Solved developer example · Part 5 of 6
Data-centre power feasibility tutorial: 150 MW at Dholera with and without BESS
A complete India data-centre power study using Gujarat policy, Atlas Data Shop evidence, a Dholera nodal scenario and a controlled 30 MW / 4 h battery comparison.
- For
- Data-centre developers, utilities, investors and energy procurement teams
- Time
- 90 minutes plus paired native-hourly solves
- Updated
- 20 Jul 2026
Direct answer
A credible data-centre power screen separates policy, observed market and grid data, derived energy arithmetic and modeled nodal impacts. For Dholera, test an untouched base, a 150 MW flat load, and the same load with a 30 MW / 4 h co-located BESS over one identical horizon.
Solved result first
What the completed analysis must expose
The solved question is not simply whether Gujarat has enough annual energy. It is whether 150 MW of constant grid draw at Dholera is deliverable over the modeled week, at what incremental operating cost and marginal carbon, with which local corridor effects, and whether a 30 MW / 4 h battery changes the result. The benchmark below uses recorded native-hourly results: storage improved reliability and cost but increased marginal emissions and local loading. That trade-off is the point of the analysis.
| Decision metric | 150 MW no BESS | 150 MW + 30 MW / 4 h | Interpretation |
|---|---|---|---|
| Facility energy | 25.20 GWh / modeled week | 25.20 GWh plus storage losses | 150 MW × 168 h; do not annualize the nodal deltas |
| Annual planning volume | 1.314 TWh | Same facility energy basis | 51% annual green arithmetic = 670.14 GWh |
| Unserved energy | 3.58 GWh | 2.94 GWh | Battery lowered the recorded shortfall by about 18% |
| Incremental operating cost | ≈₹2,148/MWh | ≈3.6% lower | Operational modeled-week comparison; not delivered tariff or NPV |
| Marginal emissions | 563 gCO₂/kWh | 607 gCO₂/kWh | Battery case was more carbon-intensive at the margin |
| Dholera corridor peak | 22.5% | 24.4% | Local loading increased but did not bind in the recorded case |
Analytical method
- 01Translate IT load, PUE, redundancy and load factor into an explicit grid-draw series; preserve whether 150 MW is IT or facility demand.
- 02Build the power-to-compute cost stack from energy, balancing, green supply, network, flexibility, carbon, policy and project-capital ledgers.
- 03Validate the named Dholera EHV yard, source-backed corridors, limits and missing utility data before interpreting local results.
- 04Solve the untouched base and 150 MW case with matched native-hourly settings; persist the result pair and full run manifest.
- 05Repeat from the same base with 30 MW / 4 h BESS; compare cost, reliability, carbon, LMP proxy, curtailment and corridor loading.
- 06Benchmark with load-shape, PUE, price/carbon, BESS and network-rating sensitivities; convert the result into a prioritized diligence request.
Who should interrogate what
Electrical engineer
Can the selected node and corridors carry the load under stress?
Interrogate bus selection, thermal ratings, losses, outages, reactive needs, redundancy and the hours of binding constraint.
Environmental engineer
Is annual green matching reducing hourly impact?
Compare marginal emissions, curtailment, storage charging source, water/land constraints and hourly CFE gaps.
Energy economist
What is the complete power-to-compute cost?
Separate bulk energy, balancing, network charges, green premium, storage, reinforcement, taxes and financing.
Policy maker
What infrastructure or market rule becomes the bottleneck?
Use the portfolio and site cases separately; test whether support shifts cost, congestion or emissions.
Energy trader
How much volume is exposed to DAM, RTM and imbalance?
Model procurement shares, ramp/flex windows and spread distributions rather than multiplying annual load by one spot price.
Before you start
- • Enterprise access to the Gujarat state model, Data Shop and Dholera workflow.
- • A fresh Gujarat — Detailed Nodal copy with the named Dholera and Pachham yards present.
- • An untouched base scenario and enough compute credits for paired native-hourly solves.
