The difference comes down to who holds the switch when the grid drops. A microgrid sits inside a defined electrical boundary as a single controllable entity and has to carry critical load for at least 2 hours on its own.[1] A VPP aggregates the same modules and batteries across sites and never disconnects. Two numbers settle which one fits: your annual peak load, and the hours a year your site goes dark.
Cost separates them fastest. A microgrid runs $2 million to $5 million per MW in the continental United States, so a 500 kW build lands between $1 million and $2.5 million.[2] The VPP side carries almost none of that. Its aggregation floor is 100 kW, so a 1 MW site clears the entry bar ten times over.[3]
US VPP capacity sits at 30 to 60 GW today against 80 to 160 GW targeted for 2030.[4] Two outcomes follow on a 1 MW load.
Above two outages a year, with a 4-hour event near $60,000, the microgrid pays for itself. Below one outage a year, with a peak-valley spread holding above 0.43 yuan/kWh, the VPP is cheaper.
They also stack. Any microgrid under a 20 MW maximum load can itself be aggregated, selling whatever capacity is left once the islanding reserve is set aside.[1]
The difference in one line
Coverage runs on two different clocks. The 4 hours you are dark belong to the microgrid case, and the 2 cycles inside a day belong to the VPP. Same modules, same batteries, two bookings $80,000 apart each year.
The two official definitions run under 200 words combined and point in opposite directions. Microgrid language keys on electrical boundary and a single controllable entity. VPP language keys on aggregation and market trading.
A microgrid is a group of interconnected loads and distributed energy resources within clearly defined electrical boundaries that acts as a single controllable entity with respect to the grid.[2]
Two hard limits sit in completely different units. Microgrids stop at 35 kV and 20 MW of maximum load, while a VPP starts at a 100 kW aggregation floor and reaches across several points of connection.[5]
Neither model is a synonym for solar plus storage, and both usually sit on top of it. The settlement boundary separates them: a microgrid settles against a distribution network it can leave, a VPP against 3 market levels it cannot.
Hardware differs too. Switchgear, protection relays, a controller and a dispatchable source all sit inside the microgrid boundary, whereas a VPP runs the same batteries with metering and telemetry solid enough to settle a market position every 15 minutes.
Four lines that never blur
Boundary comes first. Microgrid assets have to sit inside one electrical scope, generally at 35 kV or below. VPP resources can spread across several grid connection points, as long as they clear at the same market node.[5]
Islanding comes second, and cannot be bought back. The microgrid must hold critical load for 2 hours or more after the main grid fails.[1] Disconnecting means leaving the market, so islanding capability earns a VPP nothing.
Market level splits with it. Microgrid trading runs against the distribution network, while a VPP registers as an independent market entity across 3 markets with separate pricing for medium and long term, spot and ancillary services.[5]
Who pays decides who carries idle cost. Amortised at $2 million per MW over 25 years, a 1 MW microgrid costs $80,000 a year whether or not an outage happens. VPP capital goes on controllers and metering terminals instead.
Capacity rules of thumb set the scale. DOE assigns 500 kW to 200 homes or a supermarket, and 1.5 MW to a hospital.[2]
| Dimension | Virtual power plant | Microgrid |
|---|---|---|
| Physical boundary | Cross-region, no fixed scope | Within a defined boundary, up to 35 kV |
| Island mode | Not supported | Critical load held 2 hours or more |
| Market level | Medium term, spot, ancillary | Two-way services at distribution level |
| Upfront cost | Controllers and metering | $2M-$5M per MW |
| Primary return | Capacity, spread, ancillary | Avoided production loss |
What a microgrid costs and buys
Development cost in the continental United States runs $2 million to $5 million per MW, a 2.5-times spread. Converted to a 500 kW build, that is $1 million at the low end and $2.5 million at the high end, and the whole gap sits in islanding duration.[2]
Scale effects show up harder on the generation side. US solar plants commissioned in 2024 cost $8,062 per kW in the 1 to 5 MW band, $1,831 from 10 to 100 MW, and $1,496 above 100 MW, a 5.4-times spread from top to bottom.[6]
US solar averaged $1,865 per kW that year, and battery storage $1,469 per kW. Both figures cover plants of 1 MW or more, so neither includes distributed generation.[6]
China's grid-connected microgrid rules are tighter, and all four traits are hard limits. The 2017 pilot measure, which ran for 3 years rather than as a standing standard, set 35 kV or below and a maximum load of 20 MW. That voltage and capacity language still frames the sector.[1]
With self-governance, a microgrid has internal controls that guarantee independent operation of load and electrical equipment. It is able to balance supply and demand on its own and to black-start, and it must keep critical load energised for at least 2 hours when running independently. Annual energy exchanged with the external grid generally stays under 50% of annual consumption.[1]
Two of those limits bind hardest. The 2-hour islanding duty sets battery size: 2 hours at 1 MW of critical load needs about 2.3 MWh of nameplate once 90% depth of discharge and 95% discharge-path efficiency are applied.
