GPU Cloud Capacity-Clearing Leader

The mandate

Commercial teams sold accelerator capacity as interchangeable monthly units even though customer workloads require specific memory, interconnect, locality, software and start-date combinations. Several launches then competed for the same viable clusters while other reserved supply sat unusable. The capacity executive departed after provider invoices, scheduler allocations and customer commitments could not be reconciled into one defensible position.

The interim must join in Mumbai within two weeks and lead for fifteen months, with a possible three-month extension only if a strategic provider delays an approved transfer. A permanent capacity-market search begins after the clearing model survives two quarterly contracting cycles, anticipated in month eight. The successor will overlap for six weeks and chair the last supplier allocation and customer exception forum.

Handover requires every sale and reservation to map to a deliverable accelerator shape, network, software image, geography, availability window and fallback class; supply positions to reconcile monthly to provider and customer obligations; allocation exceptions to expire visibly; two scarcity exercises to pass; and the permanent leader to accept the exposure, unit-economics and relationship registers.

The interim may suspend new commitments, reallocate uncontracted supply, set clearing priorities, invoke existing provider options, approve customer remedies within policy and commit up to ₹60 crore inside the authorised recovery. New multiyear provider contracts, permanent hiring, equity commitments, customer settlements above delegated limits and decisions exceeding ₹12 crore require committee consent. The seat cannot alter content policy or model licensing.

Creative-product roadmaps, model training design, general public-cloud migration and consumer pricing outside capacity-linked terms are explicitly out of scope. The mandate owns accelerator supply classification, commitment control, allocation clearing, exception governance, service evidence, provider performance and the operating team that connects commercial promise to physically suitable capacity.

Why this seat is open

Overselling occurred because sales, procurement and scheduling counted different definitions of available compute, and the leadership departure removed the arbiter between them. Launch failures now threaten strategic creators while unused reservations continue to consume cash. An interim executive must reset commitments, operate the clearing discipline through real scarcity and leave a permanent leader with credible commercial control.

What you will own

• Reconcile provider reservations, owned clusters, scheduler inventory, customer contracts, launch calendars, fallback rights and stranded capacity.
• Define fungibility classes spanning accelerator generation, memory, topology, fabric, storage, locality, software stack, reliability and activation date.
• Establish a clearing mechanism that ranks committed rights, workload fit, customer consequence, margin, portability and recovery options.
• Decide remediation for unsupported promises through substitution, staged activation, brokered capacity, workload movement or negotiated service relief.
• Install commitment gates requiring evidence of deliverable shape, operating headroom, provider obligation, unit economics and tested fallback.
• Command scarcity exercises involving provider delay, cluster failure, fabric degradation, demand surge and simultaneous customer launch obligations.
• Transfer supply books, exposure, clearing rules, exception history, provider levers and customer decisions through successor-led allocation councils.

Candidate qualifications

• Held executive responsibility for scarce GPU, HPC or specialised cloud capacity sold or allocated across multiple customers.
• Reconciled non-fungible infrastructure supply with commercial commitments, scheduler reality and provider contracts under shortage.
• Recovered an oversold or misclassified capacity position without masking unsupported promises inside aggregate utilisation.
• Built allocation mechanisms incorporating hardware shape, fabric, locality, software, timing, fallback and customer consequence.
• Negotiated remedies with strategic customers and concentrated infrastructure providers while preserving future supply options.
• Handed a capacity marketplace or allocation function to permanent leadership after proving controls through live contracting cycles.

Non-negotiables

• Can start in Mumbai within two weeks and travel monthly to Asian provider, customer and operating hubs.
• Will accept an exclusive executive commitment and continuous escalation availability for launch-critical capacity decisions.
• Brings direct accelerator-capacity commercial operations; cloud procurement or model engineering alone is insufficient.
• Must disclose current relationships with GPU providers, clouds, brokers, model platforms, customers and infrastructure investors.

1. 49 words maximum. State your earliest Mumbai start and the largest specialised-compute shortage you have personally cleared.
2. 49 words maximum. Describe two accelerator reservations that looked equivalent financially but could not serve the same workload.
3. 49 words maximum. Which evidence would make you stop selling capacity despite unused devices appearing in a provider dashboard?