Sarvam AI: US$234m first close at US$1.5bn Series B
Sarvam AI's stand at an AI event · Image: TechCrunch
Independent report content
| COMPANY | Sarvam AI (Axonwise Private Limited), Bengaluru · founded August 2023 |
|---|---|
| FOUNDERS | Vivek Raghavan (10+ years at UIDAI/Aadhaar) and Pratyush Kumar · both from AI4Bharat at IIT Madras |
| SERIES B | US$234m first close of a ~US$300m round, 15 Jun 2026 · US$1.5bn valuation · India's newest AI unicorn |
| LEAD | HCLTech: US$150m for ~10.46% (₹1,427 Cr, cash) as lead strategic investor |
| OTHER INVESTORS | Bessemer Venture Partners (new); Khosla Ventures and Peak XV Partners (existing) · earlier: Lightspeed (Series A lead) |
| GOVERNMENT MANDATE | Selected in April 2025 as the first startup under the IndiaAI Mission to build a sovereign foundation model |
| MODELS | Sarvam-1 (2024) · Sarvam-M 24B (2025) · Sarvam-30B and Sarvam-105B, trained from scratch, open-sourced (2026) · Indus app |
| REPORTED NEXT STEP | NVIDIA expected to lead a US$74m Series B extension (Entrackr via DealStreetAsia) |
| THE ACCESS POINT Sarvam is private, and its Series B was taken by a listed strategic (HCLTech) and top-tier VCs. Individuals can reach the theme indirectly through HCLTech's listed shares, since HCLTech owns ~10.46% of Sarvam, or through VC funds that hold it. Peak XV, Khosla, Bessemer and Lightspeed are all on the cap table. |
|---|
Executive decision summary
| THE BOTTOM LINE | Sarvam AI is India's answer to a strategic question: should the country depend on foreign AI models for its languages, public services and businesses? In under three years, two AI4Bharat researchers have done several things. They raised US$41 million from Lightspeed, Peak XV and Khosla. They became the first startup chosen under the IndiaAI Mission to build a sovereign model. They trained 30B and 105B models from scratch and open-sourced them. They launched a consumer app (Indus), and they brought in HCLTech, India's third-largest IT services company, as a 10.46% strategic shareholder at a US$1.5 billion valuation. Commercial scale is only starting to show: the exchange filing by HCLTech put unaudited FY26 revenue at ₹45.1 crore, up from ₹1.5 crore in FY25. Paying customers and margins are not public. The valuation rests on strategic position and technology, and the next 18 months must turn that into a business. |
|---|---|
| WHICH INVESTOR | Sarvam fits you IF you want exposure to Indian sovereign AI. Through listed HCLTech, which owns ~10.46%, or through VC funds that hold it, you get a company with government mandates, open models and a large IT-services distribution partner. Wait IF you need proof of commercial scale. The first disclosed revenue, ₹45.1 crore (unaudited) in FY26, is small against a US$1.5 billion valuation, which implies very fast growth from here. |
| COMFORT FIT | Sarvam = Tata Consultancy Services in 1975: a national technology champion being built for an industry that does not yet fully exist. HCLTech = the dealership network. It brings Sarvam's models to thousands of enterprise clients worldwide. IndiaAI Mission = the government highway. Subsidised compute and a sovereign mandate, shared with other selected builders (Soket, Gnani, Gan.ai, BharatGen). OpenAI, Google and Anthropic = the global brands. Larger and further ahead on frontier models; Sarvam competes on Indian languages, voice, cost and sovereignty. |
| THE RUPEIA NUMBER | 10.46%. HCLTech paid about US$150 million in cash for 10.46% of Sarvam, a rare double-digit strategic stake by one of India's top IT-services companies in an Indian foundation-model builder. It turns Sarvam from a research-led startup into a distribution partner for an enterprise channel spanning global clients. |
1 CLAIM REGISTER: EVIDENCE GRADING
A = government or regulator record. B = company or counterparty filing or release. C = independent press. D = third-party database or blog.
