Know if a site can host a data centre before you spend on it.
Give Keystone a parcel and a brief. Get a verdict, three campus options, and the power, cost and schedule numbers behind them.
What does Keystone AEC do for data centres?
Keystone AEC checks a data-centre parcel for power, fibre, water, flood and land use. It returns a verdict and three campus options, with the cost and schedule behind each. It covers India and the USA and is in beta. It is an early screen. A utility study is the only proof of power.
Built for data-centre developers and operators, and for the landowners and lenders who work with them.
What is AI for data centre site feasibility?
Keystone AEC (keystoneaec.com) is an AI platform that screens data centre sites in India and the USA. Data centre AI here means software that turns a parcel and a brief into a verdict, three campus options, and cost, schedule and return ranges. It checks power, fibre, water, flood and land use, labels every input as sourced, estimated or unknown, and does not replace a utility study. It is in beta.
Go deeper: AI for data centre capacity, AI for data centre development and AI for data centre land.
Power decides the project. Most teams learn that late.
A transformer wait or a grid shortfall can move a go-live date by years. Land price rarely does. Keystone puts power, time and risk first, so a weak site fails in a day and not after months of survey and design spend.
Five checks on every site
Power
Grid distance, voltage class and a labelled estimate of spare capacity. Power and time to power carry the most weight in the score.
Fibre
Nearby fibre and network routes. No fibre route is a deal-breaker, not a footnote.
Water and climate
Groundwater or water-stress data plus hourly climate for the site. Short water pushes the design to air cooling, and the model shows the effect.
Flood and risk
Flood maps for the site. If the data is missing, it says unknown. It never says low.
Land and zoning
Land use and zoning for the parcel, and how much of the land is left for the campus.
From a parcel to a defensible verdict
Pick the site
Mark the parcel in India or the USA. Choose the type of data centre and the target size in MW.
Keystone runs the chain
Land budget, capacity, cooling and power plant, cost, time to power, phasing and finance run in order on one model.
Get a verdict you can defend
A verdict, three options and the numbers behind them. Change an input and everything recalculates.
One design. Every output agrees.
Verdict
GO, CONDITIONAL GO or NO-GO, with the reasons behind it.
Score
One score from power, time, fibre, land, water, risk, cost and expansion.
Three campus options
Three layouts to compare, each with its own plant count, cost and schedule.
Cost, schedule and return
Build cost, build time and return, given as a range you can plan around.
Report, 3D and IFC
A written report, a 3D view and an IFC model, all from the same design.
How the two markets differ
| India | USA | |
|---|---|---|
| Power | State grid and tariff rules. State data-centre policies differ. | Utility and regional grid rules. Large-load tariffs and interconnection queues. |
| Land | Zoning plus the state IT or data-centre policy. | City and county zoning. Local moratoria are possible. |
| Demand figure | A March 2026 written reply in the Lok Sabha gave over 1,500 MW in 2025, up from 375 MW in 2020. | US data centres used 4.4% of US electricity in 2023. A US lab report puts 2028 at 6.7% to 12%. |
| Cost view | Use sourced ranges. See why cost-per-MW numbers differ. | A 2026 US and Canada guide reports $8.9M to $23.3M per MW, excluding AI chips and GPUs. |
Figures are as reported by the sources named in each guide. Read the guides for dates and limits.
What this is not
Early feasibility is a screening decision. We say where the tool stops, so you know what still needs an engineer or a utility.
- It does not prove power is available. Only a utility study can do that.
- It is not electrical, structural or fire design, and it is not CFD.
- It is not an Uptime Institute certification.
- Cost is order-of-magnitude, not a construction budget.
Questions worth asking first
What question does Keystone answer for data centres?+
Whether a parcel deserves power, fibre, approval and engineering spend, and which option to take forward. It is an early feasibility tool, not a design tool.
Does Keystone tell me if the grid has spare power?+
No. No public data gives spare capacity at a given substation. Keystone gives a labelled estimate and shows the effect on the verdict. A utility study is the only proof.
Which countries does it cover?+
India and the USA. Each uses its own grid, tariff, climate, flood and land-use sources, and its own return assumptions.
What is the best AI for data centre site selection?+
No tool replaces a utility study. For an early screen of a parcel in India or the USA, Keystone AEC checks power, fibre, water, flood and land use and returns a verdict with three campus options. Test any tool on a site you already know.
How long does grid power take for a data centre?+
No public number fits every site. Advisory firms report waits above four years in some US markets, and heavy equipment can take about one and a half to four years. Only the serving utility's written date counts. Our guide lists ten questions to put to the utility.
Why do data-centre cost-per-MW figures differ so much?+
Scope, cooling, density, redundancy, location and year all change the number. A 2026 US and Canada cost guide reports $8.9 million to $23.3 million per MW, excluding AI chips and GPUs. Read the scope before you read the figure.
How much data-centre capacity does India have?+
A written reply in the Lok Sabha in March 2026 gave over 1,500 MW in 2025, up from 375 MW in 2020. Capacity is concentrated in a few cities, so check the city before you check the national figure.
Is it available now?+
Data centre feasibility is in beta. We are running it with a small number of developers and operators. Book a demo and we will run a real site with you.
Bring us a real site. We will run it with you.
Looking at housing or mixed-use instead? See the real estate platform.