Sarvam in Talks to Raise $250–300 Million at a Reported $3 Billion Valuation

Sarvam, the Bengaluru company building India's home-grown large language models, is in early talks to raise between $250 million and $300 million at a valuation of about $3 billion, The Economic Times reported on 13 October 2026, citing people aware of the discussions. If the round closes on those terms, it would double the company's valuation in roughly four months and could make Sarvam the most highly valued AI start-up in India.

Sarvam has not publicly confirmed the talks, and the figures could change as negotiations progress. But the report says a lot about how investors now see India's attempt to build its own foundation models.

From $1.5 billion to $3 billion in four months

Sarvam's last raise was its Series B in June 2026. The company announced a first close of $234 million out of a planned $300 million round, at a post-money valuation of $1.5 billion. IT services major HCLTech led that round with $150 million, alongside Bessemer Venture Partners and existing investors Khosla Ventures and Peak XV Partners.

A strategic lead from one of India's largest IT services firms was notable in itself. It signalled that Sarvam's models were being treated as infrastructure that large enterprises could build services on, not just as a research project.

The new talks, coming so soon after that close, suggest demand from investors has risen sharply. Reports describe them as early-stage, and no lead investor has been named.

Who Sarvam is

Sarvam was founded in 2023 by Vivek Raghavan and Pratyush Kumar. Both were closely involved with AI4Bharat, the open-source Indian-language AI initiative, and Raghavan previously worked on India's digital public infrastructure. The company builds Indian-language AI models and applications for government agencies and businesses.

Its profile rose sharply in April 2025, when the government's IndiaAI Mission selected Sarvam as the first company to build India's sovereign large language model ecosystem. Under that programme, Sarvam received access to subsidised GPU compute from the Mission's shared pool.

What it has shipped

The fundraising talks follow a year in which Sarvam moved from announcements to releases:

• Foundation models: At the India AI Impact Summit in New Delhi in February 2026, Sarvam launched models with 30 billion and 105 billion parameters. They were later published openly under a permissive licence and were trained in India on IndiaAI Mission compute.
• Speech and documents: In September 2026 it released Saaras V4, a speech-recognition model, and Vision 2.1, a document-understanding model, both aimed at handling India's 22 scheduled languages.
• Enterprise push: It has been building an enterprise sales team, including the hire of former Google Cloud India head Sashikumar Sreedharan, as it tries to sell what it calls sovereign AI to large Indian companies.

Why investors are paying up

Sovereignty has become a market. Governments, banks and regulated industries increasingly want AI models that run on infrastructure inside the country, can be audited, and handle Indian languages well. A domestic model maker with government backing is positioned to win that business, especially in public services such as citizen helplines, document processing and translation.

Language is a real moat. Most frontier models from US labs still perform unevenly on low-resource Indian languages and on code-mixed speech, where people switch between Hindi and English in a single sentence. Sarvam's focus on speech and Indic text is a practical differentiator for voice-first services used by hundreds of millions of people.

Capital intensity. Training and serving large models requires huge GPU budgets. A $300 million raise would give Sarvam room to train larger models and expand inference capacity without depending entirely on government compute allocations.

The caveats

Sarvam's flagship 105-billion-parameter model is not considered frontier-class when compared with the largest models from OpenAI, Google or Anthropic, a point some Indian analysts have made openly. India's sovereign AI stack also still runs largely on Nvidia hardware, so "sovereign" refers to the models, data and deployment, not the chips. Government officials have said they are targeting indigenous GPUs in three to four years.

There is also the question of revenue. A $3 billion valuation implies investors expect large enterprise and government contracts. Sarvam has not published revenue figures, and how quickly its enterprise push turns into recurring business will decide whether this valuation holds.

Why it matters for India

A well-capitalised domestic foundation-model company changes the structure of India's AI ecosystem. Start-ups and government departments get an Indian supplier for models they can fine-tune and host locally; universities and researchers get open models built for Indian languages; and the country gets a test of whether public compute subsidies under the IndiaAI Mission can produce companies that private investors will back at scale.

If the round closes, it would also be one of the largest private investments in an Indian deep-tech company this year, at a time when, according to Inc42, the country's deep-tech start-ups raised $290 million across all of the July-September quarter.

Sources

• https://analyticsindiamag.com/ai-news/sarvam-ai-in-talks-to-raise-300-mn-at-3-bn-valuation-report
• https://x.com/ETtech/status/2109853031974289501
• https://www.newsbytesapp.com/news/business/sarvam-in-talks-for-250-300-million-to-hit-3b-valuation/tldr
• https://www.cxodigitalpulse.com/sarvam-ai-funding-startup-eyes-300-million-at-3-billion-valuation/
• https://www.sarvam.ai/blogs/indias-sovereign-llm
• https://www.forbesindia.com/article/ai-tracker/five-new-sovereign-ai-models-signal-indias-bold-leap-at-ai-summit/2991615/1
• https://inc42.com/buzz/deeptech-funding-zooms-2-8x-yoy-in-q3-despite-tighter-deal-flow/