Every operator plans for redundancy. Very few write down what happens when redundancy has already failed and the entire site is dark.

A black start is the procedure for bringing a facility back from total power loss using only its own resources, with no external supply to lean on. For a grid-connected data center this is largely academic, because the utility comes back and the building follows. For an off-grid natural gas mining site it is an operational reality, and the difference between a two-hour recovery and a two-day one is almost entirely whether somebody wrote the sequence down before it was needed.

This guide covers that sequence. It assumes a containerized natural gas deployment of the kind we build, and it deliberately stops short of the equipment-selection questions covered in our guide to power redundancy, backup generators, ATS and UPS design.

Why a dead site does not simply restart

The intuition is that you start the generator and everything comes back. In practice a cold site fights you in four specific ways.

The generator needs power to start. A gas genset requires battery power for the starter and the engine control unit, and on larger units an air or hydraulic start system. If those batteries have been sitting flat since the shutdown, nothing turns over. This is the single most common reason a black start stalls at minute zero.

The gas train needs power too. Solenoid valves, the fuel control system and often a booster compressor all draw electrical power. On a wellhead site, the gas conditioning skid may also need power before it can deliver fuel of usable quality, which we cover in our guide to gas quality conditioning for mining gensets. You have a circular dependency: no power without gas, no gas without power.

Everything wants to start at once. When power returns, every miner, fan, pump and transformer tries to energize simultaneously. Inrush current on a large block of ASICs and the magnetizing inrush of distribution transformers can be several times steady-state draw. A generator sized comfortably for running load can be badly undersized for everything starting together, and it will trip on overcurrent or stall on a frequency dip.

Control systems come back in an unknown state. Controllers reboot, some hold their last configuration and some default. Miners may come up at a different power profile than you left them on. Without a defined restart state you are not restoring the site, you are discovering it.

What has to exist before the lights go out

A black start is a preparation problem that presents itself as an emergency problem. Four things have to be in place in advance.

• An independent starting supply. Maintained start batteries, verified under load rather than by voltage alone, or a small dedicated starting generator. A battery that reads correct voltage and collapses under cranking load is a battery that will fail you exactly once.
• A documented restart sequence. Written, posted in the container, and specific to your configuration. Not a general principle, an ordered list with the actual breaker and panel designations on your site.
• Manual overrides you have actually used. Every automated transfer and control function needs a manual path, and somebody needs to have operated it during a drill rather than reading about it at 3am.
• Enough fuel pressure to start. On a wellhead site, know whether the well flows without powered equipment. If it does not, you need a starting fuel source, and that is a design decision rather than a procedure.

The sequence

The principle underneath every step is the same: bring load on in blocks small enough that the generator can absorb each one and stabilize before the next.

1. Make the site safe before energizing anything. Establish why it went down. A black start into an unresolved fault repeats the fault, sometimes destructively. If anyone has been working on equipment, verify that locks and tags are cleared properly, following your lockout/tagout procedure. Confirm nobody is inside a panel.

2. Open everything downstream. Before starting any generator, open the breakers feeding miners, and ideally the distribution transformers too. You are deliberately starting into no load. Operators skip this step because it feels like extra work, and it is the step that prevents the generator from tripping the moment it closes.

3. Energize controls and the gas train. Bring up the starting supply, then the control power, then the gas train. Verify fuel pressure and that the conditioning equipment is doing its job before asking the engine for anything.

4. Start one generator and let it stabilize. One, not all of them. Let it reach rated speed and voltage and run unloaded long enough to confirm oil pressure, coolant temperature and stable frequency. A few minutes here is cheap insurance; the engine is also cold, and loading a cold engine hard is how you shorten its life, as our guide to oil analysis and condition monitoring explains.

5. Pick up house load first. Controls, lighting, ventilation, monitoring, communications. This is a small block, it confirms the generator holds load, and critically it restores your visibility. Trying to restart a site you cannot see the telemetry from is working blind.

6. Energize distribution in steps. Close transformers one at a time, allowing magnetizing inrush to settle between each. Watch frequency and voltage recover fully before the next step.

7. Bring miners on in blocks. This is where the discipline matters most. Energize a fraction of the fleet, let frequency recover, confirm the units are hashing, then take the next block. The right block size depends on your generator headroom; what matters is that the sequence is predetermined rather than improvised. Many operators also stage miners at reduced power and raise them once the full fleet is stable.

8. Parallel additional generators only once stable. If the site runs multiple units, synchronize them after the first is carrying load steadily, following the practice in our guide to generator paralleling and synchronization. Paralleling into an unstable bus creates a larger problem than the one you are solving.

Where black starts go wrong

The failures repeat across sites, and they are mostly impatience rather than ignorance.

Closing everything at once is the classic. The generator trips, the operator resets and tries again, and repeated trips damage equipment and flatten the start batteries until a recoverable situation becomes a callout.

Starting into the original fault wastes the attempt and can escalate the damage. Diagnose first.

Flat start batteries turn a one-hour recovery into a wait for someone to drive out with a charger. They are the cheapest item in the chain and the most frequently neglected.

No documented sequence means the recovery quality depends on who happens to be on site. The person who knows the site intuitively is not always the person who is there.

Never having rehearsed it is the deepest version of the same problem. A procedure that has only ever been read is a hypothesis.

Testing it

A black start procedure is only real if it has been performed. Schedule a planned shutdown and restart, ideally at a time when lost hashrate is cheapest, and run the written sequence exactly as written. Time each step and note every place the document was wrong or ambiguous, because there will be several on the first attempt.

Rotate who runs the drill. The purpose is not to prove the senior technician can restart the site; it is to prove the document can. Then revise it and date it.

Fold the result into your broader disaster recovery and business continuity planning, and revisit after any change to generation, distribution or fleet size, since all three change the block sizes in step seven.

Designing the problem down

Procedure compensates for design. Where you still have design freedom, a few choices make black starts far less eventful: generator headroom above steady-state load, distribution segmented so load can be picked up in natural blocks, a genuinely independent starting supply, and controls that default to a known safe state rather than to their last state.

Our natural gas MDU containers are built as self-contained generation and compute units for exactly this kind of off-grid deployment, where there is no utility to fall back on and the site has to be able to recover itself. If you are planning a deployment and want the restart sequence worked out before commissioning rather than after the first outage, our team can walk through it with you — see our consulting services or browse available units.

The site acceptance process is the right moment to capture it, while the equipment is fresh and the commissioning engineers are still on site. Our guide to site acceptance testing and the handover punchlist covers what else belongs in that window.