What Solar Downtime Actually Costs
The framework, the formula, and worked examples at 3 MW, 50 MW, and 200 MW — because the business case for operational excellence should take five minutes, not a quarter. The cost of solar downtime equals affected capacity × capacity factor during the outage window × duration × energy price, plus dispatch cost and any contractual penalties. A full-plant daytime outage costs roughly $40–$60 per MW per hour at typical prices — but partial, undetected faults cost far more annually than dramatic outages, because they run for months instead of hours.
Ask an asset manager what an outage costs and you will get a shrug toward the O&M contract. Ask their CFO and you will get silence. This is odd, because the number is knowable to the dollar, the formula fits on an index card, and the answer reframes the entire operations budget. Here it is — with the worked examples at three scales, and the fleet case that demonstrates why the intuitive version of this analysis gets the priorities exactly backwards.
What is the downtime cost formula?
Energy cost of an event = affected capacity (MW) × expected capacity factor during the outage window × duration (hours) × energy price ($/MWh). Add direct response cost — a truck roll runs $700–$1,500 — and any performance-guarantee or PPA penalty exposure the event triggers. The one nuance that changes everything: use the capacity factor of the outage window, not the annual average. A solar plant's annual capacity factor of 25% is an average over nights, winters, and clouds; a clear summer midday hour produces at three to four times that rate. The same four-hour inverter fault therefore costs an order of magnitude more at noon in June than at 4 p.m. in December — which is precisely why maintenance windows are a financial decision, why "we'll get to it Monday" has a computable price, and why the formula belongs in the dispatcher's head and not just the annual report.
Worked examples: 3 MW, 50 MW, 200 MW
Read the table bottom-up. The dramatic events — the ones that generate phone calls — are the small numbers. The events nobody phones about, because nobody sees them, are the large ones. A 200 MW plant losing a quiet two percent forfeits more every year than a decade of inverter trips, and it will keep doing so until someone measures it. Note also what the dispatch line implies: on the smallest events, the truck roll costs more than the energy — which is the entire economic argument for desk-level diagnosis and event clustering before anyone drives anywhere.
| Scenario | 3 MW C&I | 50 MW utility | 200 MW utility |
|---|---|---|---|
| Inverter trip, 4 daytime hrs (1 unit) | $9 energy + dispatch | $450 + dispatch | $450–$900 + dispatch |
| Full-plant grid outage, 8 daytime hrs | $650 | $10,800 | $43,200 |
| 3 dead strings, undetected 6 months | $1,900 | $1,900 per cluster | Multiple clusters: $10k+ |
| 1 inverter at −48% vs peers, 4 months | $1,000 | $1,000 per affected unit | Scales with unit count |
| Chronic 1% plant-wide loss, full year | $2,100 | $35,000 | $140,000 |
| Typical annual undetected-loss band (1–3%) | $2,000–$6,000 | $35,000–$105,000 | $140,000–$420,000 |
Why do partial faults dominate the ledger?
Duration beats magnitude. A tripped inverter is loud, binary, and fixed within days; a dead string is silent, sub-threshold, and — in fleets without string-level detection — routinely survives one to two quarters. Multiply a small loss rate by a long exposure and it outweighs every acute event on the books. This is the arithmetic behind a claim we make often: detection latency, not repair speed, is the variable that governs a portfolio's downtime economics. Cutting mean time-to-repair from three days to two saves one day of one fault. Cutting mean time-to-detect from ninety days to two saves eighty-eight days of every fault in the silent category — which, as the table shows, is where the money was all along.
The case that proves it: 121 days, 6,973 kWh, 45 minutes
A real record from the Ellume fleet, and the cleanest illustration of this essay's argument we have ever logged. One inverter — INV-044 on a 3 MW plant — developed a DC string outage in late January. No trip. No alarm. The unit stayed online, "available" by every contractual definition, producing at 48.5% below its fleet-peer average: 19 kW where its peers made 37. It ran that way for 121 days, accumulating 6,973 kWh of loss — silently, through a spring of green monthly reports. Vector's peer-deviation physics surfaced it as a case: the measured deviation, the duration, the accumulated energy, the pattern signature pointing to a string outage, and a recommendation — 45 minutes of repair effort, 92% confidence, 14.1 kWh per day of recoverable yield. It was clustered with two neighboring faults on the same combiner row and a soiling job into one truck roll. The repair took less than an hour. The fault had run for a third of a year. That ratio — 121 days of exposure against 45 minutes of repair — is the entire economics of solar downtime in one event. Nothing about the fix was hard. Everything about the detection was, for any system watching availability instead of physics.
From the Ellume fleet — the monthly ledger
The same plant's running ledger for the period: 16,156 kWh and $1,777 in active fault losses surfaced and priced on the executive dashboard; 58 kWh/day recoverable across open recommendations if actioned; five truck rolls avoided and $3,500 saved through false-positive suppression and clustering. Downtime economics stop being an annual estimate when they are a standing monthly number — and start being managed the month they get one.
How does this build the business case?
Take your portfolio's capacity, apply the one-to-three percent undetected-loss band — or better, measure it with a trailing-year energy reconciliation — price it at your PPA, and add measured dispatch waste. That figure, recurring, annual, and conservative, is the budget line that operational intelligence competes against. For most portfolios above 20 MW it clears the cost of the entire detection stack several times over, which is why the honest framing was never "can we afford the tooling." It is "how long do we keep paying for its absence" — and the table above prices every month of the answer.
Frequently Asked Questions
- What energy price should the calculation use?
- Your actual contracted or realized price — PPA rate, merchant capture, or blended. At $25/MWh the table's figures shrink but the structure holds: partial faults still dominate, because duration still beats magnitude.
- Do performance guarantees recover these losses?
- Only the losses your measurement regime can prove, within the guarantee's methodology. Availability-based guarantees recover almost none of the partial-fault ledger — one more argument for energy-based contracts with attribution both parties can audit.
- Is nighttime downtime free?
- For energy, essentially yes — which is why smart maintenance windows and the capacity factor of the outage window belong in every dispatch decision. The formula prices this automatically; intuition does not.
- How do I estimate my portfolio's undetected-loss rate without a platform?
- Run a trailing-year reconciliation on one representative plant: expected generation from measured or satellite weather versus metered actuals, with known outages removed. The unexplained residual is your first estimate — and in our onboarding experience, it lands in the 1–3% band far more often than owners expect.
- Does the analysis change for merchant versus PPA assets?
- The structure holds; the stakes rise. Merchant exposure adds price-shape risk — losses during high-price hours cost disproportionately — which strengthens the case for intraday detection and weather-aware maintenance scheduling further.
