iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
The right purchase starts with the service you need, not a generator or battery model. Define the loads that must stay online, the outage and islanding duration, growth plans, normal operating mode, and utility constraints. Then compare a complete architecture—UPS, transfer and distribution equipment, generation, storage, controls, fuel systems, and maintenance—against reliability, partial-load performance, emissions, space, delivery risk, and lifetime cost.
What should I evaluate before buying an on-site power system for a data center?
Use a written requirements brief before asking suppliers for proposals. It should state the present and future electrical load, the consequence of losing each load, the ride-through and backup duration, the required redundancy, and whether on-site assets will run only during outages or also during normal grid operation. This prevents a common error: comparing a UPS, a standby generator, and a microgrid as if they were interchangeable products.
A UPS conditions power and bridges short interruptions. Standby generation normally starts when utility power is unavailable and can run as long as fuel and maintenance plans permit. A microgrid coordinates local generation, storage, controls, and loads; it can remain connected to the utility or operate as an island. These layers are often combined in one data-center design.
1. Define the service before choosing technology
Identify the loads that cannot go off
Separate information-technology loads from cooling, pumps, fire and life-safety systems, controls, security, lighting, and other building services. Classify each load as critical, conditionally critical, or shed-able. Record both the electrical rating and the operational consequence of a short interruption. A server may tolerate a brief transfer only if its power path, UPS, and internal power supplies are designed for it; a chiller or pump may require a different sequence.
#1 Best Overall
- Critical loads: must remain within their voltage, frequency, and transfer limits.
- Conditional loads: can be restarted or transferred after a defined delay.
- Interruptible loads: may be curtailed to preserve capacity for critical systems.
Specify the outage scenario
State whether the objective is momentary ride-through, emergency backup, resilience to a defined regional event, or operation independent of the grid. A requirement such as “eight hours of backup” is incomplete unless it says which loads are included, the expected weather and ambient conditions, and whether refueling is possible during the event.
Choose the normal operating mode
Decide whether generation is standby-only, peak-shaving, demand-response capable, or intended to operate continuously behind the meter. Normal operation changes fuel use, maintenance intervals, emissions permitting, controls, utility tariffs, and the economic model. A system sized only for rare outages may have very different equipment and contracts from one dispatched every day.
2. Size for current demand, growth, and partial load
Build a load profile, not a single peak number
Measure or model peak, sustained, minimum, and ramping demand over a day and a year. Include power factor, harmonic content, inrush currents, cooling-season changes, and the sequence in which equipment starts after an outage. Record the load at the utility service, each critical distribution bus, and each planned expansion area.
DOE’s data-center design guidance recommends considering initial and future loads, especially part- and low-load conditions, when selecting electrical equipment. A facility that eventually reaches 20 MW may operate at a fraction of that value for years. Generators, UPS modules, inverters, and transformers can have materially different efficiency, minimum-loading, and maintenance behavior at that operating point.
Include expansion and rack-density changes
Document planned halls, higher-density racks, liquid-cooling pumps, new chillers, and compute changes. Define the date and capacity of each expansion and whether the electrical plant must be installed in phases. Ask suppliers to show the capacity available after one module is unavailable and after the planned growth step, rather than quoting only a nameplate total.
Test normal, partial, and peak cases
Require a model that reports voltage, frequency, efficiency, fuel consumption, heat rejection, and reserve margin at the actual operating points. A design optimized only for peak demand can be inefficient or poorly matched in ordinary operation. Ask how controls prevent unstable low loading, excessive wet-stacking or equivalent maintenance issues, and repeated start-stop cycling where those concerns apply.
