What four nines means for power.

Four nines — 99.99 percent availability — allows a facility only minutes of downtime in a year. Three nines allows hours. Each added nine shrinks the allowance by an order of magnitude, and each one has to be engineered rather than promised. The foundation is always the power system, because power is the one failure that takes everything down at once.

The nines, in plain terms

Availability is the fraction of time a system is up. Three nines means roughly hours of outage a year; four nines, minutes; five nines, barely moments. The vocabulary matters because the last nine always costs the most: getting from hours to minutes means eliminating every single point of failure, and the closer a design gets to always-on, the more it must assume components will fail and route around them without stopping.

Tiers as a design vocabulary

The data center industry's tier language describes escalating resilience: from a single path for power and cooling, to redundant components, to systems that can be maintained without shutting down, to architectures that ride through any single failure without the load noticing. The useful insight is that tiers describe topology, not luck — how many independent paths exist between the load and its sources of power and cooling, and whether maintenance ever requires an outage. A tier is a wiring diagram before it is a certificate.

Power is the foundation

Cooling failures unfold over minutes; network failures affect some traffic; a power failure is instantaneous and total. That is why availability engineering concentrates at the power system: redundant generation or feeds, equipment sized N+1 or better, uninterruptible power bridging every transition, and commissioning rigorous enough to prove the failure modes were actually designed out. Whether the electrons come from a utility or from dedicated on-site generation, the same discipline applies — the design determines the nines, not the source. What changes off-grid is ownership: the operator engineers redundancy directly and contracts for availability explicitly, instead of inheriting whatever the surrounding grid provides. The cost side of that trade is in behind-the-meter vs grid: the cost math.

Match the target to the workload

Not every AI workload needs the same nines. Training checkpoints and resumes; a brief, managed interruption costs time, not the run. Customer-facing inference is stricter — downtime is visible the moment it happens. Buying more nines than a workload needs means buying redundancy it will never use, so the sharpest operators set availability targets per workload and let the power architecture follow, rather than paying for the most severe tier everywhere. The provider conversation should start there — a fuller checklist is in how to evaluate a BTM power provider.

About Corley Energy

Corley Energy is a behind-the-meter independent power producer, founded in 2024 by Jake Corley, Tim Bozeman, and Mark Meyer. We convert stranded Permian Basin natural gas into firm, contracted electricity for AI data centers at Power Foundry, our ~1,000-acre development in Upton County, Texas. Start with what a power foundry is, see the company facts, or check current capacity on the Sites page.

Keep reading

What is an interconnection cluster study? · What is an AI factory? · What is driving the power shortage for AI data centers? · Browse the full library