Sarvam AI

Sarvam AI is an Indian artificial intelligence company founded in August 2023 by Vivek Raghavan and Pratyush Kumar, both former senior researchers at the IIT Madras-based AI4Bharat research initiative. The company is headquartered in Bangalore and develops the Sarvam family of large language models, the Indus 105-billion-parameter open-weights model, and full-stack sovereign-AI tooling spanning speech-to-text, text-to-speech, translation, and conversational agents across 22 Indian languages. In April 2025, Sarvam was selected by the Government of India under the IndiaAI Mission to build India's sovereign large language model. As of March 2026, the company is in talks for a $250 million funding round at a reported $1.5 billion valuation, led by NVIDIA, Accel, and HCLTech.

At a glance

• Founded: August 2023 in Bangalore, India, by Vivek Raghavan and Pratyush Kumar.
• Status: Private. India's principal sovereign-AI research startup.
• Funding: Approximately $71 million raised in early rounds (including a $30 million Series A in 2024 led by Lightspeed Venture Partners and Khosla Ventures). $250 million round in process as of March 2026 at a reported $1.5 billion valuation, led by NVIDIA, Accel, and HCLTech.
• CEO: Vivek Raghavan (co-founder; previously contributed to India's digital public infrastructure and to AI4Bharat at IIT Madras).
• Other notable leadership: Pratyush Kumar (co-founder; previously led India's open-source AI efforts on Indian languages at AI4Bharat).
• Open weights: Yes. The Sarvam family is released open-weights through Hugging Face. The Indus 105-billion-parameter model launched as open-weights in February 2026.
• Flagship models: Sarvam 105B / Indus (February 2026, 105-billion-parameter open-weights model), Sarvam 30B (February 2026, 30-billion-parameter open-weights), Sarvam multilingual conversation and speech model line.

Origins

Sarvam AI was founded in August 2023 by Vivek Raghavan and Pratyush Kumar. Both founders had been senior figures at the AI4Bharat research initiative at the Indian Institute of Technology Madras, one of India's principal academic research institutions. AI4Bharat had been the locus of open-source Indic-language AI research in India through 2018 to 2023, contributing speech, translation, and language-modeling research focused on India's 22 official languages. Raghavan had earlier been a senior contributor to India's digital public infrastructure (the technology stack underpinning Aadhaar, the Unified Payments Interface, and the broader DigiLocker ecosystem), bringing public-sector technology and government-engagement credentials to the company's founding.

The 2024 trajectory established Sarvam's research and commercial credentials. The Series A round of $30 million in February 2024 was led by Lightspeed Venture Partners with Khosla Ventures, Peak XV Partners, and other participants. The 2024 release sequence included multiple Sarvam model variants spanning 1.3-billion to 22-billion parameter scales, focused on Indic-language capability and speech, translation, and conversational-AI tooling. The company also joined the Nemotron Coalition for collaboration on open-weights model training infrastructure.

In April 2025, the Government of India selected Sarvam under the IndiaAI Mission to build India's sovereign large language model. The selection provided dedicated compute resources and government-funded model-development support, framed as an indigenous foundational-model effort distinct from the wider open-weights international ecosystem. The selection placed Sarvam in a structurally distinctive position as the principal Indian-government-aligned AI research effort.

The February 2026 release of Sarvam 30B and Sarvam 105B (also branded as Indus, in beta from February 20, 2026) was the most consequential model launch from any Indian AI lab to date. Both models were trained from scratch on datasets focused on Indian languages, with capability across reasoning, multilingual tasks, mathematics, and coding. The 105-billion-parameter scale represents the largest open-weights model trained primarily for Indic-language coverage globally.

In March 2026, reports emerged that Sarvam was in talks for a $250 million funding round led by NVIDIA, Accel, and HCLTech at a reported $1.5 billion valuation. The round, if closed, would make Sarvam India's second AI unicorn after Krutrim, with a structurally distinctive investor mix combining frontier compute infrastructure (NVIDIA), top-tier US venture capital (Accel), and Indian IT-services scale (HCLTech).

Mission and strategy

Sarvam AI's stated mission is to build a "full-stack sovereign AI platform" for India, with capabilities spanning speech-to-text, text-to-speech, translation, conversational agents, and large language models across 22 Indian languages. Vivek Raghavan and Pratyush Kumar have framed the strategy publicly as the development of indigenous AI capability in alignment with India's "Atmanirbhar Bharat" (self-reliant India) policy framing.

