Sarvam Signs a Five-Year AI Pact With HQ IDS to Build an 'AI Fabric' for the Army, Navy and Air Force
Sarvam has signed a five-year MoU with Headquarters Integrated Defence Staff to design, develop and deploy AI applications for India's three armed services. The work will run through its firewalled defence unit, Chanakya, on the Sarvam 105B model and sovereign infrastructure.

Sarvam, the Bengaluru company building India's sovereign large language models, has signed a five-year memorandum of understanding with Headquarters Integrated Defence Staff (HQ IDS). Under the deal, Sarvam will develop and deploy artificial-intelligence applications for all three armed services. HQ IDS announced the MoU on 26 October, and Sarvam made it public on 27 October.
The framework puts Indian models and infrastructure at the centre of the proposed projects. It does not establish that operational systems have already been fielded.
Who signed with whom
Early reports described the agreement as a deal with the Ministry of Defence. On 28 October, Sarvam issued a clarification: the MoU was executed with HQ IDS, which it described as an organisation under the ministry, not with the ministry itself.
The distinction matters. HQ IDS is the tri-service headquarters that coordinates joint planning and capability development across the Army, Navy and Air Force. A framework agreement signed there can be used by all three services and by the joint organisations under HQ IDS. A deal with a single service could not.
What the MoU covers
According to the company's announcement and reporting by ThePrint, Business Standard and others, the MoU sets up a common framework under which Sarvam will:
- design, develop and deploy AI applications for operational, training and decision-support needs across the three services;
- provide follow-on technical support after systems are deployed; and
- work on projects including an "AI fabric", a shared layer meant to connect different parts of India's defence capabilities.
Reporting by The Week and others lists target acquisition and sensor-to-shooter data fusion among the possible applications, with humans kept in charge of operational decisions. ThePrint reported that the services would get access to a shared stack of foundation models, agentic AI infrastructure, and geospatial and perception systems.
The parties have not disclosed the agreement's value, the first projects, or a timeline for operational deployment. An MoU is a framework, not a procurement contract. Individual projects will still need their own approvals and orders.
Chanakya: Sarvam's firewalled defence unit
The work will be handled by Chanakya, Sarvam's separate, firewalled division for national security and defence. ThePrint reported that the unit was announced in March and focuses on air-gapped, on-premise deployments, meaning systems that run inside the military's own networks with no connection to the public internet.
Chanakya will use Sarvam's own technology stack, including the Sarvam 105B foundation model. The stated aim is an AI layer for the forces built with Indian partners and run on sovereign infrastructure.
The MoU builds on earlier work. ThePrint reported that in August, Lt Gen Manish Erry launched an upgraded Sarvam-built platform at the Defence Services Staff College (DSSC), with Sarvam co-founder Vivek Raghavan present, and the two sides signed an MoU for further development. Sarvam has described a system it built for a military training institution that runs on a GPU server inside the institution's own network. It says the system was certified by HQ IDS and the Defence Cyber Agency after security audits.
Why sovereignty matters for military AI
Militaries are cautious about AI they cannot inspect. A model hosted by a foreign cloud provider, or one whose weights and training data are opaque, raises obvious questions about where data goes, who can switch the service off, and how the system behaves under pressure. Models trained and hosted in India, running air-gapped on government hardware, avoid many of those concerns.
What it means for Sarvam and the ecosystem
For Sarvam, the MoU adds the national security market to its government and enterprise work, which already includes Indian-language speech and text models and sovereign data-centre projects with state governments. A five-year tri-service framework gives a start-up something it rarely gets: a long-horizon customer relationship with India's most demanding user.
What to watch
The tests are concrete. Will HQ IDS name the first projects? Will any of them move from pilots into funded procurement? And how will the services write rules for human oversight of AI-assisted targeting? The MoU gives the forces a framework. Results in the field will decide whether the "AI fabric" becomes real.
Sources
- https://www.sarvam.ai/blogs/sarvam-hq-ids-mou
- https://theprint.in/defence/design-develop-deploy-sarvam-signs-5-year-mou-with-mod-to-build-ai-for-army-navy-iaf/3083107/