Source-linked AI summary
Minimizing Grid Interconnection Capacity Requirements for AI Data Centers: A Developer-Side Planning Framework with Onsite Resources and Workload Flexibility
Hassan Zahid Butt, Rida Fatima, Xingpeng Li
TL;DR
AI data center projects face prolonged grid-interconnection constraints, motivating a developer-side method for planning required capacity. ICP-AI jointly sizes PV and BESS and schedules deadline-constrained workload flexibility under an onsite budget. Capacity reductions vary by planning environment, while flexibility can substitute substantially for storage.
Problem
Grid interconnection can take years, and existing studies provide limited direct evidence on the minimum capacity a fixed data center project can achieve through onsite investment and workload flexibility.
Method
ICP-AI minimizes grid import capacity under a prescribed onsite budget while jointly sizing PV and BESS and scheduling workload deferrals, then selects the least-investment portfolio preserving that capacity.
Results
At $100 million, capacity reduction is about 6% under the baseline, exceeds 10% with monthly average solar availability, and reaches about 13.3% for the Mixed Business profile.
Takeaways & Limitations
Workload flexibility provides both direct interconnection-capacity reduction and substitution for physical storage, supporting investment-versus-interconnection planning.
Takeaways & Limitations
The framework excludes tariffs, operating revenues, degradation, augmentation, and operation and maintenance costs, so its linear investment model is not a lifecycle economic model.
Abstract
from arXiv · showhide
Securing grid interconnection capacity has become a bottleneck for AI data center projects and can take longer than constructing the facilities themselves. This mismatch can delay deployment for years, making early interconnection planning essential. This paper develops ICP-AI, an interconnection capacity planning framework from a data center developer's perspective. The framework minimizes grid import capacity under a prescribed onsite investment budget while jointly sizing photovoltaic (PV) and battery energy storage system (BESS) resources and scheduling deadline constrained workload flexibility. A secondary refinement fixes the minimum grid capacity and selects the minimum-investment PV-BESS portfolio among solutions that achieve that capacity. The framework is evaluated using monthly composite stress profiles across varying temporal assumptions, load shapes, flexible load fractions, and deferral windows. Results show that interconnection capacity reduction depends strongly on the planning environment: at a $100M budget, it is about 6% for the high load factor baseline, exceeds 10% under monthly average solar availability, and reaches 13.3% for a more diurnal load. At a $10M budget, 5% flexible load with a 1 h workload deferral window reduces BESS capacity from 15.30 to 4.87 MWh while increasing capacity reduction from 4.43% to 4.84%. To test sensitivity to temporal compression, the model is also solved over the full 8,760 h chronology, which preserves the main capacity and flexibility trends. Overall, ICP-AI quantifies the interconnection capacity and infrastructure substitution value of workload flexibility, providing an investment-interconnection frontier to support capital allocation and early project planning in constrained grid environments.
Nomenclature
The nomenclature defines indices, facility and PV inputs, investment parameters, flexibility variables, and installed-resource capacities used by ICP-AI.
- Indices m, d, h represent month, day, and hourly period; a and k represent flexible-load arrival hour and deferral index.
- L_m,h denotes native aggregate facility load, while CF_m,h denotes the PV capacity factor.
- B, c_PV, and c_B specify the onsite budget and PV-BESS capital-cost factors.
- S_PV and S_B denote installed PV capacity and BESS energy capacity, respectively.
- P_IC is the grid interconnection capacity variable, defined as the maximum permitted grid import.
- X_m,a,k represents flexible load served k hours after arrival, while E_m,h represents stored BESS energy.
1. Introduction
AI data center deployment is increasingly constrained by the time and capacity required for grid interconnection. The paper frames onsite resources and workload flexibility as developer-side tools for reducing that requirement.
- Grid planning, permitting, and infrastructure completion can require 5–15 years, creating a timing mismatch with data center development.
- Around 20% of global data center capacity planned through 2030 could face connection delays under current grid constraints.
- Developers therefore need to plan the grid capacity required by a fixed facility alongside the onsite investment needed to reduce it.