| Claim | Source | Grade | Rupeia assessment |
|---|---|---|---|
| US$234m first close at US$1.5bn (Jun 2026) | TechCrunch; DealStreetAsia; Business Standard | C | Established |
| HCLTech US$150m for ~10.46% | HCLTech disclosure via IIFL, Sahi and others | B | Established (₹1,427 Cr, cash) |
| Selected under IndiaAI Mission (Apr 2025) | MeitY selection; Sarvam announcement; Wikipedia | A | Established |
| Compute bill ₹246.71 Cr for 4,096 H100s; ₹98.68 Cr subsidy | Analytics India Magazine | C | Reported |
| Sarvam-30B and 105B trained from scratch; open-sourced under Apache 2.0 | Company; Hugging Face; IndiaAI | B | Established |
| $41m seed + Series A (Dec 2023) | TechCrunch | C | Established |
| NVIDIA to lead US$74m extension | Entrackr via DealStreetAsia | C | Reported; completion not announced |
| Pixxel orbital data-centre satellite (Q4 2026) | Sarvam partnership page | B | Announced |
| FY26 revenue ₹45.1 Cr (unaudited); FY25 ₹1.5 Cr | HCLTech exchange filing, via Inc42 and Muthoot Securities | B | Disclosed (unaudited) |
2 COMPANY OVERVIEW & FOUNDERS
Sarvam AI (legal entity Axonwise Private Limited) was founded in mid-2023 (Inc42 called it five months old in December 2023) in Bengaluru. Its name means "all" or "wholeness". It builds large language models, speech and translation systems, and AI agents for Indian languages and use cases. Its slogan is "AI for all from India".
| Founder | Background (published) | What it brings |
|---|---|---|
| Vivek Raghavan | More than a decade at UIDAI, the body running Aadhaar; AI4Bharat at IIT Madras | Experience deploying technology at population scale, and government relationships |
| Pratyush Kumar | AI4Bharat at IIT Madras (backed by Nandan Nilekani) | Indian-language AI research and datasets |
At its December 2023 launch Sarvam had about 18 employees. Raghavan described the technical approach: adapting model architectures and training them for Indian languages, so that models use fewer tokens and work more efficiently in those languages than general models. Raghavan also said Sarvam had seen "the enormous value in innovating at foundational layers and deploying at population scale".
| RUPEIA READ The founders' background explains the company's positioning. AI4Bharat built much of the open Indian-language data and model infrastructure. Aadhaar is the clearest example of Indian technology built for a billion users. Sarvam combines both: Indian-language AI built to run inside public services and large enterprises. That is why government (IndiaAI, UIDAI) and a large IT-services firm (HCLTech) are its two most important partners. |
|---|
3 WHY INDIAN LANGUAGES ARE A DIFFERENT AI PROBLEM
India's Constitution recognises 22 scheduled languages, written in many scripts, and most Indians use AI through voice in their own language rather than English text. General-purpose models are trained mainly on English-language internet data. They often need more tokens to process Indian languages, which raises cost and lowers quality.
| Problem | Sarvam's approach (published) |
|---|---|
| Token inefficiency in Indian scripts | Architectures and training adapted for Indian languages, so models use fewer tokens (Raghavan, 2023) |
| Voice-first users | Speech recognition (Saaras), text-to-speech, voice agents; sovereign model designed for voice |
| Scarce training data | IndicVoices dataset with IIT Madras and AI4Bharat; models trained from scratch on Indian-language data |
| Documents in Indian scripts | Sarvam Vision for OCR and document understanding |
| Cost for population-scale services | Mixture-of-experts models that activate only part of the network per token |
| RUPEIA READ This is Sarvam's strongest argument. A model that handles Hindi, Tamil or Bengali voice queries more cheaply and accurately than a global model has a real advantage in Indian public services (Aadhaar, government helplines) and in mass-market enterprise use (banking, telecoms, retail). Global labs can close the gap, but they have less incentive to optimise for one country's languages. |
|---|
4 TIMELINE: 2023–2026
| Date | Milestone |
|---|---|
| Aug 2023 | Sarvam founded by Vivek Raghavan and Pratyush Kumar |
| Dec 2023 | US$41m across seed and Series A: Lightspeed led the A and co-led the seed with Peak XV; Khosla participated |
| Sep 2024 | Named among Indian members of the AI Alliance |
| Oct 2024 | Sarvam-1 released |
