Sarvam AI Unicorn

"Build, deploy, and run AI with full control, developed and operated entirely in India" — models, data and infrastructure never leave Indian jurisdiction.

Industry-leading text-to-speech, speech-to-text, translation and document-digitisation models built specifically for India's languages, culture and context (11-23 Indic languages depending on product).

Ambition to build the "full-stack" for Generative AI — from research-led innovation in training custom AI models to an enterprise-grade platform for authoring and deployment.

Forward-deployed engineers work alongside client teams to design, integrate and continuously optimise agents — "not a handoff, a partnership" — backed by SLA production support.

Banking, insurance and financial-services companies deploying Sarvam's multilingual voice agents and conversational platform.

Central government (selected under the IndiaAI Mission to build India's sovereign LLM) and state governments (e.g. Tamil Nadu's Sovereign AI Park); used for citizen-facing programs like Ministry of Agriculture farmer feedback and policy renewal campaigns.

Developers building Indic-language AI into their own products via Sarvam's REST APIs, Python SDK (pip install sarvamai) and browser Playground.

Tata Capital's Chief Digital Officer describes using Sarvam to scale multilingual voice interactions across consumer loan products.

"Dedicated engineer from day one, joint workflow design and build, ongoing accuracy and cost optimisation, SLA-backed production support" for enterprise/government accounts.

Developers interact through documentation, an API playground and SDK rather than a dedicated account team.

Customers can run Sarvam Cloud, Private Cloud (VPC) or fully on-premises/air-gapped, and can "swap vendors without rewriting workflows" — built to reduce lock-in concerns for regulated clients.

Dedicated engineers embedded with client teams from day one for joint workflow design, integration and go-live support across banking, insurance, gov-tech and defence.

REST API, Python SDK (pip install sarvamai) and an in-browser Playground let developers add Indic AI to their products without a sales conversation.

IndiaAI Mission (central government) and state MoUs (e.g. Tamil Nadu's Rs 10,000 crore Sovereign AI Park) serve as a distribution channel for population-scale deployment.

Continued research on training its next frontier model for agentic, coding, and cybersecurity use-cases, requiring access to compute at scale.

Engineers work alongside client teams designing and building agent workflows across banking, insurance, gov-tech and defence verticals.

Conversational platform handles 2M+ interactions/day (usage doubling in two months); developer inference platform processes 10M+ API calls daily (usage tripling in three months); speech models transcribe 500K+ hours of audio/month.

Sarvam 105B (matches/outperforms larger reasoning models on knowledge, reasoning and agentic benchmarks) and Sarvam 30B (optimised to run on consumer/edge hardware), both trained from scratch in India.

Co-founder Pratyush Kumar led AI4Bharat, India's open-source Indian-language AI research initiative; co-founder Vivek Raghavan brings a decade of experience building India's digital public infrastructure (Aadhaar/UIDAI).

Secured 4,096 Nvidia H100 SXM GPUs through cloud/data-center partner Yotta Data Services under the IndiaAI Mission.

$150M lead strategic investor in the Series B; combines its enterprise-transformation expertise, global client relationships and engineering depth with Sarvam's AI research to build a sovereign AI ecosystem.

Selected Sarvam to build India's sovereign LLM and provided subsidised compute (4,096 H100 GPUs) in exchange for CCDs expected to convert into an equity stake.

Rs 10,000 crore MoU to build India's first full-stack Sovereign AI Park in Chennai, with Sarvam anchoring the technology stack.

Deployed Sarvam's multilingual voice agents across consumer loan products; featured as a named customer case study.

Supplied the 4,096 Nvidia H100 SXM GPUs Sarvam uses under the IndiaAI Mission compute allocation.

Paid deployments of conversational (Samvaad), content (Studio), document-digitisation (Akshar) and agentic (Arya) platforms, delivered via Sarvam Cloud, Private Cloud (VPC) or on-premises.

Developer-facing inference platform ("token factory") serving 10B+ tokens with <100ms median latency and a 99.9% uptime SLA — usage-metered access to Sarvam's models.

Compute-heavy cost base: a single six-month IndiaAI Mission allocation of 4,096 Nvidia H100 GPUs carried a total compute bill of Rs 246.71 crore (government subsidised Rs 98.68 crore of it).