- Deadline-tolerant and batch-oriented workloads can be shifted in time, making computational flexibility a potential planning resource.
- PV reduces contemporaneous grid demand, while BESS shifts energy across hours that would otherwise establish the interconnection peak.
- ICP-AI minimizes grid import capacity while jointly sizing PV and BESS under a capital budget and scheduling eligible computation within a deferral window.
2. Literature Review
Prior research establishes computational and electrical flexibility as useful resources but generally studies operation, system planning, or broader infrastructure decisions. ICP-AI addresses the project-level tradeoff between onsite investment and minimum interconnection capacity.
- Prior studies use workload shifting for electricity cost reduction, demand response, carbon-free energy matching, and renewable coordination.
- AI-specific evidence includes a 25% power reduction for 3 h on a 256 GPU cluster while maintaining quality-of-service requirements.
- Storage research has jointly planned BESS with data centers and optimized storage for grid-flexibility services.
- Many AI-storage studies operate under utility-imposed peak limits or predefined interconnection envelopes rather than choosing the capacity limit itself.
- Recent interconnection studies examine siting, system expansion, network congestion, or uncertainty-driven capacity decisions.
- ICP-AI instead quantifies how a prescribed onsite budget changes interconnection capacity for a fixed facility while jointly optimizing PV, BESS, and workload flexibility.
- Research Gaps and Contributions: The paper identifies gaps in developer-side capacity planning and in systematic characterization of computational-flexibility and BESS substitution.
3. ICP-AI FRAMEWORK
ICP-AI models one grid-connected AI data center campus and jointly sizes onsite resources with flexible workload scheduling. Its primary optimization minimizes grid import capacity under a budget, while a refinement selects the least-cost portfolio that preserves that capacity.
- The framework takes facility load, PV profiles, flexibility parameters, and an onsite budget as inputs and returns resource sizes, workload schedules, and grid capacity.
- Grid Interconnection and Investment Formulation: The primary problem minimizes P_IC subject to an onsite capital-investment constraint.
- Grid Interconnection and Investment Formulation: The secondary refinement fixes P_IC at its optimum and selects the minimum-investment PV-BESS portfolio among capacity-optimal solutions.
- Grid Interconnection and Investment Formulation: The resulting P_IC is a modeled minimum grid import capacity, most applicable where an enforceable import cap or flexible service arrangement is recognized.
- Aggregate Workload Flexibility: Workload flexibility is represented as an aggregate facility fraction, with eligible demand deferred but not curtailed or served before arrival.
- PV, BESS, and Power Balance: The model uses PV capacity factors, BESS charging and discharging, hourly power balance, and stored-energy dynamics to satisfy facility demand without grid export or load shedding.
- Monthly Composite Design Conditions: The primary formulation uses twelve monthly 24 h stress profiles, pairing monthly peak-load days with independently selected minimum-daily-PV days.
4. CASE DESCRIPTION AND EXPERIMENTAL DESIGN
The study evaluates a 100 MW nominal IT AI data center using constructed load and PV profiles, baseline planning assumptions, and sensitivity cases for temporal design and workload flexibility. It solves each case as a mixed-integer linear program for minimum grid import capacity, followed by a minimum-investment refinement.
- Case description: The case represents a 100 MW nominal IT AI data center with fixed PUE of 1.2 and annual profiles generated from configurable usage and diurnal characteristics.The profile tool includes training, inference, and mixed workloads, with patterns ranging from nearly flat demand to business-oriented profiles.
- Case description: The formulation can represent other data center types and large flexible loads when their demand profile, flexible fraction, and allowable shifting window are characterized.The optimization does not depend on a specific training or inference workload archetype.
- Planning assumptions: PV and BESS capital costs use adopted 2026 NREL Annual Technology Baseline projections, while the study excludes tariffs, operating revenues, flexibility payments, degradation, augmentation, and operation and maintenance costs.The linear investment-cost representation and fixed 4 h BESS duration are capacity-planning assumptions rather than a lifecycle economic model.
