In brief: Backup generators were designed to sit idle until the grid failed. Grid connection delays of more than five years, outage costs that regularly exceed $100,000, and new rules such as Texas SB 6 are pushing onsite generation into daily operation. Most standby diesel engines are permitted for only limited annual hours, so sites that want generation to work every day need equipment designed and permitted for continuous duty, typically multi-engine gas-fired plants that can run grid-connected, semi-islanded or fully islanded.
For decades, in markets with developed power grids, resilience had a simple set up. A facility bought its power from the grid. A diesel generator supported the facility, tested once a month and otherwise ignored. If the lights went out, it started. If they didn’t, it was an insurance policy nobody wanted to think about. That model worked because outages were rare, short, and someone else’s problem to fix. All three assumptions are now under pressure.
Extreme weather is testing networks built for a milder climate. During Winter Storm Uri in February 2021, more than 1,000 generating units across Texas and the South Central US experienced over 4,000 outages, derates or failures to start. Federal investigators found it was the fourth cold weather event in a decade to put the bulk power system at risk 1.
New grid capacity has been slow to deploy. Lawrence Berkeley National Laboratory reports that the median time from interconnection request to commercial operation passed five years for projects completed in 2025 2.
The cost of failure keeps rising. In Uptime Institute’s most recent survey, 57% of operators said their latest major outage cost more than $100,000 3. Power issues remain the most common cause of serious data center outages 4.
In the latest edition of Powering Resilience, we argued that reliability is becoming something you design for rather than something the grid guarantees. This edition follows that thread to its practical conclusion: the backup system is being asked to do far more than wait.

From standby to working asset
The old division was clean. Prime power ran continuously and reliably. Standby power ran when it had to.
Regulators are starting to blur that line. Texas Senate Bill 6, signed in June 2025, applies to large loads with onsite backup generation able to cover at least half of their demand. Those loads must disclose that generation, and ERCOT can direct them to run it or curtail load during a grid emergency 5. In effect, a private generator becomes part of the grid operator’s emergency toolkit. Other markets are watching closely.
Owners are reaching a similar conclusion from the other direction. If a site already has significant onsite generation, why leave it idle for almost every hour of the year? It could shave peaks, support a constrained grid connection, or bridge the years before full grid capacity arrives.
There is a catch, and the conversation about turning backup into prime power often skips it. Most standby diesel sets are permitted as emergency engines. Under EPA rules, those engines are limited to 100 hours a year for maintenance and testing. Only 50 of those hours can be used in non-emergency situations, and those hours generally cannot be used for peak shaving 6. An asset permitted to wait cannot simply be told to work.
Turning resilience into a daily operating capability usually means choosing equipment designed and permitted for continuous duty, typically gas-fired engines, with the fuel supply and maintenance regime to match. That decision has to be made early. It is very hard to retrofit.
“The first question we hear from data center developers has changed. It used to be how many hours of fuel storage they needed. Now it is how many hours a year they can actually run, and what the permit will allow. That tells you backup power has moved into the operating plan.”
– Alex Marshall, Group Business Development and Marketing Director, Clarke Energy, Rehlko
Many engines, not one big one
The second shift is in how redundancy is built. The traditional approach concentrates protection in a few large units. It works, but it creates single points of failure and awkward maintenance windows. Take one large set offline for service and much of the site’s redundancy goes with it.
A multi-engine plant behaves differently. With ten or twenty units running in parallel, one can be out for maintenance and the site barely notices. Capacity can be added in steps as demand grows, rather than bought years ahead. Add battery storage for fast response, and a control system that decides second by second which asset does what. At that point, resilience no longer lives in one machine. It becomes a property of the whole system.
Anyone who watched IT move from individual servers to distributed computing will recognize the pattern. Fault tolerance moved out of the box and into the architecture. Power is following the same path, and it is the principle behind the multi-engine power plants Clarke Energy has delivered for many years.
Between connected and islanded
The idea we find most useful here is semi-islanded operation. Most facilities are still designed for two states: on the grid, or off it during an outage. Real conditions are rarely that binary. A site might run fully grid-connected on a mild Tuesday. During a congestion event or a price spike, it might carry part of its own load. In a storm, it might island completely. The engineering challenge is to move smoothly across that whole range, not just to survive the switch from one end to the other.