3. Compare complete power architectures
The following options occupy different layers of a resilient power system. The descriptions are evaluation categories, not a recommendation that one technology is best for every site.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problems| Architecture or component | Primary role | Questions to resolve | Typical integration issues |
|---|---|---|---|
| UPS (battery, rotary, fuel-cell or other design) | Conditioned power and short-term ride-through | Required bridge time, redundancy, efficiency at actual load, battery or energy-storage replacement, bypass behavior, and serviceability | Coordination with generators, transfer equipment, harmonics, cooling, and maintenance bypasses |
| Standby generators | Supply loads when utility power is unavailable | Start and transfer time, sustained rating, fuel access and storage, black-start sequence, redundancy, testing, and emissions limits | Switchgear and paralleling controls, load steps, noise and heat rejection, refueling logistics, and shared-failure risks |
| Battery or other energy storage | UPS duty, longer-duration backup, grid support, or several services | Power rating, energy duration, charging source, state-of-charge reserve, replacement cycle, safety design, and controls | Fire protection, thermal management, inverter limits, degradation, utility interconnection, and dispatch priorities |
| Fuel cells and other on-site generation | Local electricity, potentially for standby or normal operation | Fuel availability, ramp and start behavior, efficiency at expected load, service support, emissions, water and heat rejection, and replacement plan | Fuel infrastructure, permitting, controls integration, spare parts, and evidence from comparable duty |
| Microgrid | Coordinates local generation, storage, controls, and loads while grid-connected or islanded | Islanding sequence, protection, black start, load prioritization, utility rules, operating staffing, and lifecycle responsibility | Greater controls and integration complexity, cybersecurity, interconnection studies, and potentially higher upfront cost |
DOE’s advisory board recommends evaluating cost, performance, reliability, availability, space, emissions, and supply-chain concerns. It lists renewable diesel, natural gas, renewable natural gas, fuel cells, battery storage, enhanced geothermal, and long-duration storage as technologies to assess. That list is not an endorsement or a finding that every option is commercially ready, permitted, or serviceable at every site.
4. Set a reliability and resilience objective
Define failure boundaries
Specify what must survive the loss of one generator, UPS module, inverter, transformer, switchboard section, fuel pump, control network, or shared utility connection. A nominally redundant system can still have a single fuel, cooling, protection, or controls dependency. Request a failure-modes-and-effects analysis and a one-line diagram showing independent paths.
Specify maintenance operation
Explain how equipment will be serviced without interrupting critical loads. Ask for maintenance-bypass arrangements, spare-module procedures, test-load requirements, and the maximum time a component may remain unavailable. Include planned maintenance of fuel systems, batteries, switchgear, controls, and fire-protection equipment—not just engine service.
Demonstrate performance
Put acceptance tests and recurring integrated-system tests in the contract. They should cover utility loss, start and transfer, load sequencing, islanding, re-connection, black start where applicable, failed-module scenarios, alarm reporting, and restoration to normal operation. Define pass/fail limits for voltage, frequency, transfer time, fuel autonomy, temperature, and load shed.
Recommended Free Tools
A historical DOE case illustrates why context matters. The First National Bank of Omaha Technology Center combined four 200 kW phosphoric-acid fuel cells (800 kW total) with four rotary UPS units and other equipment. The case reports more than 89,000 operating hours through October 2009 and independently verified designed total-system availability exceeding 99.9999%. Those figures describe that site-specific configuration and period; they are not a current benchmark or a guarantee for another facility.
5. Evaluate technology against the operating profile
Start, ramp, and load-following behavior
Ask how long each source takes to start, synchronize, accept a step load, and follow a changing load. Match that behavior to the UPS bridge time and the cooling system’s restart sequence. A source that is attractive at steady state may not meet a fast transfer or repeated cycling requirement without storage or additional controls.
Efficiency and minimum loading
Obtain efficiency and fuel-consumption curves at normal, partial, and peak load, not just a peak rating. Include parasitic loads such as pumps, fans, battery HVAC, fuel conditioning, and controls. Model the effect of running multiple smaller units versus fewer large units, including the ability to shut down unneeded modules while retaining the required reserve.
Service evidence and warranties
Request references with a comparable duty cycle and climate, documented availability terms, warranty exclusions, planned replacement intervals, and a parts and technician-response plan. Do not treat theoretical capability as proof that a permitted, maintainable site solution is available.
6. Check fuel, emissions, and physical constraints
Fuel resilience
Test whether the fuel supply will remain available during the same regional emergency that interrupts the grid. Document on-site storage volume, usable rather than nominal capacity, delivery contracts, access routes, pumping power, fuel quality, and priority during shortages. For batteries, document usable energy at the required temperature, reserve state of charge, recharge time, and the source of charging energy.