| Nov 2024 | IndicVoices dataset launched with IIT Madras and AI4Bharat |
| Mar 2025 | UIDAI collaboration on AI voice and multilingual support for Aadhaar services |
| Apr 2025 | Selected by MeitY as the first startup under the IndiaAI Mission to build a sovereign foundation model |
| May 2025 | Sarvam-M (24B parameters) open-sourced under Apache 2.0 |
| 18 Feb 2026 | Sarvam-30B and Sarvam-105B unveiled at the India AI Impact Summit, Bharat Mandapam |
| 20 Feb 2026 | Indus app (beta, on Sarvam-105B) launched on iOS, Android and web |
| Mar 2026 | 30B and 105B models open-sourced on Hugging Face and AIKosh; Sarvam Startup Program launched |
| May 2026 | Partnership with Pixxel for India's first orbital data-centre satellite (Q4 2026); Sarvam Kaze AI glasses planned |
| 15 Jun 2026 | Series B first close: US$234m at US$1.5bn, led by HCLTech (US$150m) |
| 2026 | NVIDIA reported to lead a US$74m Series B extension |
5 TECHNOLOGY: THE MODEL FAMILY
| Model / product | Released | Size and design | Licence / availability |
|---|---|---|---|
| Sarvam-1 | Oct 2024 | Indian-language model | Sarvam AI research licence |
| Sarvam-M | May 2025 | 24 billion parameters | Apache 2.0 (open) |
| Sarvam-30B | Feb 2026 | Mixture-of-experts; ~1B parameters active per token; 32,000-token context | Apache 2.0; Hugging Face and AIKosh |
| Sarvam-105B | Feb 2026 | Mixture-of-experts; ~9–10B parameters active per token; 128,000-token context | Apache 2.0; Hugging Face and AIKosh |
| Indus | Feb 2026 | Consumer assistant app running on Sarvam-105B | iOS, Android, web |
| Saaras V3 | — | Speech recognition across Indian languages | API |
| Sarvam Vision | — | Vision-language model for documents and OCR in Indian scripts | API |
| Sarvam Kaze | Planned May 2026 | AI glasses supporting 10+ Indian languages | Device |
The 30B and 105B models were trained from scratch on datasets focused on Indian languages, with support for reasoning, multilingual tasks, mathematics and coding. Their mixture-of-experts design activates only a fraction of the parameters per token, which keeps inference cheaper than a dense model of the same size.
| RUPEIA READ Open-sourcing the flagship models under Apache 2.0 is a deliberate strategy. It makes Sarvam the default Indian-language base for developers, government departments and enterprises. It gives up licence revenue on the models themselves, and moves monetisation to APIs, agents, enterprise deployments and services (Document AI, Work Agents, Coding Agents, Model Training). The commercial bet is that owning the ecosystem pays more than charging for the weights. |
|---|
6 PRODUCTS & GO-TO-MARKET
| Channel | Offer (published) | Customer |
|---|---|---|
| APIs | Text-to-speech, speech-to-text, document digitisation, translation, model access | Developers and enterprises |
| Enterprise products | Content Studio, Token Factory, Document AI, Work Agents, Coding Agents, Sarvam Edge, Model Training | Enterprises, government departments |
| Consumer | Indus app; Sarvam Kaze glasses | Indian consumers |
| Government | IndiaAI sovereign model; UIDAI voice and multilingual Aadhaar services | Central government |
| Strategic distribution | HCLTech industry models for its clients | HCLTech's global enterprise base |
| Startups | Sarvam Startup Program: 6–12 months of API credits, engineering support | Early-stage companies |
Sarvam holds ISO and SOC 2 Type I certifications, which enterprise and government buyers usually require.
7 FUNDING HISTORY & INVESTOR FORENSICS
| Round | Date | Amount | Investors | Valuation |
|---|---|---|---|---|
| Seed + Series A | Dec 2023 | US$41m | Lightspeed (Series A lead, seed co-lead), Peak XV (seed co-lead), Khosla Ventures | Not published |
| Series B (first close) | 15 Jun 2026 | US$234m of ~US$300m | HCLTech (US$150m, lead strategic), Bessemer (new), Khosla, Peak XV | US$1.5bn post-money |
| Series B extension (reported) | 2026 | US$74m | NVIDIA (expected lead), Glade Brook, Gaja Capital, Indigo Ventures | — |
Speed of conviction: Peak XV and Lightspeed reportedly took less than a week to commit to the seed round in 2023.
Khosla Ventures: OpenAI's first institutional investor backed Sarvam from the start. Vinod Khosla said India needs companies like Sarvam to build deep AI expertise for and in India.
Lightspeed sat out the Series B after leading the Series A, according to reports. That is a notable absence for an earlier lead.
