Availability and Hyperscale Reliability

The gap between the burner's studied uptime and what data centers require is one of its four honest gates.

The number that decides the use case

For most power customers, occasional downtime is an inconvenience. For a hyperscale data center, it can mean lost training runs, dropped services, and cascading failures. Tier III facilities target availability near 0.99982 — minutes of unplanned downtime per year. The burner (Aegis / MetroVolt), in current design and simulation studies, sits at an availability of 0.86 to 0.995.

That range is 30 to 100 times more downtime than a Tier III standard allows. It is not a rounding issue; it is the difference between a source you can build a data center's uptime guarantee on and one you cannot. We state it as a gate, not a caveat.

Why fusion uptime is hard

• Plasma sustainment, magnet systems, and thermal cycling all introduce failure modes
• Fuel handling and vacuum systems add maintenance demand
• Early machines lack the operational history that drives reliability engineering
• A first-of-a-kind is where these problems are found, not where they are already solved

The honest reading

On current analysis the burner is not a sole Tier-III-grade source. That does not make it useless to compute — as one firm source in a diversified supply, or for loads that tolerate more downtime, it could still contribute. But the claim that fusion powers hyperscale AI, full stop, is not supported by the numbers today. Closing this gate is future work, and we do not pretend it is done.

See the availability gate and the honest gates and impact.