, counterparties are verified, offers are normalised to comparable units, and introductions happen under NDA. Buyer and seller contract directly; we handle matching, due diligence and commercial support.
What trades
GPU clusters are leased, reserved or sold by accelerator class, cluster size, interconnect and term, from short project bursts to multi-year reservations. Bare-metal and dedicated servers trade on configuration and term. Colocation trades by the rack, cage or suite and increasingly by the kW, with power either included or passed through. Wholesale capacity of 1 MW and above trades as powered shell or fitted suites on long terms. A growing segment is the resale or sublease of committed capacity under existing contracts, subject to the master agreement's transfer rights.
How pricing is normalised
Offers in this market are rarely comparable as quoted: a per-rack price with 5 kW included is not a per-rack price with 10 kW and metered power, and a GPU-hour with storage and interconnect is not a GPU-hour without. We restate every offer in standard units: INR per kW per month including power and the stated redundancy for colocation, INR per MW per year for wholesale, and per GPU-hour or per month by accelerator class with the interconnect, storage and support terms specified for compute. Indicatively, colocation in the major metros trades at INR 7,000 to 12,000 per kW per month including power, and wholesale at INR 9 to 14 crore per MW per year; GPU pricing moves weekly with availability.
Verification and diligence
Facilities are verified on Tier or rating, certifications (ISO 27001, SOC 2, MeitY where relevant), connectivity and location. GPU and bare-metal capacity is verified by inspection or a remote test of the actual cluster rather than a specification sheet. For resale of committed capacity we review the master agreement for assignment and sublease rights before any introduction. Buyers are qualified on use case, compliance constraints such as data residency under the DPDP Act and sector rules, and creditworthiness.
Contract and onboarding
Term sheets prepared by MyDCNeeds set out capacity, pricing in normalised units, term and exit provisions, SLAs for availability, support and replacement, and onboarding responsibilities. The parties contract directly, and we support the negotiation, the technical onboarding and any transfer of committed contracts. Short-term and reserved structures with defined exit terms match the project cycles that drive most compute demand.
How to list or enquire
Sellers describe the capacity (type, quantity, location, term and any constraints) and buyers describe the need, both confidentially through the form on this page. Verification and normalisation take a few days; qualified introductions follow under NDA, typically within five to ten business days.
Nothing is published. We verify, normalise, match under NDA and support the contract between the parties.
What we cover
Scope of the service line
- GPU clusters: lease, reserve or buy
- Bare-metal and dedicated servers
- Colocation: racks, cages and suites by the kW
- Wholesale capacity: 1 MW and above
- Short-term burst and project capacity
- Capacity resale and sublease of committed contracts
How procurement works
From brief to contract
Listing or requirement
Sellers describe capacity (GPU type and count, kW, racks or MW, location, term) and buyers describe need, both confidentially.
Verification
Operators and facilities are verified; GPU and bare-metal capacity is confirmed by inspection or remote test.
Normalisation
Every offer is restated in comparable units: INR per GPU-hour, per kW per month including power, or per MW per year.
Matching under NDA
Qualified counterparties are introduced with term sheets prepared by MyDCNeeds.
Contract and onboarding
Buyer and seller contract directly; we support SLA, exit and transfer terms and the onboarding.
Advisor note
Price by the kW, not the rack
Colocation deals in India are moving from per-rack to per-kW pricing as densities rise. We normalise every offer to INR per kW per month, including power pass-through, so you compare like with like.
What to evaluate
Criteria that separate a safe award from an expensive one
| Criterion | What good looks like | Why it matters |
|---|---|---|
| Verified capacity | Physical or remote verification completed | No speculative listings |
| Comparable pricing | Normalised to standard units with power and bandwidth stated | Like-for-like comparison |
| Facility standard | Tier level, certifications, connectivity and location | Fit for the workload |
| Term flexibility | Short-term, reserved and committed options with exit terms | Matches project cycles |
| SLA | Availability, support and replacement commitments in the contract | Enforceable performance |
| Confidentiality | NDA before any introduction | Protects both sides' positions |
Benchmarks
Indicative lead times and cost ranges
| Item | Typical range | Note |
|---|---|---|
| Colocation, per kW per month, major metros | INR 7,000 to 12,000 including power | Density and term dependent |
| Wholesale, per MW per year | INR 9 to 14 crore | Long-term, shell and power |
| GPU cluster lease | Quoted per GPU-hour or per month by accelerator class | Availability changes weekly |
| Time to first introduction | 5 to 10 business days | After verification |
Indicative figures from MyDCNeeds engagements and supplier quotations, 2025 to 2026. Project-specific numbers are confirmed during scoping.
Common mistakes
What we see go wrong
- 01Comparing per-rack prices across facilities with different power allocations.
- 02Leasing GPU capacity without a tested interconnect and storage path.
- 03Reselling committed capacity without checking the master agreement's transfer rights.
- 04Publishing spare capacity publicly and anchoring the market against yourself.
Standards and regulation
The framework this work sits inside
- Uptime Tier and TIA-942 ratings for facilities
- ISO 27001 and SOC 2 for operators
- MeitY empanelment for government workloads
- DPDP Act 2023 data residency considerations
Glossary
Terms used in this guide
- Bare metal
- Dedicated physical servers leased without a hypervisor layer.
- Wholesale colocation
- Large blocks of capacity, usually 1 MW and above, leased as powered shell or fitted suites.
- Reserved capacity
- Capacity committed for a term at a fixed price, as opposed to on-demand.
- GPU-hour
- The standard pricing unit for accelerator compute.
Questions
Frequently asked
Is my capacity listing public?+–
No. Capacity, pricing and identity are shared only with qualified counterparties under NDA after verification. Nothing is published on the website or elsewhere.
What does colocation cost in India?+–
Indicatively INR 7,000 to 12,000 per kW per month including power in the major metros, depending on density, redundancy and term. Wholesale capacity trades at roughly INR 9 to 14 crore per MW per year. Offers should be compared on normalised units, not per rack.
What GPU capacity is available?+–
Availability changes weekly. Tell us the accelerator class, cluster size, interconnect and term and we respond with current verified options and normalised pricing.
Can we resell capacity we have already committed to?+–
Often, subject to the assignment and sublease provisions of your master agreement. We review the contract before any introduction and structure the transfer accordingly.
Who signs the contract?+–
Buyer and seller contract directly. MyDCNeeds prepares the term sheet, supports negotiation and onboarding and is not a party to the agreement.
How quickly can a match be made?+–
Verification and normalisation take a few days; qualified introductions under NDA typically follow within five to ten business days.
