Accelerated-Compute Portfolio Economics Board Adviser

The mandate

The board repeatedly debates which institutions deserve scarce accelerated compute, what users should pay and when older clusters should be renewed. Each research centre reports different utilisation, publications, models and national-benefit measures; reserved cloud capacity and owned systems sit in separate cases; and political visibility sometimes outweighs the evidence that a workload requires the requested architecture. The standing question is how to govern one portfolio without pretending unlike missions are interchangeable.

The adviser will contribute three working days each month, hold a monthly challenge session with the portfolio office, attend four research-capital committee meetings and visit one funded programme each quarter. Time-sensitive questions on a reservation or renewal receive a response within two business days; routine papers receive a five-day review. The retainer covers preparation, committee attendance and domestic visits.

The term is twelve months and finishes without automatic continuation. In month eleven, the board chair may recommend a fresh mandate if a new national compute programme has been formally approved, but renewal requires full-board consent and new conflict review. Deferred meetings, unused days or an unfinished procurement cannot roll the current appointment forward.

The adviser holds no line authority and has no executive responsibility for cluster operation, research selection, grant award, procurement or contract signature. Institute executives remain accountable for delivery and the committee decides allocations. The adviser may question metrics, test economic cases and recommend portfolio rules, but cannot award capacity, rank individual scientists or negotiate suppliers.

The role may sit alongside no more than three other material commitments. Current work for an accelerator manufacturer, hyperscale provider, capacity broker, funded institute, university applicant, research cloud or investor in relevant infrastructure must be disclosed and could preclude participation in a decision. Referral fees, reseller margin and compensation linked to procurement or grant outcomes are not permitted.

Why the board wants this voice

The board has scientific and financial members, yet lacks a leader who has governed accelerated compute as a changing portfolio rather than a succession of hardware purchases. Existing dashboards reward allocation and nominal utilisation without exposing blocked science, unsuitable architecture, failed work or long-term concentration. An independent economics voice can help directors make defensible trade-offs across public value, capability and cost.

What you will own

• Press institutions to state which scientific decision requires each accelerator class, interconnect, memory profile, data locality and service window.
• Test portfolio measures across useful work, queue delay, failed computation, research outcome, capability development, energy and fully burdened cost.
• Challenge subsidy and chargeback designs for gaming, exclusion of early research, administrative burden and incentives to retain idle allocations.
• Examine owned, reserved, elastic and shared capacity choices under demand uncertainty, technology obsolescence, software lock-in and supply concentration.
• Shape renewal gates that distinguish hardware age from economic usefulness, unmet capability and credible future workload.
• Probe cross-institution allocation rules for conflicts, priority overrides, emergency access, unused reservations, reproducibility and appeal.
• Frame an annual board portfolio view with allocation principles, disputed assumptions, concentration exposure and decisions requiring explicit public-value judgment.

Candidate qualifications

• Governed a multi-institution GPU, HPC or advanced-research infrastructure portfolio with material capital and operating commitments.
• Designed allocation or chargeback mechanisms that balanced early research, strategic missions, scarce capability and evidence of productive use.
• Compared owned and external accelerated capacity including utilisation quality, software, data movement, support and obsolescence.
• Challenged a high-profile compute request that lacked architecture or workload evidence and preserved stakeholder trust after refusal.
• Reported research-infrastructure value beyond publications, device occupancy or headline model size to a public or mission-led board.
• Remained independent while navigating concentrated provider, academic, government and investor relationships around advanced compute.

Non-negotiables

• Can meet the three-day monthly commitment, four committee dates and quarterly UAE site visits for the full year.
• Will disclose vendor, institute, university, cloud, broker, investor and grant relationships before accessing allocation proposals.
• Accepts that capacity, grant, procurement and research decisions remain wholly with executives and the board.
• Brings portfolio-level compute economics; single-cluster engineering or conventional technology-budget advice alone is insufficient.

1. 49 words maximum. Which measure best exposed an institution that occupied accelerators without producing proportionate useful work?
2. 49 words maximum. Name every current provider, institute, university or investment relationship relevant to this portfolio.
3. 49 words maximum. Confirm the Dubai cadence and describe one allocation rule that protected exploratory research without enabling capacity hoarding.