The exchange cap is the counterweight. A grid-connected microgrid can trade no more than 50% of annual consumption with the external grid.
Duration is the next cost frontier. The DOE long-duration storage shot targets a 90% cut in grid-scale storage costs within the decade for systems delivering 10+ hours.[7]
What a microgrid sells is exactly one thing: not going dark. Berkeley Lab quantified that insurance through customer surveys, and for a medium-to-large commercial or industrial user the loss from a 1-hour outage on a summer weekday afternoon is $20,360, in 2008 dollars.[8]
Stretch the duration and the cost climbs faster than the clock. Extending that to 4 hours costs $59,188 and to 8 hours $93,890, so an 8-hour event runs 4.6 times a 1-hour one.[8]
Manufacturing sits higher again. A 1-hour outage costs $37,238, 4 hours cost $104,019, and 8 hours cost $164,033, which at the 4-hour mark is 1.76 times the all-industry figure.[8]
The insurance reads differently at each scale. A $5 million-per-MW build costs $200,000 a year and needs four 4-hour outages to justify; at $2 million per MW the same build costs $80,000, and two outages cover it.
| Duration | All industries | Manufacturing |
|---|---|---|
| 30 minutes | 15,709 | 29,098 |
| 1 hour | 20,360 | 37,238 |
| 4 hours | 59,188 | 104,019 |
| 8 hours | 93,890 | 164,033 |
Where a VPP's money comes from
US virtual power plant capacity sits between 30 and 60 GW. The Department of Energy puts the 2030 target at 80 to 160 GW, covering 10% to 20% of peak demand and saving roughly $10 billion a year in grid costs.[4]
A VPP is an aggregation of distributed energy resources such as rooftop solar with customer-sited batteries, electric vehicles and chargers, smart buildings and equipment and their controls, and flexible commercial and industrial loads that can balance electricity demand and supply and provide utility-scale and utility-grade grid services like traditional power plants.[4]
At the 120 GW midpoint, that $10 billion works out to $83 per kW per year, money that would otherwise be spent on new peaking plants.[4]
The gap has a concrete source. The US needs more than 200 GW of new peaking capacity before 2030, while new generation queues average 5 years.[4]
China works to policy targets instead. National VPP adjustable capacity is set at 20 GW or more by 2027 and 50 GW or more by 2030, a 2.5-times rise over 3 years.[5]
Only one of the three revenue layers is broadly proven today. Most Chinese projects still earn from demand response alone, and Guangdong, Shandong and Shanxi have moved fastest.[9]
Berkeley Lab profiled 214 US utilities with peak demand above 100 MW. The top 10% already shave more than 12% off system peak. Lifting the remaining 90% of that same 214 to the same rate adds 50 GW or more of flexible capacity. Otter Tail Power, a Minnesota utility, shaves 15% off system peak on its own.[10]
Individual devices can be tiny. Ontario built a 90 MW virtual power plant from 100,000 smart thermostats in six months. That averages 0.9 kW per device.[10]
- Capacity payments, priced on available flexible capacity and settled monthly
- Demand response, paid on delivered reduction or shaved peak
- Spot spread, earned by charging cheap and discharging dear
- Ancillary services, covering frequency regulation, reserve and black start
A 1 MW rooftop makes the numbers concrete. what a battery does to a 1 MW rooftop's returns puts 1,100 full-load hours at 1.1 million kWh and 572,000 yuan a year grid-tied, rising to 1.22 million yuan on two daily cycles with a 2.2 million yuan battery.
The spread gates sit at 0.43 and 0.6 yuan/kWh. Two cycles a day at a spread above 0.6 yuan/kWh favours PV plus storage. If a second cycle cannot be scheduled and the spread stays under 0.43 yuan/kWh, plain grid-tied wins.
Storage pricing needs the right yardstick. A 2-hour system cleared at 602.1 yuan/kWh on hardware and 1,040.4 yuan/kWh on EPC, while an industrial cabinet reached 1,176 yuan/kWh on EPC. Cabinet against container is 13% more on EPC and 26% more on hardware alone.
Industrial sites carry revenue per account. A single 500 kW unit from the standard 100, 125, 250 and 500 kW frames beats 500 thermostats on flexible capacity.
The 1.8 sizing factor is the easiest number to get wrong. It comes from multiplying 85% round-trip efficiency, 80% end-of-life capacity and an 80% usable state-of-charge window to get 0.54, then taking the reciprocal.
Replacement sits in the same ledger. A battery cycling once a day lasts 18.2 years, so 25 years needs one swap; at two cycles a day two.