The strategy combines four threads. First, foundation-model research with explicit support for 22 Indian languages, building on AI4Bharat's open-source Indic-language research foundation. Second, full-stack speech and language tooling (STT, TTS, translation, conversational agents) optimized for Indian-language and Indian-cultural contexts. Third, government and enterprise distribution through the IndiaAI Mission selection and through partnerships with Indian and international enterprises. Fourth, open-weights distribution as a structural commitment, with the February 2026 Sarvam 30B and 105B releases as the principal demonstration.

The competitive premise is that India's combination of large multilingual population, growing digital infrastructure, and sovereign-AI policy interest creates a structural opening for an Indian-domiciled research-led AI provider with deep Indic-language capability. The company's research-first founder team contrasts with Krutrim's founder-driven commercial strategy, and the IndiaAI Mission selection provides a structurally distinctive government-engagement position.

The "frugal AI" framing in industry coverage (most prominently in Rest of World's 2026 piece) characterizes both Sarvam and Krutrim as building meaningful AI capability with capital expenditure substantially below US frontier-lab equivalents. For Sarvam, the frugal-AI capability is paired with research-led founder credibility and government partnership.

Models and products

• Sarvam 105B / Indus. Released February 20, 2026 (beta) as the company's largest open-weights model. 105 billion parameters, trained from scratch on Indic-language-focused datasets. Supports reasoning, multilingual tasks, mathematics, and coding.
• Sarvam 30B. Released February 2026. 30-billion-parameter open-weights model.
• Sarvam family of smaller models. 1.3-billion, 2-billion, and 22-billion-parameter variants released through 2024 and 2025.
• Sarvam Speech and Language tooling. STT (speech-to-text), TTS (text-to-speech), and translation models specialized for 22 Indian languages.
• Conversational-agent platform. Voice-and-text conversational-AI tooling for Indian-language deployment in enterprise and government applications.
• Samvaad. Sarvam's conversational-AI product brand.

The principal distribution channels are Hugging Face for open-weights releases, the Sarvam developer platform for paying customers, and government-aligned distribution through the IndiaAI Mission engagement.

Benchmarks and standing

Sarvam's benchmark profile emphasizes Indic-language evaluation rather than the standardized English-language leaderboards. Industry coverage has characterized Sarvam's Indian-language capability as among the leading Indic-language model implementations, alongside Krutrim, AI4Bharat (the IIT Madras research initiative), and international labs whose Indic-language fine-tunes are based on Llama or Qwen base models.

The Sarvam 105B / Indus release in February 2026 was characterized in Indian and international AI coverage as the most ambitious open-weights model trained primarily for Indian-language coverage. The model's standardized-benchmark performance has been less comprehensively evaluated than the leading frontier-tier releases, with public coverage focusing on Indic-language capability and speech-stack integration rather than English-language general-knowledge benchmark scores.

The company's standing in the global AI landscape rests on the founder-team's AI4Bharat research credentials, the IndiaAI Mission government selection, the open-weights distribution commitment, and the reported NVIDIA-led 2026 funding round. Industry coverage has frequently characterized Sarvam as the most credible Indian sovereign-AI research-led organization, distinct from Krutrim's founder-driven commercial-and-cloud strategy.

Leadership

As of April 2026, Sarvam AI's senior leadership includes:

• Vivek Raghavan, Chief Executive Officer and co-founder. Former senior contributor to India's digital public infrastructure (Aadhaar, UPI, DigiLocker ecosystem). Senior figure at AI4Bharat at IIT Madras prior to founding Sarvam. Public face for Sarvam on the sovereign-AI strategic vision and the IndiaAI Mission engagement.
• Pratyush Kumar, co-founder. Former lead of India's open-source AI efforts on Indian languages at AI4Bharat. Research-and-engineering leadership for the Sarvam model line and the Indic-language-specific capability.

The company has hired aggressively from AI4Bharat alumni networks, IIT Madras and other Indian research universities, and from Indian and international AI organizations. Specific senior research leadership beyond the two co-founders has been less broadly profiled in international media.

Funding and backers

Sarvam AI's funding history through April 2026 includes approximately $71 million in private rounds prior to the reported March 2026 round. The Series A of $30 million in 2024 was led by Lightspeed Venture Partners and Khosla Ventures, with Peak XV Partners and other participants. Subsequent rounds added approximately $40 million to the company's capital position before March 2026.