What you will have
You will produce a policy-and-market evidence screen, a 150 MW no-BESS result, a 150 MW plus 30 MW / 4 h BESS result, and a four-dimension decision table covering cost, reliability, carbon and grid loading.
Step-by-step
Complete the workflow
Fix the facility definition
Treat 150 MW as constant grid draw unless the project evidence says it is IT load, sanctioned demand or another quantity.
Do this
- 1.Set facility grid draw to 150 MW and document 100% load factor for the tutorial arithmetic.
- 2.Set the green-energy share to 51% only as a reported-policy study input.
- 3.Define cost, reliability, carbon and local grid loading as the four pass/fail dimensions.
Expected evidence
- ✓ A facility definition that distinguishes grid draw, IT load, PUE, redundancy and flexible workload.
Worked calculation
Annual facility energy = 150 MW × 8,760 h = 1,314,000 MWh = 1.314 TWh 51% green-energy arithmetic = 1,314,000 × 0.51 = 670,140 MWh
Trust check
Do not convert 150 MW of grid draw into IT capacity without a PUE and auxiliary-load model.
Convert compute demand into a grid-load shape
Separate IT load from facility grid draw using PUE, auxiliary demand, redundancy and workload flexibility before the network solve.
Do this
- 1.Ask whether 150 MW means IT nameplate, metered facility load, sanctioned demand or contracted supply.
- 2.If 150 MW is facility draw, divide by PUE to estimate IT load; if it is IT load, multiply by PUE to estimate facility draw.
- 3.Create hourly baseload, ramp, maintenance and flexible-compute segments rather than assuming every project is perfectly flat.
- 4.Document UPS/BESS boundary, backup generation and whether redundancy assets are normally energized.
Expected evidence
- ✓ One hourly MW series with explicit PUE and load-factor assumptions.
- ✓ A separate flexible block with maximum MW, duration, notice and rebound constraints.
Worked calculation
IT load = facility grid draw / PUE At 150 MW facility draw and PUE 1.30: IT ≈115.4 MW; overhead ≈34.6 MW Annual energy = Σ hourly facility MW × Δt
How to read the result
- • PUE is not constant under all weather and load conditions; test a range rather than one certification value.
- • A flexible workload can shift energy but may create rebound peaks; preserve total compute-service requirements.
Trust check
Do not size interconnection, clean-energy procurement or emissions from IT MW when the grid sees facility MW.
Collect policy and market evidence before solving the network
Use official policy for targets and Data Shop/Atlas evidence for current demand, IEX and carbon windows.
Do this
- 1.Record the official Gujarat policy URL, publication date and exact terms used.
- 2.In the Data Shop, attach the appropriate Gujarat demand, IEX-derived and carbon signals to the study model where compatible.
- 3.Record coverage percentages and retrieval times before quoting averages or percentiles.
Expected evidence
- ✓ A dated source block that separates official policy, observed data and derived calibration series.
Trust check
A recent-market mean is a benchmark, not a procurement quote; a reported policy support is not a guaranteed project entitlement.
Build the eight-ledger Power-to-Compute Cost Stack
Do not collapse unlike risks into a single unsupported power-cost number.
Do this
- 1.Calculate annual volume and recent-market bulk-energy benchmark.
- 2.Add balancing exposure, green procurement, network delivery and flexibility as separate rows.
- 3.Keep carbon, policy support and BESS/interconnection capex in their own ledgers.
- 4.Mark unknown commercial inputs TBD instead of zero.
Expected evidence
- ✓ An evidence range whose observed, derived, modeled, policy and TBD rows are visually distinct.
| Ledger | Minimum evidence | Truth label |
|---|---|---|
| 1 · Compute-to-grid volume | MW × load factor × 8,760 | Derived assumption |
| 2 · Bulk energy | Volume × dated DAM benchmark | Derived from observed |
| 3 · Balancing | Chosen imbalance share × DAM–RTM spread | Derived scenario |
| 4 · Green energy | 51% volume plus sourced premium | Policy + TBD commercial |
| 5 · Network | Base/load solve deltas and charges | Modeled + TBD commercial |
| 6 · Flexibility/BESS | Shift envelope and paired solve | Derived + modeled |
| 7 · Carbon | Recent signal and nodal marginal effect | Derived + modeled |
| 8 · Policy/capex gaps | Support terms, bay, reinforcement, capex, finance | Policy + TBD |
Trust check
Green premium, network charges, losses and financing are not zero merely because public data are missing.
