Pillar 3
Zero unplanned downtimeDesigned so a failed part isn't downtime.
Every pump, fan and power supply has a partner already running (design). Break parts and watch the output line.
Pumps10/101 per loop can stop
Fans16/161 per pair can stop
PSUs4/4full power on any 3
Design
All parts running. Full output, nothing flagged.
24hstandard response through our service partner, on your schedule
4hpriority response for critical deployments

Plan Service levels · supplier life ratings being confirmed
Investor read Reliability is what turns a pilot into a fleet order.
Server output, tokens/s Measured53,984 one-hour run, replayed
Left to right: 16 fans in 8 push-pull pairs around the radiator · distribution plate · 10 pumps in 5 loops of 2 · 8 GPUs at 600 W each · 1 CPU · 4 × 3 kW PSUs
Design, not yet failure-tested: 10 pumps in 5 loops of 2 in series (one per loop can stop, 5 at most), 16 fans in 8 push-pull pairs (one per pair, 8 at most), 4 × 3 kW 80 PLUS Titanium PSUs: full output on any three, power-capped on two (typical draw ~6.4 kW, estimate). Behaviour, as the Nova 8 deck puts it: a few parts down, full power; a few more, still running, just not at peak. The simulator treats up to 2 pumps, 3 fans and 1 PSU down as the design range (the development brief's claim) and shows untested output beyond that. No failure-injection test has been run yet. With every part running, the output line replays the measured one-hour run (5 Oct 2026); with parts flagged, the dashed line is the design claim, not a recording under failure. Pumps and fans selected for continuous duty; supplier life ratings being confirmed. The manifold and radiator are not duplicated. Service: 24 h standard and 4 h priority response through a service partner (plan, indicative and subject to agreement in each region). Schematic simplified; the 16 fans are two walls of 4 × 2.
V-3 · Zero unplanned downtime