Where the two overlap
Every microgrid is itself a distributed energy resource. Once a 1 MW microgrid joins a VPP, it bids on flexible capacity rather than islanding, and both revenue streams settle separately.[2]
China restricts double counting. One resource cannot be aggregated by two or more VPPs at once, and resources bidding into the spot market must in principle sit at the same clearing node, so one battery signs one aggregation agreement.[5]
The overlap is not a niche. Global renewable capacity grows by 4,600 GW before 2030, which spreads to 920 GW a year over five years, and solar PV takes close to 80% of that.[11]
Some manufacturers already run solar, storage and virtual power plants as one business line, with PV-plus-storage, virtual power plants and zero-carbon industrial parks covering more than 56 power stations and over 5.1 GW of grid-connected capacity.
Solar sits under both
Commercial rooftop module classes run from 480 W up to 802.43 W. The modules rated 615–635 W for commercial and industrial rooftops cover 615 W to 635 W at 22.8% to 23.5% efficiency and a Pmax temperature coefficient of -0.28% per degree C.
Module class sets the piece count. A 1 MW array needs 1,575 modules at 635 W or 1,626 at 615 W, a 51-module difference, which at 31 modules per pallet comes to 1.6 pallets.
Footprint follows from the class. One 635 W module measures 2.7 sq m, so 1 MW occupies about 4,255 sq m before racking and walkways.
Twenty-five years of output needs a discount. Degradation runs 2% in year one and 0.55% a year after, leaving 84.8% of nameplate at year 25, and that 84.8% goes straight into the denominator of any revenue model.
Shadow length tracks solar altitude, so the row-spacing bill only arrives in winter. the row clearance that keeps modules out of each other's shade turns into a shaded band across the bottom row when it is too small, and on a 1,100 full-load-hour roof a 5% assumption is 55,000 kWh.
Decide with your own numbers
Turning your own data into an answer takes four steps and usually 3 to 5 days.
- Pull 12 months of 15-minute interval data, 35,040 points a year or 96 a day, and work out annual peak, hours above the limit and peak-valley spread.
- Count outages over the last three years and their duration, multiply by the matching interruption cost, and get an annual outage loss.
- Check eligibility. Aggregation depends on whether 100 kW-scale bids are accepted locally; a microgrid depends on staying at 35 kV or below with maximum load under 20 MW.
- Price the overlap. A built or planned microgrid can join a VPP and sell the capacity left after islanding reserve.
Interval length matters. Demand charges settle on 15-minute averages, and welders, crushers or large motor starts can hide more than 20% of a spike inside that window.
Headroom adds another 10% to 20% above required discharge power, and derating takes 5% to 10% above 40 degrees C.
Running both is the common case. Build PV plus storage first, then push the battery into aggregation, where sizing storage power against storage energy keeps the two numbers from being read as one.
| Your situation | Route to look at first | Key figures |
|---|---|---|
| 2 or more outages a year, heavy downtime cost | Microgrid | Islanding 2 hours or more; loss above $80,000 |
| 1 MW load, outages rare | Virtual power plant | Aggregation floor 100 kW; loss below $80,000 |
| Peak-valley spread 0.6 yuan/kWh or more | PV, storage and aggregation | Two cycles a day |
| Microgrid and storage already built | Join a virtual power plant | One aggregation per resource |
Fix the cost yardstick before choosing a route. commercial and utility-scale economics put the two 43% apart per watt, so any quote needs its DC or AC basis stated up front.
DC against AC is the more common trap. The same $1.56 million bill reads as $1.20 per watt on 1.3 MW of DC and $1.56 per watt on 1 MW of AC.
The full budget also carries work outside the EPC scope. On a 1 MW plant what a 1 MW build costs shows every $0.10 per watt adding $100,000, with interconnection routes, land preparation and contingency regularly falling outside the quoted price.
What reaches procurement should be a parameter list checked line by line. Module model, 5 power classes, temperature coefficients and warranty terms all belong in a contract annex, so run them past the product specification sheet first.
Built projects make good reference points. a commercial rooftop already running in Switzerland sits in the same 1 MW band, and its 60% self-consumption ratio is directly comparable.
Keep the two ledgers apart, because they settle on different clocks. The only shared hardware is the modules, the batteries and the inverters, and most sites in a set of industrial and commercial rooftops already built have developed just one of the two layers.
Put both numbers into one sum and the answer falls out. A 1 MW load with three 4-hour outages a year represents $178,000 of loss at the medium and large C&I average, while a $2 million-per-MW microgrid amortises to $80,000 a year, so three outages already exceed that by more than double.
Reverse the case and the same $80,000 buys insurance that sits idle most of the year. A site whose annual outage loss stays below $80,000 should finish the aggregation route before it prices switchgear.
Site assessment
Two numbers settle the route: your peak load and how many hours a year the site goes dark. Send us 12 months of bills at 15-minute resolution, your transformer rating and the outage log for the last three years. Our engineers will run the same arithmetic against your site, the 100 kW aggregation floor, the 2-hour islanding duty and the 0.43 yuan/kWh spread, before you commit a single yuan to switchgear.