Benchmark energy and balancing exposure
Translate the load shape into dated DAM/RTM exposure without pretending one spot-market statistic is a delivered procurement quote.
Do this
- 1.Choose explicit procurement shares for PPA/RTC, DAM, RTM and imbalance settlement.
- 2.Use mean, median and p95 DAM/RTM spreads from the declared source window.
- 3.Stress 5% and 10% imbalance volumes rather than applying p95 spread to all annual energy.
- 4.Keep open-access, transmission, losses, distribution, taxes and green premium as separate inputs.
Expected evidence
- ✓ A range of energy and balancing cost with source windows and volume shares.
- ✓ No unsupported conversion from a derived 0–1 Data Shop band into ₹/MWh.
Worked calculation
Balancing stress (₹ crore) = annual MWh × imbalance share × selected DAM–RTM spread (₹/MWh) ÷ 10⁷ Run at least 5% and 10% imbalance shares with median and p95 spreads.
| Layer | Volume basis | Price basis | Not included automatically |
|---|---|---|---|
| PPA / RTC | Contracted share | Contract schedule | Shape, losses and deviation |
| DAM | Day-ahead exposed share | Dated observed distribution | Future forecast or guaranteed clearing |
| RTM | Balancing share | Dated observed distribution | DSM and executable liquidity |
| Network | Delivered MWh | Utility/open-access inputs | Reinforcement and bay capex |
| Green | Matched MWh | PPA/certificate premium | 24×7 hourly matching proof |
Trust check
A procurement benchmark must show both volume and price distributions; one annual average hides balancing risk.
Load and verify the Gujarat network
Use a fresh Gujarat — Detailed Nodal copy and verify the named Dholera area before launching the workflow.
Do this
- 1.Open File → State Model Library and load Gujarat — Detailed Nodal into the active project.
- 2.Confirm Dholera (TPL-D) 220 kV and Pachham (Fedra) 400 kV appear by name.
- 3.Read the network fingerprint, source notes and zero-proxy-corridor claim from the current card/result—not from an old screenshot.
Expected evidence
- ✓ The workflow’s intended EHV yard and surrounding corridors are present in the active model.
Trust check
Call it an evidence-backed transmission-planning backbone, not a complete as-built or utility-validated network.
Solve the 150 MW no-BESS case
Run the typed Dholera workflow from the untouched base with facility 150 MW, storage 0 MW and snapshot stride 1.
Do this
- 1.Open Workflows → Dholera data-centre impact.
- 2.Set Facility load 150 MW, Co-located BESS 0 MW and Snapshot stride 1.
- 3.Run and save base result ID, Dholera result ID, selected yard, distance, modeled hours and solver status.
- 4.Record cost, unserved energy, curtailment, marginal emissions, local price and incident-corridor deltas.
Expected evidence
- ✓ A persisted cloned scenario carrying a 150 MW named load and the truth label modeled:nodal-economic-dispatch.
Copy into Modeler Agent
Build and solve a fresh 150 MW nodal Dholera data-centre scenario with no battery and snapshot stride 1. Report the selected EHV yard and distance, modeled horizon, facility energy, cost and emissions deltas, unserved energy, curtailment, Dholera LMP proxy, binding lines and incident corridors. Keep the nodal economic-dispatch truth label visible.
Trust check
This is not CERC SCED, AC power flow, N-1 approval or a utility interconnection decision.
Audit the no-BESS result before adding mitigation
Establish the unmitigated physical mechanism—when the shortfall or cost occurs, which generation responds and which corridor changes—before testing storage.
Do this
- 1.Confirm the modeled facility energy equals 150 MW × 168 h = 25.20 GWh for the tutorial week.