Site and building requirements
- Footprint, structural loading, access for replacement, and crane or lifting paths.
- Heat rejection, ventilation, water use, drainage, and winterization.
- Noise, vibration, exhaust routing, clearances, and fire protection.
- Battery thermal management, detection, suppression, separation, and emergency response.
- Space for fuel tanks, switchgear, transformers, controls, spare parts, and future modules.
Permits and emissions claims
Identify local air permits, operating-hour limits, emissions reporting, fuel rules, water approvals, and noise requirements before selecting equipment. DOE explicitly includes emissions, space, and supply-chain issues in technology assessment criteria, but there is no universal emissions ranking or permitting limit that applies to every jurisdiction. Require a site-specific permitting review and a credible method for measuring emissions, including fuel origin and the actual operating profile.
7. Engage the utility before design is locked
Ask the serving utility for its process and schedule before committing to equipment. A project can be technically sound yet infeasible if interconnection studies, protection upgrades, construction work, tariffs, or islanding rules are overlooked.
- What application, study, and construction steps are required?
- Who pays for studies, protection equipment, metering, communications, and utility-side upgrades?
- What queue, equipment-lead-time, and construction dates apply?
- Can the site export power, and are there export limits or prohibitions?
- What protection, synchronization, telemetry, and anti-islanding functions are mandatory?
- What tariff, demand-charge, standby-service, or rate changes result from on-site generation?
- What tests are required before parallel operation, islanding, and reconnection?
- Are grid services or demand response permitted, and who controls dispatch?
DOE FEMP interconnection and microgrid checklists are useful for organizing these questions. Keep written utility assumptions in the design basis and update the financial model when the utility provides costs or schedule constraints.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware match8. Build a full-life cost model
Include every cost category
| Cost area | Items to model |
|---|---|
| Capital and installation | Equipment, civil work, fuel systems, switchgear, transformers, controls, protection, engineering, commissioning, permits, and utility upgrades |
| Ongoing operation | Fuel, electricity, demand charges, water, staffing, monitoring, software, communications, and routine testing |
| Maintenance and replacement | Service contracts, consumables, batteries, engines or stacks, inverters, controls, major overhauls, and spare parts |
| Financial and end of life | Financing, incentives, insurance, decommissioning, disposal, site restoration, and residual value |
| Risk value | Expected outage cost, curtailment, fuel-price exposure, permit risk, schedule delay, and the value of flexibility |
Run sensitivity cases
Show results for fuel prices, operating hours, load growth, outage duration, interest rates, replacement timing, and the value assigned to avoided downtime. Use transparent assumptions and separate one-time costs from recurring costs. Current project prices are not established by the available DOE material; obtain competitive, site-specific quotations rather than applying a generic price per kilowatt.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.9. Decide who owns and operates the system
Compare owner-purchased equipment, an energy-service or project-finance structure, and third-party ownership with sale of on-site electricity. DOE FEMP notes that some on-site contracts can place operations, maintenance, repair, and replacement with the provider. The contract—not the label—determines who carries fuel, performance, permit, labor, insurance, uptime, and end-of-life risk.
Require clear language for availability guarantees, response times, planned outages, testing, spare parts, software access, cybersecurity incidents, change orders, utility curtailment, fuel-quality failures, force majeure, and termination. Confirm whether electricity-market rules restrict third-party sales or dispatch in the facility’s jurisdiction.
10. Check delivery, supply chain, and operations
Ask for lead times and alternates for transformers, switchgear, generators, storage, controls, and protection devices. DOE’s advisory board flags electrical-equipment supply-chain concerns. A single long-lead component can determine the schedule even when generation equipment is available.
Free tools Windows power users keep installed
One-click scans. No signup required.
- Assign one party responsibility for the integrated design and interface schedule.
- Define factory testing, site commissioning, operator training, and documentation deliverables.
- Verify local technicians, critical spares, service response, and escalation procedures.
- Require network segmentation, access control, patching, logging, and recovery plans for control systems.
- Plan phased construction so temporary power, tie-ins, and cutovers do not endanger live loads.