- Baseline cases and sensitivity design: C1 excludes workload flexibility, whereas C2 allows 5% of aggregate facility load to be deferred by up to 1 h under deliberately moderate assumptions.The experimental design also includes sensitivity scenarios for temporal profile construction and a chronological robustness check.
- Optimization and computation: Each 288 h case first minimizes grid import capacity and then applies a minimum-investment refinement, with models implemented in Pyomo and solved as MILPs using Gurobi 12.0.3.The refinement fixes the capacity optimum and identifies the corresponding infrastructure portfolio; reported cases use a zero requested optimality gap.
- Optimization and computation: The 288 h design cases and the separate full 8,760 h chronological robustness check solve to optimality in under one minute on the specified hardware.Solutions are checked for power balance, grid capacity, terminal SOC, facility capacity, and flexible workload allocation residuals.
5. RESULTS AND DISCUSSION
Results show that onsite investment, temporal design, native load shape, and workload flexibility jointly determine grid-capacity reduction and the PV-BESS portfolio. Workload flexibility often changes infrastructure composition more substantially than final interconnection capacity, especially by substituting for BESS.
- Baseline Investment and Interconnection Capacity Frontier: 3.34%: Allowing 5% of aggregate demand to shift by up to 1 h reduces required grid capacity from 87.915 to 84.975 MW without PV or BESS.This isolates the direct interconnection capacity value of workload flexibility.
- Baseline Investment and Interconnection Capacity Frontier: 72.9%: A $10 million budget captures this share of the 5.340 MW reduction achieved at $100 million without workload flexibility, while flexibility raises the share to 77.8%.The results indicate steep initial benefits followed by diminishing returns at higher investment levels.
- Baseline Investment and Interconnection Capacity Frontier: 68.2%: At $10 million, workload flexibility reduces BESS from 15.301 to 4.872 MWh while lowering required grid capacity by an additional 0.363 MW.Flexible computation performs short-duration temporal shifting, allowing more investment to go to PV.
- Temporal Design Condition Sensitivity: 10.39% and 10.67%: Under D3 monthly average PV availability, C1 and C2 achieve these reductions at $100 million, compared with 6.07% and 6.22% under D1.The temporal pairing and solar-availability assumptions materially affect achievable capacity reduction.
- Load Shape and Load Factor Sensitivity: 13.31% and 13.37%: The Mixed Business profile achieves these $100 million reductions, more than doubling the Mixed Flat results of 6.07% and 6.22%.The value of onsite resources depends strongly on the temporal structure of native facility demand.
- Flexible Workload Sensitivity: 4.872 MWh: At $10 million and 5% flexibility, BESS falls from 15.301 MWh with no flexibility; increasing flexibility to 10% reduces BESS to 0.413 MWh while grid capacity falls only 0.143 MW further.Workload flexibility continues reshaping the minimum-investment portfolio after most direct capacity value is captured.
- Flexible Workload Sensitivity: 2.940 MW: A 1 h deferral window captures 77.0% of the maximum scheduling benefit, while reducing BESS from 15.301 to 4.872 MWh at $10 million.A 2 h window captures 97.9% of the capacity benefit and leaves 0.391 MWh of BESS; storage is fully displaced at 3 h.
6. Conclusion
ICP-AI shows that onsite resources and workload flexibility can reduce required grid interconnection capacity, but the achievable reduction depends strongly on the planning environment. The framework also quantifies workload flexibility’s value both for capacity reduction and for substituting for physical storage infrastructure.
- At a $100 million budget, interconnection capacity reduction is about 6% under the conservative baseline, exceeds 10% with monthly average solar availability, and reaches about 13.3% for the Mixed Business profile.These results show that temporal design conditions and native load shape materially affect the achievable benefit.
- At a $10 million budget, 5% flexibility with a 1 h deferral window reduces BESS capacity from 15.30 to 4.87 MWh while increasing capacity reduction from 4.43% to 4.84%.This demonstrates both direct interconnection-capacity value and substitution for physical storage infrastructure.
- The sensitivity results show diminishing returns in direct capacity value, while storage substitution can remain substantial as flexibility increases.
- A full 8,760 h chronological analysis preserves the main interconnection-capacity and workload-flexibility trends.