Bessemer entered as a new financial investor; NVIDIA, the main supplier of AI compute, was reported as leading the extension.
| RUPEIA READ The cap table has moved from VC-led (2023) to strategic-led (2026). HCLTech brings enterprise distribution, and NVIDIA (if the extension closes) brings a compute-supply relationship. Strategic investors are valuable for an AI model company, because distribution and compute are its two hardest inputs. They also often have commercial agreements alongside their equity, so the economics of the HCLTech partnership matter as much as the share price paid. |
|---|
8 THE HCLTECH DEAL: TERMS AND LOGIC
| Term | Published detail |
|---|---|
| Amount | About US$150m (₹1,427 Cr), paid in cash |
| Stake | ~10.46% (rounded to 10.5% in coverage) |
| Role | Lead strategic investor in the Series B |
| Use of funds | Sarvam R&D on agentic AI, coding and cybersecurity models |
| Commercial angle | HCLTech to build industry-specific models and solutions for its clients |
| Analyst view | Morgan Stanley reportedly sees limited direct financial impact on HCLTech given the minority stake; Business Standard noted that gains hinge on execution and scale-up |
| RUPEIA CALCULATION US$150 million for 10.46% implies a post-money value of about US$1.43 billion, consistent with the reported US$1.5 billion once the full round is counted. For a company of HCLTech's size it is a modest outlay. For Sarvam, it is about 64% of the Series B first close, which makes HCLTech the single most important shareholder decision of the round. |
|---|
9 WHAT HCLTECH GETS: ITS AI BUSINESS IN NUMBERS
| HCLTech measure (FY26) | Published figure |
|---|---|
| Revenue growth (constant currency) | +3.9% |
| Annualised Advanced AI revenue (year-end) | US$620m |
| New deal wins (total contract value) | US$9.3bn |
| AI platform | AI Force / AI Force 2.0 (software engineering, IT operations, business processes) |
| Large AI deal | US$1.14bn AI-operating-model deal with a Fortune Global 50 company |
| FY27 guidance | 1–4% revenue growth (constant currency) |
HCLTech said the Sarvam stake will let it build industry- and client-specific language models and AI solutions for global clients. It will also speed up sovereign AI solutions for governments, regulated industries and enterprises that need localised, secure and compliant AI. Separately, HCLTech announced an AI data centre in Bhubaneswar in partnership with Sarvam and the Government of Odisha.
| RUPEIA CALCULATION HCLTech's US$150 million investment equals about 24% of its annualised Advanced AI revenue of US$620 million. With overall growth guided at only 1–4%, HCLTech is betting that AI, and sovereign AI in particular, is where its next growth comes from. That makes Sarvam strategically important to HCLTech, not just a financial stake. It is the best reason to expect HCLTech to push Sarvam's models into client work. |
|---|
10 THE INDIAAI MISSION: SCALE AND STRUCTURE
The Union Cabinet approved the IndiaAI Mission on 7 March 2024 with an outlay of ₹10,371.92 crore over five years. It has pillars for compute capacity, an innovation centre for foundation models, datasets (AIKosh), application development, future skills and startup financing. It is run by the IndiaAI Independent Business Division under Digital India Corporation. Its compute pillar targets 10,000 or more GPUs through public-private partnerships.
| IndiaAI pillar | Relevance to Sarvam |
|---|---|
| Compute capacity (10,000+ GPUs via PPP) | Sarvam's 4,096 H100 allocation came through this pillar |
| Innovation centre / foundation models | Sarvam selected first in April 2025 to build the sovereign model |
| Datasets platform (AIKosh) | Sarvam's 30B and 105B weights are published on AIKosh |
| Startup financing | Government equity in exchange for support was reported in coverage |
| RUPEIA CALCULATION Sarvam's reported compute bill of ₹246.71 crore is about 2.4% of the IndiaAI Mission's total ₹10,371.92 crore outlay. The ₹98.68 crore subsidy initially attached to it is about 1%. For a model-builder, being the first and largest compute beneficiary of a national programme is a strategic asset. For the programme, Sarvam is its flagship. |
|---|
11 THE INDIAAI MISSION & COMPUTE ECONOMICS
In April 2025 MeitY selected Sarvam, from 67 applicants, as the first startup to build India's sovereign foundation model under the IndiaAI Mission. Sarvam described the target model as fluent in Indian languages, designed for voice and capable of reasoning. The mission added Soket AI Labs, Gnani.ai and Gan.ai about a month later. Other selections include BharatGen (an IIT Bombay consortium), Fractal Analytics and Tech Mahindra. Coverage reported that the government takes an equity stake in exchange for its support.