- 2.Locate every unserved-energy hour and the corresponding Dholera demand, generation and incident-line state.
- 3.Trace incremental cost and emissions to the marginal dispatch change.
- 4.Compare local LMP proxy and corridor peak against the untouched base.
- 5.Save a no-BESS scorecard with result IDs and hourly evidence links.
Expected evidence
- ✓ A mechanism-backed baseline for cost, reliability, carbon and loading.
- ✓ Recorded native-hourly benchmark: ≈₹2,148/MWh incremental cost, 3.58 GWh unserved, 563 gCO₂/kWh marginal emissions and 22.5% local peak loading.
How to read the result
- • A 3.58 GWh shortfall over 25.20 GWh facility energy is material in the recorded screen; do not bury it beneath annual-policy arithmetic.
- • A non-binding local corridor does not prove deliverability when system resource adequacy or a different constraint causes unserved energy.
- • The LMP proxy is an optimization output at the modeled node, not a Gujarat market or retail tariff.
Trust check
Do not introduce BESS until the no-BESS cause is understood; otherwise mitigation can improve a number for the wrong reason.
Solve the controlled BESS mitigation case
Return to the same untouched base and change only the co-located storage input to 30 MW / 4 h.
Do this
- 1.Re-select the untouched base scenario.
- 2.Set Facility 150 MW, BESS 30 MW, Duration 4 h and Snapshot stride 1.
- 3.Run and confirm the clone contains 30 MW / 4 h = 120 MWh at the selected Dholera bus.
- 4.Save both new result IDs and compare them with the no-BESS pair.
Expected evidence
- ✓ A second persisted clone with identical load and horizon, plus one 30 MW / 4 h BESS.
Copy into Modeler Agent
Build and solve a fresh 150 MW nodal Dholera data-centre scenario with a 30 MW / 4 hour battery and snapshot stride 1. Compare it with the same untouched base and report the modeled-week operational cost, emissions, unserved energy, curtailment, local price and congestion effects. Do not claim battery ROI.
Trust check
The storage implementation uses 92% charge and 92% discharge efficiency and excludes capex, degradation and financing.
Interpret the result across all four dimensions
A battery can improve reliability and cost while worsening marginal emissions or local corridor loading; report the full trade-off.
Do this
- 1.Compare unserved energy and operating cost first.
- 2.Then compare marginal emissions and the same incident corridor.
- 3.Explain the dispatch mechanism behind each direction of change.
- 4.Keep every result tied to the modeled horizon and result ID.
Expected evidence
- ✓ A decision table that can contain both green and red outcomes without forcing one headline verdict.
| Metric | No BESS | 30 MW / 4 h BESS | Interpretation |
|---|---|---|---|
| Unserved energy | 3.58 GWh | 2.94 GWh | about 18% lower in the recorded modeled week |
| Incremental operating cost | recorded baseline | about 3.6% lower | roughly ₹19 lakh lower in the recorded week |
| Marginal emissions | 563 gCO₂/kWh | 607 gCO₂/kWh | higher in the recorded dispatch |
| Local corridor loading | 22.5% | 24.4% | higher but not binding in the recorded case |
Trust check
The recorded demonstration found a mixed outcome; rerun current data before using any historical value in a live project.
Dive into hourly price, carbon and corridor behavior
Aggregate deltas become actionable only when the critical hours, marginal generators and affected network elements are identified.
Do this
- 1.Plot facility load, unserved energy, local LMP proxy, system marginal emissions and corridor loading on one clock.
- 2.Mark BESS charge/discharge and SOC when comparing the mitigation case.
- 3.Identify whether the worst reliability, price, carbon and loading hours coincide or trade off.
- 4.Open the cloned scenario on Map and inspect the selected Dholera bus and incident corridors.
Expected evidence
- ✓ A critical-hours table with timestamp, driver, affected asset and mitigation response.
- ✓ A statement of which objective improves and which worsens in each critical interval.
How to read the result
- • The battery can reduce shortfall while increasing loading if charging or local injection changes corridor flow direction.
