Direct-Liquid-Cooled Compute Hall Commissioning Leader

The mandate

First energisation of a direct-liquid-cooled accelerator hall was stopped after particulate contamination crossed cold-plate limits and manifold controls oscillated under mixed rack demand. Facility, server and cooling vendors dispute responsibility, while the prior commissioning head departed before integrated systems testing. Hundreds of engineers now share constrained legacy compute as capital equipment waits for an evidence-based acceptance route.

The interim must take control in Seoul within three weeks and remain for eighteen months, with an extension possible only for a delayed utility or equipment acceptance outside management control. The permanent compute-facilities search launches after one production row completes sustained-load testing, expected around month ten. The successful candidate will overlap with the interim through final hall acceptance for at least eight weeks.

Handover requires coolant chemistry and cleanliness to remain within approved envelopes through two maintenance cycles; facility and technology cooling loops to pass transient, leak, isolation and heat-rejection tests; every rack class to have an accepted operating curve; operators to complete supervised fault drills; capacity to meet signed workload service levels; and the permanent leader to chair the final integrated test review.

The interim may quarantine equipment, stop commissioning, set test sequence, reject vendor evidence, redirect internal teams and commit up to KRW 45 billion inside the approved recovery budget. Design departures, claims settlement, permanent hiring, new hardware platforms and individual decisions above KRW 8 billion require committee approval. Safety interlocks, environmental limits and warranty conditions may not be waived.

Chip-design workflows, accelerator procurement strategy, fabrication systems and construction of the group’s other data centres are explicitly outside scope. The assignment covers this hall’s liquid loops, rack interfaces, integrated commissioning, vendor accountability, operational readiness, capacity release and transfer into stable HPC service.

Why this seat is open

The failed energisation exposed an accountability gap between facility completion and compute acceptance, then the commissioning head’s departure removed the integrator for both. Schedule pressure is increasing because legacy clusters cannot absorb the planned verification workload. Temporary executive authority is required to resolve evidence, enforce safe sequence and leave permanent ownership after the capital asset proves operable.

What you will own

• Establish a single defect and evidence baseline across coolant manufacture, flushing, filtration, manifolds, controls, cold plates, heat exchangers and monitoring.
• Decide quarantine, remediation, retest and acceptance disposition for each affected loop, rack class and supplier boundary.
• Define operating envelopes for flow, pressure, temperature, chemistry, heat load, ramp, partial occupancy, isolation and emergency shutdown.
• Command integrated tests covering mixed workloads, pump loss, valve failure, leak detection, fouling, facility-water derate and maintenance bypass.
• Reconcile compute release with power, cooling, network, scheduler, firmware, safety and operator-readiness gates instead of facility completion alone.
• Enforce vendor root-cause, warranty, spares and corrective-action commitments through measured acceptance evidence and expiry-controlled concessions.
• Transfer accepted curves, defect status, maintenance standards, capacity limits, test records and escalation rights through successor-led review.

Candidate qualifications

• Held executive commissioning or operations accountability for direct-liquid-cooled HPC, AI or high-density data-centre capacity.
• Resolved coolant contamination, flow instability, cold-plate, manifold or heat-rejection defects across multiple supplier boundaries.
• Designed integrated systems tests joining facility infrastructure, rack hardware, controls, workload behaviour and operator response.
• Rejected incomplete vendor evidence under schedule pressure while preserving warranty, safety and commercial rights.
• Released compute capacity against measured thermal envelopes and production workload requirements rather than nameplate facility ratings.
• Transferred a newly commissioned critical facility to permanent operations with defects, concessions and decision rights visibly owned.

Non-negotiables

• Can relocate to Seoul within three weeks and remain site-led through testing, initial operation and successor transition.
• Will accept exclusive executive responsibility and continuous escalation duty during energisation and high-risk integrated tests.
• Brings production direct-liquid cooling at meaningful rack density; conventional chilled-water facility delivery alone is insufficient.
• Must disclose ties to cooling, server, facility, commissioning and monitoring suppliers engaged in the disputed acceptance.

1. 49 words maximum. State your earliest Seoul start and the highest-density liquid-cooled facility you personally accepted into service.
2. 49 words maximum. Describe a coolant-quality defect whose true source crossed more than one vendor boundary.
3. 49 words maximum. Which transient test would you refuse to omit before releasing an occupied accelerator row?