11. A practical buying process
- Write the requirements brief. Record critical loads, present and future demand, outage scenarios, operating mode, redundancy, environmental conditions, and utility assumptions.
- Gather load and site data. Use interval measurements where available; document power quality, short-circuit levels, space, fuel access, cooling, and expansion boundaries.
- Screen architectures. Compare UPS-plus-generation, storage-centered, generation-centered, and microgrid designs against the same criteria.
- Engage the utility and authorities. Start interconnection, permitting, tariff, and emergency-fuel discussions before equipment is selected.
- Issue a performance-based request for proposals. Require part-load curves, start and transfer behavior, redundancy diagrams, autonomy calculations, emissions data, maintenance plans, warranties, lead times, and lifecycle pricing.
- Model lifecycle and risk. Run sensitivity cases for fuel, growth, outages, replacements, financing, and schedule delay.
- Validate the preferred design. Review protection, controls, cybersecurity, fire safety, maintainability, and failure scenarios with independent engineering support where appropriate.
- Contract acceptance and operations. Specify integrated tests, training, documentation, recurring tests, service levels, reporting, and responsibility for every interface.
Should a data center use generators or a microgrid?
Choose generators when the requirement is primarily dependable standby power and the site does not need coordinated island operation, multiple distributed resources, or routine dispatch. Choose a microgrid when controls must coordinate several sources and loads, the facility may operate islanded, or normal-operation services justify the added integration work. In either case, a microgrid may still include generators and UPS equipment; the comparison is between a coordinated architecture and a standalone source, not between mutually exclusive boxes.
Make the decision with the same load, outage, interconnection, permitting, lifecycle-cost, and operations evidence. A microgrid’s broader capability brings more controls, protection, commissioning, and cybersecurity obligations, while a generator-only design may provide less flexibility for load management or multiple resources.
How do I size backup power for a data center?
Start with the loads that must remain energized, then add future demand, starting and transfer behavior, redundancy losses, auxiliary loads, environmental derating, and the required reserve. Calculate both power in kilowatts or megawatts and energy over time for the specified outage. For generators, verify fuel autonomy and refueling; for storage, verify usable energy, state-of-charge reserve, recharge source, degradation, and temperature effects. Validate the result at minimum, normal, and peak load rather than using a single nameplate value.
Common buying mistakes and fixes
- Buying to peak load only: require part-load and growth cases.
- Treating UPS autonomy as generator runtime: specify the bridge, start, transfer, and sustained-duration sequence separately.
- Ignoring shared dependencies: analyze common fuel, cooling, switchgear, controls, and communications.
- Leaving the utility until late: obtain interconnection, tariff, protection, and schedule information before design lock.
- Comparing equipment prices only: include installation, fuel, maintenance, replacements, staffing, financing, and outage value.
- Assuming a technology list is an endorsement: demand site-specific permits, references, service coverage, and performance evidence.
- Failing to test integrated behavior: make islanding, reconnection, failed-module, and load-shed tests contractual acceptance criteria.
- Leaving responsibility vague: assign ownership of operations, maintenance, fuel, software, cybersecurity, permits, and end of life in the contract.
Final decision checklist
- Required critical, conditional, and shed-able loads are documented.
- Current, future, minimum, sustained, and peak load profiles are modeled.
- Outage duration, ride-through, islanding, black-start, and reconnection requirements are explicit.
- Redundancy and single-failure boundaries are shown on a reviewed one-line diagram.
- Part-load efficiency, ramping, start behavior, fuel or storage autonomy, and environmental derating are evidenced.
- Permits, emissions, noise, water, fire safety, footprint, and access are confirmed for the site.
- Utility interconnection, tariffs, protection, export, testing, costs, and schedule are documented.
- Lifecycle cost includes capital, installation, fuel, maintenance, replacement, staffing, financing, and end of life.
- Lead times, spares, service response, training, controls integration, and cybersecurity are assigned.
- Acceptance tests and recurring integrated tests have measurable pass/fail criteria.
The buying decision is sound when the proposed architecture meets the defined service at normal and future loads, survives the specified failures, can be permitted and interconnected on the required schedule, and remains supportable for its full life. The equipment choice follows from that evidence; it should not substitute for it.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
