Below roughly 500 kW the question mostly answers itself – string inverters. Above a few megawatts, central. The interesting range is the commercial and industrial band in between, where both work and the decision is genuinely about how you want the plant to behave.
This is the part that matters most and gets discussed least.
A central inverter failing on a 1 MW plant takes the entire plant offline. One event, total loss, until a service engineer arrives with the right part. Depending on your contract and location, that is somewhere between two days and three weeks.
A string inverter failing on the same plant, configured as say twelve 80 kW units, takes roughly 8% of production offline. Annoying, not urgent. You keep generating while you arrange the swap, and swapping often means a two-person crew and a few hours rather than a specialist visit.
For a site where the solar is offsetting expensive imported power, that difference compounds. I generally weight it heavily.
Cost per watt, though the gap has narrowed considerably. Central units remain cheaper per kW at scale, and the DC-side wiring is simpler and cheaper because you are combining strings into fewer, larger runs.
Grid support capability. Central inverters generally offer more sophisticated reactive power control, low-voltage ride-through and grid-forming functions. If your network operator imposes serious grid code obligations – increasingly common in Europe – central hardware often handles them more readily.
Fewer things to monitor. Two central inverters produce two data streams. Twenty string inverters produce twenty. That is more visibility, but also more to manage.
One maintenance relationship. Central inverters usually come with a service contract and a defined response time, because the manufacturer knows an outage is total.
Roof geometry. Commercial roofs are rarely clean rectangles. Plant rooms, skylights, differing orientations, shading from adjacent structures. String inverters with multiple MPP trackers handle segmentation naturally; each tracker optimises its own portion. Forcing a fragmented roof onto a central inverter means mismatch losses you cannot recover.
Getting hardware onto the roof. A central inverter is a substantial cabinet needing a plinth, a route, and often a crane. String inverters are wall-mountable, carried up by hand.
Serviceability without specialists. Most competent electrical contractors can replace a string inverter. Central units usually require manufacturer-authorised service, which means their schedule, not yours.
Phased expansion. If you might add capacity later, string architecture extends cleanly.
Modular or multi-MPPT "string-central" units in the 150-350 kW range have taken a lot of this market, and often sensibly. You get partial redundancy and reasonable cost per watt without twenty separate devices.
Optimisers and microinverters solve a real problem – per-module mismatch and shading – at the cost of putting active electronics on the roof, under the modules, where the temperature cycling is worst and access is poor.
On a heavily shaded commercial roof they can be worth it. On a clean, unshaded industrial roof they add failure points and lifetime cost for benefit you will struggle to measure. Decide based on the actual shading analysis, not on a general preference.
Ask: what happens on the worst day? If a total plant outage for two weeks in July is merely irritating, central is fine and cheaper. If it means buying peak power at a punitive rate while your production line runs, distributed architecture is worth its premium.
Then check the roof geometry honestly. If the array splits into more than three or four distinct orientations or shading zones, the decision has largely been made for you.