The reported March 2026 round of $250 million at a $1.5 billion valuation, with NVIDIA, Accel, and HCLTech as lead-investor candidates, would be the company's largest round to date and would establish Sarvam as India's second AI unicorn. The investor mix is structurally distinctive: NVIDIA's strategic-investor participation provides compute infrastructure access, Accel provides top-tier US venture capital, and HCLTech provides Indian IT-services scale and enterprise-distribution capability. As of April 2026, the round's closing terms had not been publicly confirmed.

Beyond private capital, the IndiaAI Mission selection provides government-funded compute resources and model-development support, an unusual public-private arrangement that complements the private-investor capital.

Industry position

Sarvam AI occupies a structurally distinctive position in the global sovereign-AI landscape. The combination of the AI4Bharat-founder research credibility, the IndiaAI Mission government partnership, the open-weights model distribution, the Sarvam 105B / Indus capability scale, and the reported $1.5 billion valuation produces a profile no other Indian AI organization matches and few global sovereign-AI organizations match across all dimensions.

Industry coverage in 2025 and 2026 has characterized Sarvam as the most credible Indian sovereign-AI research-led organization, with the frugal-AI capability framing as the principal differentiator from US-frontier-lab capital intensity. The combination of government engagement and private capital is widely viewed as evidence of a maturing Indian sovereign-AI policy.

Strategic risks include the operational complexity of running both a research-led foundation-model program and a full-stack speech and language tooling business, intensifying competition from Krutrim (the other Indian sovereign-AI vehicle) and from international open-weights labs, and the open question of whether Indic-language capability translates to commercial wins outside India. Strategic strengths include the founder-team research credentials, the IndiaAI Mission government partnership, the open-weights distribution commitment, the AI4Bharat research-and-data heritage, and the reported NVIDIA-led 2026 funding round.

Competitive landscape

Sarvam AI competes with several AI labs:

• Krutrim. Direct Indian sovereign-AI competitor. Krutrim's founder-driven full-stack commercial strategy contrasts with Sarvam's research-led foundation-model focus. Both target Indic-language sovereign-AI markets.
• AI4Bharat. IIT Madras-based academic research initiative producing open-source Indic-language AI models. Sarvam's founder team came from AI4Bharat; ongoing relationship combines collaboration and capability competition.
• OpenAI, Anthropic, Google DeepMind, Microsoft AI. US frontier-and-incumbent competitors operating in India through cloud and consumer-AI distribution.
• Aleph Alpha, Mistral AI, AI Singapore. Peer national-and-regional sovereign-AI organizations.
• Meta AI / FAIR. Llama family is open-weights and has been used as a base for Indic-language fine-tuning by multiple Indian organizations.
• DeepSeek, Alibaba Qwen. Chinese open-weights competitors. Indian developers have used Chinese open-weights models as alternatives to Sarvam.
• NVIDIA. Strategic investor and partner in the reported March 2026 round; also the principal compute-infrastructure provider for the model-training program.

Outlook

Several open questions affect Sarvam AI's trajectory in 2026 and 2027:

• The closing of the reported $250 million March 2026 funding round and the lead-investor identity confirmation.
• The release cadence of successor models in the Sarvam line, including potential Sarvam 200B+ scaled releases.
• The IndiaAI Mission engagement progression, including any expansion of government commitments and any India-government enterprise-AI deployment.
• The commercial trajectory of the Sarvam developer platform and the Samvaad conversational-AI product line.
• Continued senior-talent recruitment from AI4Bharat alumni networks, Indian research universities, and international AI organizations.
• The competitive dynamic with Krutrim, particularly as Krutrim navigates its 2025 to 2026 organizational difficulties.
• The international adoption of Sarvam's open-weights releases beyond Indic-language deployment, particularly in Southeast Asian markets where Indian-language adjacency is relevant.

Sources

• Sarvam AI: Sarvam to build India's sovereign large language model. IndiaAI Mission selection announcement.
• Wikipedia: Sarvam AI. Comprehensive company history reference.
• Rest of World: India's frugal AI startups Sarvam and Krutrim build sovereign models. Indian sovereign-AI ecosystem context.
• Founderpin: Sarvam AI (2026): India's $1.5 Billion Sovereign AI Unicorn. 2026 valuation and funding context.
• Inc42: Sarvam AI Cofounder Vivek Raghavan Bats For 'Sovereign AI'. Founder strategic perspective.
• Inc42: Sarvam And The Sovereign AI Dream. Strategic-context analysis.
• YourStory: India taps Sarvam to build sovereign large language model. IndiaAI Mission selection coverage.
• Sarvam AI official site. Company overview and product information.