Sovereign Arabic Model Compute and Talent Leader

The mandate

The appointed programme head withdrew for personal reasons six weeks before acceptance testing of a new accelerator cluster and the first international research hires. Infrastructure delivery, Arabic data partnerships and model teams now report through separate executives, leaving no one authorised to integrate compute reliability, research priorities, talent formation and controlled model release.

The interim must relocate to Riyadh within four weeks and lead a twenty-four-month fixed build. A permanent international search launches after the first governed model release and two stable training cycles, expected near month seventeen, with eight weeks reserved for the successor to chair capacity allocation, evidence review and a partner-governance session.

Handover requires the cluster to complete two reproducible distributed training runs inside signed reliability and cost corridors, Arabic data and evaluation assets to carry traceable permissions and regional coverage, four technical directors to operate independently, one model release to pass agreed safety and usefulness gates, and the successor to accept the following annual compute-and-research plan.

The interim may allocate sanctioned compute, stop training or release, define team topology, approve hiring within authorised bands, appoint temporary technical leads and move up to SAR 90 million within the programme envelope. New strategic compute awards, capital above SAR 35 million, public model claims, permanent executives and changes to national data policy require council approval; the role cannot promise benchmark leadership.

National AI policy, unrelated government digitisation, commercial cloud sales and domain-specific deployment products are outside scope. The seat owns the Arabic foundation-model systems, their compute and data operations, evaluation and release mechanics, technical leadership bench and research-partner interfaces, without displacing accountable policy, security or application authorities.

Why this seat is open

The programme’s chosen head withdrew after critical infrastructure and hiring commitments were already irreversible. Allowing each workstream to optimise independently would make cluster acceptance look like programme readiness while data, talent and release decisions remain disconnected. A temporary executive must build one functioning institution before permanent leadership is selected for its longer research horizon.

What you will own

• Convert the programme ambition into bounded model, data, evaluation, compute and talent outcomes with explicit owners and expiring assumptions.
• Accept or reject cluster readiness through distributed-training reliability, utilisation, checkpoint recovery, networking, storage, observability and cost evidence.
• Allocate accelerator capacity among pretraining, adaptation, evaluation and research experiments using scientific value, evidence readiness and opportunity cost.
• Establish Arabic corpus and evaluation governance across regional language variation, source rights, cultural review, contamination and controlled refresh.
• Build the leadership and skills architecture through targeted global hiring, local development, university pathways and time-limited external expertise.
• Govern model promotion through capability, safety, usefulness, serving, documentation, red-team and rollback gates before any public or partner release.
• Transfer the compute plan, model registry, data rights, evaluation assets, partner commitments and talent map through three successor-led councils.

Candidate qualifications

• Built or led a large foundation-model research and engineering platform spanning distributed training, data, evaluation and production serving.
• Accepted high-value accelerator infrastructure against measured training reliability, recovery, utilisation and cost rather than installed hardware counts.
• Governed Arabic or another morphologically and regionally varied language through corpus design, evaluation and release evidence.
• Recruited senior research and systems leaders internationally while building durable local technical capability and succession.
• Held authority to stop a training run or model release when data, safety, reliability or evidence conditions failed.
• Transferred a multi-year AI institution or platform to permanent leadership with repeatable allocation and governance routines.

Non-negotiables

• Can relocate to Riyadh within four weeks, commit exclusively for twenty-four months and travel monthly to the designated Saudi partner site.
• Will not claim sovereign control, Arabic leadership or release readiness beyond the rights, infrastructure and evaluation evidence actually held.
• Brings direct foundation-model systems leadership; conventional data-centre, enterprise AI or academic research management alone is insufficient.
• Must disclose current model-lab, cloud, accelerator, university and data-provider relationships before programme details are shared.

1. 49 words maximum. State your earliest Riyadh relocation date and any obligation incompatible with a two-year exclusive executive assignment.
2. 49 words maximum. Describe one distributed training platform you accepted or rejected and the recovery evidence that decided it.
3. 49 words maximum. Which corpus, evaluation or talent condition would stop a planned Arabic model release despite available compute?