Gujarat rolls out statewide GPU-as-a-Service pool for government and AI startups

GANDHINAGAR: Gujarat has switched on a statewide GPU-as-a-Service framework, putting shared graphics processors on tap for government offices, public campuses and a slice of the state’s AI startup pipeline. The Science and Technology Department decision took effect on 23 September 2026 and names Gujarat Informatics Limited as the operator of the pooled capacity.

Chief Minister Bhupendra Patel’s office cast the move as infrastructure for the state’s AI Action Plan, not a one-off lab purchase. The plan’s six pillars cover data, infrastructure, capacity building, foundational research and use cases, deeptech startup facilitation, and safe and trusted AI. GPUaaS is meant to lock in the infrastructure pillar with a managed service instead of leaving every department to buy its own accelerators.

Who can draw on the shared GPUs

Eligible state government departments, government institutions and government educational institutions are to get GPU access on a fully subsidised basis. Startups selected under the AI Innovation Challenge at the AI Centre of Excellence also qualify for 100 per cent subsidy when they take an approved use case into production.

Other AI startups may still request capacity if spare GPUs remain, but they will pay the actual GPU service cost plus GIL’s consultancy fee. Autonomous bodies, boards, corporations and other organisations cleared from time to time by the Science and Technology Department can also be onboarded for approved AI and data projects.

The political framing is practical: AI projects stall when each secretariat waits for its own tender for high-end cards. A shared pool lets a health analytics pilot and a campus research group book time on the same managed fabric, subject to feasibility checks.

How GIL will run allocation and cost

GIL will handle provisioning, operations, monitoring and day-to-day management. Allocation is supposed to weigh technical feasibility, available stock, approved use cases, priority and existing government policy. Departments may use the GPUs only for cleared purposes and must follow information security, privacy, cybersecurity and responsible-AI rules.

On commercial structure, GIL can charge a 7 per cent consultancy fee on the actual cost of GPU services bought from a cloud provider empanelled under IndiaAI. Detailed eligibility, application, charging and monitoring guidelines are to be issued periodically with Science and Technology Department approval.

Science and Technology Minister Arjun Modhwadia is listed as steering the framework under the Chief Minister’s guidance. The brief does not publish a public inventory of how many GPUs are live on day one or which cloud SKUs sit behind the pool. Those numbers will matter when multiple departments queue for training jobs at once.

For Indian readers watching state AI plans, the signal is that Gujarat is trying to turn policy text into a booking desk. Shared, subsidised compute will only move the needle if GIL’s queue rules stay transparent and if subsidised startup seats are not swallowed by low-value demos. Shared GPU pools are becoming a recurring theme in state digital plans, but Gujarat’s brief is unusually explicit on subsidy tiers. Fully funded seats for departments and campuses remove the usual excuse that a pilot cannot start until hardware lands. At the same time, the paid lane for non-selected startups keeps the pool from becoming an unlimited free draw.

The IndiaAI empanelment link also matters. By routing cloud GPU purchase through a nationally listed provider and letting GIL add a fixed consultancy slice, the state is trying to avoid a patchwork of one-off cloud contracts. Whether that lowers unit cost will show up only after the first quarter of invoices.

Residents will not feel GPUaaS directly. They will feel it if a transport department model clears congestion faster, or if a campus lab stops waiting months for a single training box. The administrative test is simpler: publish the application form, show queue times, and keep subsidised startup seats tied to working products rather than slide decks.