| Compute item | Reported figure | Rupeia calculation |
|---|---|---|
| GPUs allocated to Sarvam | 4,096 NVIDIA H100s for six months | ~17.9 million GPU-hours |
| Compute bill | ₹246.71 Cr | ~₹138 per H100 GPU-hour |
| Subsidy at the time | ₹98.68 Cr | ~40% of the bill |
| Later policy | 100% compute subsidy for foundational-model builders (Inc42) | Up from 40%; 40% continues for other GPU uses |
| Total IndiaAI GPU subsidy disbursed (reported) | ~₹111.85 Cr | Sarvam's ₹98.68 Cr is most of it |
| RUPEIA READ At about ₹138 per H100 GPU-hour (roughly US$1.6), IndiaAI's contracted compute is cheap by global standards. With 40–100% of it subsidised, Sarvam's cost of training a large model is a fraction of what a foreign rival pays. That is a real, government-created cost advantage. It also ties Sarvam to policy. If subsidies change, or GPUs are allocated to more builders, the advantage narrows. Coverage has also reported delays in delivering GPUs to some selected startups. |
|---|
12 MARKET & COMPETITION
| Competitor group | Examples (published) | How Sarvam competes |
|---|---|---|
| Global frontier labs | OpenAI, Google, Anthropic | Indian languages, voice, cost, open weights and data sovereignty |
| IndiaAI-selected builders | Soket AI Labs (120B model in development), Gnani.ai (Vachana voice stack), Gan.ai, BharatGen (Param2) | First selected; largest compute allocation; open 30B/105B models already released |
| Open-weight models used in India | Meta's Llama, Mistral, Qwen, DeepSeek | Models trained from scratch for Indian languages; local support |
| IT-services in-house AI | Large IT firms building their own stacks | HCLTech chose to partner and invest instead |
The Series B was announced amid concern about access to foreign AI models. MediaNama's headline linked the round to restrictions on a foreign model provider. Coverage framed sovereign AI as having gained urgency.
13 THE COMPETITION IN DETAIL: KRUTRIM AND THE INDIAAI COHORT
The closest Indian comparison is Krutrim, founded by Ola's Bhavish Aggarwal. It became India's first AI unicorn in January 2024, raising US$50 million at a US$1 billion valuation in a round led by Matrix Partners India. Its path has since diverged from Sarvam's. In April 2025, Business Standard reported that Krutrim had cut its fundraising target from US$500 million to US$300 million. In late 2025 it paused its chip-design work, and in May 2026 it announced a pivot to AI cloud services. It reported FY26 revenue of about ₹300 crore, roughly three times FY25, and its first annual net profit, with a profit margin above 10%.
| Company | Backing | Latest published position (2026) |
|---|---|---|
| Sarvam AI | Lightspeed, Peak XV, Khosla, NVIDIA, HCLTech; IndiaAI Mission | US$1.5bn valuation; ₹45.1 Cr FY26 revenue (unaudited); open-weight 30B and 105B models |
| Krutrim | Bhavish Aggarwal and the Ola group; Matrix Partners India | ~₹300 Cr FY26 revenue and first net profit; pivoted to AI cloud services |
| BharatGen | IIT Bombay-led consortium; government-funded | Param2 launched publicly |
| Soket AI Labs | IndiaAI Mission selection (May 2025) | Reported to be building a 120B model, limited by GPU access |
| Gnani.ai | IndiaAI Mission selection (May 2025) | Vachana voice stack launched publicly |
| RUPEIA READ The two best-funded Indian AI startups have answered the sovereign-AI question differently. Krutrim moved towards renting computing capacity, a utility business that produces revenue sooner: about ₹300 crore in FY26, six to seven times Sarvam's ₹45.1 crore. Sarvam kept building models and is using HCLTech to turn them into enterprise revenue. That is the riskier path, and it is the one the US$1.5 billion valuation is paying for. |
|---|
14 GLOBAL COMPARABLE: MISTRAL AND ASML, SARVAM AND HCLTECH
The closest international parallel to Sarvam's Series B is Mistral AI, Europe's sovereign-AI champion. In September 2025 Mistral raised €1.7 billion at an €11.7 billion post-money valuation. The Dutch chip-equipment maker ASML led with about €1.3 billion for roughly 11%, becoming Mistral's largest shareholder.
