Sarvam AI signs deal to develop AI systems for Indian military

Sarvam AI has signed a memorandum of understanding (MoU) with Headquarters Integrated Defence Staff (HQ IDS) to establish an umbrella framework for AI projects across the Indian Army, Navy and Air Force. The company announced the agreement on October 27.

Sarvam issued a clarification on October 28 following reports that the agreement had been signed directly with the Ministry of Defence. The MoU was signed with HQ IDS, which Sarvam identifies as an organisation under the Ministry. However, Sarvam’s post continues to refer to the Ministry as the parent body for the agreement.

What the agreement covers

Scope: AI projects across the three Services, as well as joint, tri-service and integrated organisations under HQ IDS.
Work: Sarvam will design, develop and deploy AI applications for operational, training and decision-support needs and provide follow-on technical support.
Structure: The MoU covers several projects across all three Services, with HQ IDS facilitating the engagement.
Shared stack: The Services and joint organisations would use the same foundation models, agentic infrastructure, and geospatial and perception systems. Sarvam says lessons from one Service’s projects can then inform projects for the others.

Chanakya, a firewalled division dedicated to national security and defence, will oversee Sarvam’s defence work. It will leverage Sarvam’s AI stack, including foundation models such as Sarvam 105B.

Stated direction and focus areas

Sarvam states that the MoU supports the Ministry’s strategy for military AI. It will enable the armed forces to develop an AI layer across the Services with Indian partners, ensuring that solutions are built in India and deployed on sovereign infrastructure. Sarvam lists the following focus areas:

Decision support
Target acquisition and data fusion across the sensor-to-shooter loop
Human oversight retained on operational decisions

Sarvam says it will collaborate with HQ IDS to integrate elements of India’s defence capabilities, referring to this as an “AI fabric”. However, the post does not provide technical details about the concept.

What the announcement does not say

Sarvam’s post leaves out several details:

The signing date and the names of the signatories
The contract value, funding source or duration
Which specific projects are under way and which Service each project serves
Whether any systems have already been deployed
How “human oversight” will be defined or enforced
Any statement from HQ IDS or the Ministry

Neither HQ IDS nor the Ministry has independently confirmed the details. Therefore, the information above is based on Sarvam’s account.

Sarvam’s sovereign AI push

The MoU follows Sarvam’s June 2026 funding round, which established the company as India’s newest AI unicorn. Sarvam raised $234 million in the first close of a $300 million Series B round at a valuation of $1.5 billion. HCLTech led the round with a $150 million investment for a 10.46% stake (Rs 1,427.25 crore). The timeline for closing the remaining $66 million has not been disclosed.

Sarvam will use the funds to develop a next-generation frontier model for agentic AI, coding and cybersecurity, and to expand its secure computing capacity.

Sarvam reported revenue of Rs 45.1 crore in 2025-26, following an earlier valuation of $196 million in 2025. The company’s government partnerships include the IndiaAI Mission, which selected Sarvam in April 2025 to build India’s first sovereign large language model. The mission now funds Sarvam as one of 12 organisations developing foundation models.

In January 2026, Sarvam signed MoUs with Tamil Nadu and Odisha for sovereign AI infrastructure. The Tamil Nadu agreement covers a research park with IIT Madras, while the Odisha agreement covers a 50-megawatt computing facility.

The company initially faced scepticism over its pace of development and download numbers. Critics also noted that its May 2025 model, Sarvam-M, was a fine-tuned version of a foreign foundation model. In response, Sarvam launched models in February 2026 that were trained from scratch on Indian-language data. These included Sarvam 105B, which the company referenced in its announcement about the HQ IDS agreement.