That changes the question owners ask. “Can we ride through an outage?” is a compliance question. “How well do we operate when the grid is constrained, stressed or gone?” is a business question, and it is the one that protects revenue.
Why this has reached the boardroom
For data centers, the case is obvious. AI training runs do not pause politely, and customers notice minutes, not hours. The same logic now applies to automated production lines, hospitals, cold chain logistics and wastewater plants. In each of them, a power event shows up directly in revenue, regulatory standing and reputation.
That is why the resilience conversations we have with customers have moved from the facilities team to the executive table. The question is no longer which generator to buy. It is what operating posture the business needs, and what architecture delivers it.
What comes next
The sites that come through the next decade well will not be the ones with the biggest backup generator. They will be the ones that treated resilience as a design input from day one. That means multiple assets, built for real running hours, controlled as a single system, and able to work alongside the grid on good days and without it on bad ones.
Resilience used to mean having power when things went wrong. Now it means keeping the business running, and earning its keep, whatever the grid is doing.
Frequently asked questions
What is the difference between standby power and prime power?
Standby power is designed to run only when the grid fails, and in the US standby engines are usually permitted for limited annual hours. Prime power is designed and permitted to run continuously as a main or shared source of supply. The distinction matters because a standby generator cannot simply be repurposed for daily operation.
Can an emergency backup generator be used for peak shaving?
Generally not in the US. Under EPA rules at 40 CFR 63.6640(f), emergency stationary engines are limited to 100 hours a year for maintenance and testing, of which up to 50 hours may be used in non-emergency situations, and those hours generally cannot be used for peak shaving. Sites that want onsite generation to run daily typically need engines designed and permitted for continuous duty.
What does Texas SB 6 mean for data center backup generation?
Texas Senate Bill 6, signed in June 2025, requires large loads in the ERCOT region with onsite backup generation able to serve at least half of their demand to disclose that generation. During a grid emergency, ERCOT can direct those loads, after reasonable notice, to run their backup generation or curtail load.
What is semi-islanded operation?
Semi-islanded operation describes a facility designed to work across a range of grid conditions rather than just two states. It can run fully grid-connected, carry part of its own load during congestion or price spikes, and island completely during major disruptions, moving smoothly between these modes.
Why do multi-engine power plants improve resilience?
Spreading capacity across many smaller engines removes single points of failure. One unit can be taken offline for maintenance with little impact on the site, capacity can be added in steps as demand grows, and with battery storage and intelligent controls, resilience becomes a property of the whole system rather than one machine.
How much does a data center outage cost?
In Uptime Institute’s 2025 survey, 57% of operators said their most recent major outage cost more than $100,000. Power issues remain the most common cause of serious data center outages.
References
- FERC, NERC and Regional Entity Staff. The February 2021 Cold Weather Outages in Texas and the South Central United States. November 2021. https://www.nerc.com/pa/rrm/ea/Documents/February_2021_Cold_Weather_Report.pdf ↩︎
- Lawrence Berkeley National Laboratory. Queued Up: 2026 Edition, Characteristics of Power Plants Seeking Transmission Interconnection as of the End of 2025. 2026. https://emp.lbl.gov/queues ↩︎
- Uptime Institute. Annual Outage Analysis 2026 (announcement). May 2026. https://www.businesswire.com/news/home/20260513289344/en ↩︎
- Uptime Institute. Annual Outage Analysis 2025, UII Keynote Report 173. May 2025. https://intelligence.uptimeinstitute.com/resource/annual-outage-analysis-2025 ↩︎
- [5] Texas Legislature. S.B. 6, 89th Legislature, Regular Session (2025), enrolled version. https://capitol.texas.gov/tlodocs/89R/billtext/html/SB00006F.htm ↩︎
- [6] US Code of Federal Regulations. 40 CFR 63.6640(f), Requirements for emergency stationary RICE. https://www.ecfr.gov/current/title-40/chapter-I/subchapter-C/part-63/subpart-ZZZZ/subject-group-ECFR24cf3a3de5678d3/section-63.6640 ↩︎