| Measure | Mistral AI (Sep 2025) | Sarvam AI (Jun 2026) |
|---|---|---|
| Strategic lead | ASML (industrial technology champion) | HCLTech (IT-services champion) |
| Strategic stake | ~11% | ~10.46% |
| Round size | €1.7bn | US$234m first close (~US$300m planned) |
| Valuation | €11.7bn post-money (~US$13.7bn) | US$1.5bn |
| Positioning | European sovereign AI, open and commercial models | Indian sovereign AI, open models, Indian languages and voice |
| RUPEIA READ The structure is almost identical: a national industrial champion takes about a tenth of the country's leading model-builder, framed as strategic sovereignty. The valuation gap, Mistral at roughly nine times Sarvam, reflects Mistral's head start, its European enterprise revenue and the size of its round. If Sarvam follows Mistral's path, the next 18 months should bring enterprise revenue through HCLTech, which would justify a step-up. |
|---|
15 PARTNERSHIPS: WHERE SARVAM IS ALREADY DEPLOYED
| Partner | What it is | Date |
|---|---|---|
| UIDAI (Aadhaar) | AI voice interactions and multilingual support for Aadhaar services | Mar 2025 |
| MeitY / IndiaAI Mission | Sovereign foundation model | Apr 2025 |
| IIT Madras and AI4Bharat | IndicVoices dataset and language research | Nov 2024 |
| Pixxel | "Pathfinder" orbital data-centre satellite: Sarvam's models run on data-centre-class GPUs in orbit, analysing hyperspectral imagery in real time | May 2026 (orbit as early as Q4 2026) |
| HCLTech | Strategic investor; industry-specific models for HCLTech clients | Jun 2026 |
| RUPEIA READ The Pixxel partnership (see Report 03) is a signal of how Sarvam positions itself. Its models are meant to run anywhere, including on an Indian-built satellite with no foreign cloud in the loop. That is the purest form of the "sovereign stack" argument. It also links two of the companies in this batch of Rupeia reports. |
|---|
16 HOW SARVAM MAKES MONEY: REVENUE MODEL MAP
| Revenue stream | Product / evidence | Customer | Maturity (Rupeia view) |
|---|---|---|---|
| APIs (usage-based) | Speech-to-text, text-to-speech, translation, document digitisation, model access | Developers, startups, enterprises | Live; scale not public |
| Enterprise agents and tools | Work Agents, Coding Agents, Document AI, Content Studio, Token Factory | Enterprises, BFSI, government departments | Early commercial |
| Custom model training | Model Training service; HCLTech industry models | Large enterprises via HCLTech | Starting (2026) |
| Government programmes | IndiaAI sovereign model; UIDAI voice services | Central government agencies | Live |
| Edge and devices | Sarvam Edge; Sarvam Kaze glasses | Device makers, consumers | Pilot / launch stage |
| Consumer | Indus app | Indian consumers | Beta; monetisation not announced |
| Orbital / sovereign infrastructure | Pixxel Pathfinder satellite | Space and defence users | Demonstration (Q4 2026) |
The breadth is notable for a company less than three years old. The most important line is custom models through HCLTech. It combines Sarvam's models with HCLTech's client access and is the most direct path to large enterprise contracts.
17 MOAT ANALYSIS
| Moat | Evidence | Strength | Half-life |
|---|---|---|---|
| Government mandate and trust | First IndiaAI selection; UIDAI work | HIGH | Long, but shared with other selected builders |
| Indian-language data and research lineage | AI4Bharat founders; IndicVoices | HIGH | Medium: global labs are adding Indian languages |
| Subsidised compute | ₹138/GPU-hour, 40–100% subsidised | MEDIUM | Policy-dependent |
| Open-weight ecosystem | Apache 2.0 30B/105B on Hugging Face and AIKosh | MEDIUM | Grows with developer adoption |
| Enterprise distribution | HCLTech partnership | HIGH | Long, if commercial terms are durable |
| Frontier capability | 105B MoE vs global frontier models | LOW-MED | Short: frontier models move fast |
18 THE FIRST REVENUE DISCLOSURE: WHAT HCLTECH'S FILING SHOWED
When HCLTech announced its investment in June 2026, its stock-exchange filing included Sarvam's turnover: ₹45.1 crore in FY26 (unaudited), up from ₹1.5 crore in FY25. HCLTech paid ₹1,427.3 crore in cash for 41,421 equity shares, a 10.46% stake, and expected to complete the purchase within two weeks. Inc42 has separately estimated Sarvam's annualised recurring revenue at about US$12 million.
