Zum Inhalt springen
Zugang anfragen

Dubai (DIFC)

Head of Compute Infrastructure & On-Prem Fleet

Owns the physical compute substrate the entire AI workforce runs on — the on-prem Apple Silicon fleet (~50 Mac Studios and Mac Minis today, scaling into the hundreds) that hosts the agent swarms and local models. Spans the atoms (racking, power, cooling, network, hardware lifecycle), the fleet plane (MDM, zero-touch provisioning, fleet-wide OS and security updates), and local-inference operations (distributed model serving, capacity and compute strategy). This is the one role that touches hardware, and — critically — the human accountable for recovering the fleet when the agents that would otherwise self-heal are themselves offline.

Reports to COO

Why this role exists.

At a glance
  • TeamCompute Infrastructure
  • HubDubai (DIFC)
  • Reports toCOO
  • ArchetypeArchitectDesigns the systems everything else is composed from.
  • Hub focus
    • DIFC (UAE) — Global HQ
    • Governance Kernel
    • Corporate Centre

Most AI companies rent their compute and never see it. Adaptic runs a growing on-prem Apple Silicon fleet — the physical machines our agent workforce and local models actually execute on — and that fleet needs a human owner. You will own it end to end: the atoms (racking and shelving Apple Silicon, power, cooling, structured cabling, physical security, hardware RMA and lifecycle), the fleet plane (MDM with Jamf/Mosyle/Addigy, zero-touch provisioning, fleet-wide OS and security update orchestration, on-device identity and secrets, monitoring), and local-inference operations (standing up and operating distributed model serving across the cluster, model distribution, utilisation, and the capacity plan as the fleet scales). The role exists for a reason no amount of automation removes: the agent workforce runs ON this substrate, so the substrate cannot be maintained solely by the agents — you are the human outside the dependency loop who recovers it when they can't. Apple Silicon makes this sharper, not softer: no out-of-band management, no ECC, consumer warranty paths. You will also own the compute strategy — when on-prem Apple Silicon still wins on perf-per-watt and cost, and when to move to colocation or managed hardware. This is a build-from-bare-metal mandate for someone who wants to own the physical foundation of an autonomous enterprise.

What the first year looks like.

Concrete milestones your interview panel will evaluate against.

  • First 30 days
    • Inventory and document the existing fleet — every machine, its role, power draw, network position, and health
    • Establish a single source of truth for asset, warranty, and lifecycle tracking
    • Assess physical environment (power headroom, cooling, network capacity, physical security) and flag immediate risks
    • Build the recovery runbook for fleet-down scenarios, validated independently of the agent platform
  • 60-90 days
    • Bring the fleet under MDM with zero-touch provisioning and fleet-wide, staged OS/security update orchestration
    • Instrument fleet-level observability (node health, thermals, utilisation, inference throughput) with alerting
    • Harden power and network (UPS, redundancy, structured cabling) and document the physical topology
  • 6 months
    • Deliver a documented capacity plan and a costed make-vs-buy/colocation analysis for the next 12 months of growth
    • Achieve a target node-recovery time and a tested bus-factor mitigation (cross-trained backup or contracted smart-hands)
    • Operate distributed local-inference serving across the cluster at defined utilisation and reliability targets
  • 12 months
    • Scale the fleet to plan with no unplanned capacity-driven outages
    • Pass a physical-infrastructure and security-controls review
    • Establish the compute roadmap (Apple Silicon vs colocation vs managed hardware) the firm executes against

The experience we cannot coach.

The bar is specific. If you are close on required but strong on preferred, apply anyway. We read every application.

Required
  • 8+ years in data-centre, on-prem, or large-fleet hardware operations, including hands-on physical work (racking, power, structured cabling, hardware diagnostics and RMA)
  • Proven ownership of a managed device fleet at scale, ideally Apple Silicon, with deep MDM experience (Jamf, Mosyle, or Addigy) and zero-touch provisioning
  • Strong networking fundamentals (switching, VLANs, firewalling, structured cabling) and physical-environment knowledge (power, UPS, cooling)
  • Experience operating compute for ML/AI workloads, including local LLM/model serving and GPU/accelerator utilisation
  • Capacity planning and make-vs-buy/colocation analysis for owned compute
  • A recovery-first mindset: you design for the case where the automation that manages the fleet is itself down
Preferred
  • Experience standing up distributed inference on Apple Silicon (MLX, EXO-style clusters) or comparable accelerator fleets
  • Background bridging physical infrastructure and software-defined/cloud operations
  • Familiarity with colocation procurement and smart-hands/remote-hands operating models
  • Experience designing physical security and access controls for sensitive compute

We are hiring the people who govern the machine.

People write the mandate and set each account’s trading policy. A model makes each trading call inside it, the kernel checks every order against that policy before it reaches the broker, and every decision is recorded. We build that operating system for institutional asset management.

Where we hire
  • Kernel & policy engineering
  • Agent research
  • Risk, compliance & controls
  • Fund operations & close
  • Product design & interpretability
The operating model

The principles every role at Adaptic works to.

  • Senior leaders design playbooks; agents and partners execute them with approvals, telemetry and QA.
  • We are control‑first: policies, registers and evidence are audit‑ready at all times.
  • We prefer metrics over headcount—you'll own outcomes and a dashboard, not a large team.
  • Sensitive actions are hard‑gated by policy engines; we maintain human‑in‑the‑loop for regulated steps.
  • Incidents are handled with blameless post‑mortems, action tracking and time‑boxed remediation.

What we will ask you.

The application form asks these questions. They are what your interview panel already wants to hear about.

  1. The agent workforce that would normally self-heal infrastructure runs on the very fleet you maintain. Walk through how you design fleet recovery so that a fleet-down event is resolvable by a human operating entirely outside the agent platform's dependencies — what runs out-of-band, what you keep on paper, and how you test it.
  2. We run ~50 Apple Silicon machines today and expect to scale into the hundreds. In your first 90 days, what would you instrument to measure fleet health and inference capacity, and at what thresholds would you trigger the make-vs-buy / colocation decision?

If this is your work, apply.

Applying takes a résumé, a cover letter and your answers to the role’s screening questions. We read every application.

Not the right role? Write to careers@adaptic.ai with what you would build.