- • Annual green matching cannot answer whether critical-hour marginal supply is clean or deliverable.
| Hour type | Inspect | Mechanism question | Decision consequence |
|---|---|---|---|
| Peak shortfall | Demand, supply, SOC, constraints | Was energy or deliverability limiting? | Firm supply / flexibility need |
| Highest LMP | Marginal unit and binding line | Scarcity or congestion? | Procurement and siting risk |
| Highest marginal carbon | Charging and displaced generator | Did storage shift into dirtier supply? | 24×7 CFE strategy |
| Highest loading | Incident corridor and direction | Local injection or regional flow? | Utility rating / reinforcement request |
Trust check
Keep the timestamp and asset identity beside every critical-hour claim so reviewers can reproduce it.
Run the developer sensitivity matrix
Test the inputs most likely to change siting, procurement and reinforcement decisions before requesting final diligence.
Do this
- 1.Run PUE 1.20/1.30/1.40 and load factor 70%/85%/100% or the project’s credible ranges.
- 2.Run flat load versus bounded workload flexibility and explicit rebound.
- 3.Run BESS power/duration and efficiency ranges from the same untouched base.
- 4.Stress corridor ratings and renewable availability only as clearly labeled assumptions.
- 5.Report the threshold at which unserved energy, cost or corridor loading becomes unacceptable.
Expected evidence
- ✓ A low/base/high decision table with result IDs and one-variable-at-a-time changes.
- ✓ A ranked list of developer, utility and supplier inputs by decision sensitivity.
How to read the result
- • PUE changes both annual procurement and local peak; load factor changes the chronology and may alter the marginal generator.
- • If a small rating change flips the result, utility-validated seasonal ratings become the highest-value evidence request.
- • If BESS value collapses under modest spread compression, an investment claim requires contracted revenue rather than merchant extrapolation.
Trust check
A robust screen states the tested range and failure threshold, not just the preferred base-case number.
Convert the screen into a diligence request
Use the model to identify which private or utility data would most change the decision.
Do this
- 1.Request sanctioned-demand definition, PUE, load shape, redundancy and flexible-compute limits from the developer.
- 2.Request bay availability, thermal ratings, outages, protection, losses, charges and reinforcement options from the utility.
- 3.Request PPA/RTC/green premium, DSM exposure and BESS commercial terms from procurement counterparties.
- 4.Rerun the same evidence chain when those inputs arrive.
Expected evidence
- ✓ A prioritized diligence list tied to model sensitivity rather than a generic data request.
Trust check
The value of the screen is knowing which missing evidence can overturn the decision—not pretending public inputs are final.
Before you share
Completion checklist
- 150 MW is defined as grid draw or another explicit quantity.
- Policy terms, observed windows, derived inputs and modeled outputs are separated.
- The Gujarat network fingerprint and named Dholera yard are rechecked live.
- No-BESS and BESS cases start from the same untouched base.
- Both cases use snapshot stride 1 and the same horizon.
- Cost, reliability, carbon and grid loading are reported together.
- The modeled:nodal-economic-dispatch limit is visible.
- Utility, procurement and investment evidence gaps are listed as next diligence.
Answer desk
Frequently asked questions
Can Modeler prove that a 150 MW Dholera data centre can connect?
No. It can produce a planning screen on the available network. Connection proof requires utility-validated ratings, outages, bay and protection data, reactive-power/AC studies and the formal approval process.
Why does the tutorial test 30 MW / 4 h BESS?
It is a material but legible mitigation sensitivity: 30 MW power and 120 MWh energy against a 150 MW flat facility. It is not asserted to be the optimum investment size.
Are the demonstration result numbers current forecasts?
No. They are dated modeled outputs from a recorded workflow and illustrate interpretation. Rerun the current model and sources before a decision; observed market evidence is not a forecast.
Does 51% green energy mean 24×7 carbon-free power?
No. A 51% annual energy match is policy arithmetic. A 24×7 claim requires hourly matching, contractual evidence, losses and residual-supply accounting.