| Measure | Figure | Basis |
|---|---|---|
| FY25 revenue | ₹1.5 Cr | HCLTech filing |
| FY26 revenue | ₹45.1 Cr (unaudited) | HCLTech filing |
| Revenue growth, FY25 to FY26 | ~30x | Rupeia calculation |
| Price per share paid by HCLTech | ~₹3.45 lakh | Rupeia calculation: ₹1,427.3 Cr ÷ 41,421 shares |
| Implied post-money value | ~₹13,645 Cr | Rupeia calculation: ₹1,427.3 Cr ÷ 10.46% |
| Valuation ÷ FY26 revenue | ~300x | Rupeia calculation |
| Multiple if revenue reaches ₹300 Cr / ₹500 Cr | ~45x / ~27x | Rupeia calculation at today's valuation |
| RUPEIA READ The filing moves Sarvam from a company with no public revenue to one with a small but fast-growing revenue line. At about 300 times FY26 revenue, the price reflects strategic value, not current earnings. What matters now is the growth rate. If revenue reaches ₹300–500 crore within two to three years through HCLTech and government work, the multiple at today's valuation falls to about 27–45 times. If growth stalls, the valuation has little to fall back on. |
|---|
19 VALUATION ANALYSIS
| Measure | Figure | Rupeia calculation |
|---|---|---|
| Series B valuation | US$1.5bn | ~34 months from founding |
| Capital raised | US$41m + US$234m = US$275m (US$349m with reported extension) | Valuation ≈ 5.5x capital raised (before extension) |
| HCLTech-implied value | US$150m / 10.46% | ~US$1.43bn post-money |
| Revenue multiple | About 300x FY26 revenue of ₹45.1 Cr (unaudited); Rupeia calculation | Valuation priced on mandate, technology and partnerships, not current revenue |
| RUPEIA READ Coverage described Sarvam's US$1.5 billion as the highest reported Series B valuation in Indian startup history. Its justification is strategic rather than financial: a government-backed sovereign model, an open-weight ecosystem and a top-three IT-services partner. A future round at a higher price will need revenue to keep compounding from the ₹45.1 crore disclosed for FY26. The most likely sources are the HCLTech channel and government deployments. |
|---|
20 THE INDIAN FOUNDATION-MODEL RACE: A TIMELINE
| Date | Event |
|---|---|
| Mar 2024 | IndiaAI Mission approved (₹10,371.92 Cr) |
| Apr 2025 | Sarvam selected as the first startup to build India's sovereign foundation model |
| May 2025 | Soket AI Labs, Gnani.ai and Gan.ai added; BharatGen (IIT Bombay consortium) among other selections |
| May 2025 | Sarvam-M (24B) open-sourced |
| Feb 2026 | India AI Impact Summit: Sarvam-30B and Sarvam-105B unveiled |
| 2026 | BharatGen's Param2 and Gnani.ai's Vachana voice stack launched publicly |
| Mar 2026 | Sarvam's models open-sourced on Hugging Face and AIKosh |
| Jun 2026 | Sarvam's US$234m Series B at US$1.5bn; HCLTech takes ~10.46% |
| 2026 | Soket reported developing a 120B model, constrained by GPU access |
Sarvam is ahead of the other IndiaAI-selected builders on three visible measures: the first selection, the largest disclosed compute allocation, and the largest private round. Its open-weight strategy means its models can also become the base that others build on.
21 RISK ANALYSIS: RED FLAG LEDGER & KILL CONDITIONS
| Red flag | Evidence | Severity | Kill condition | What would resolve it |
|---|---|---|---|---|
| Commercial proof | FY26 revenue ₹45.1 Cr (unaudited), up from ₹1.5 Cr in FY25 (HCLTech filing) | HIGH | Revenue fails to compound through FY27 | HCLTech deployments and paying customers disclosed |
| Frontier gap | Global labs advance quickly | MEDIUM | Customers prefer foreign models on quality | Leadership on Indian languages, voice and cost |
| Policy dependence | Subsidised compute and mandate | MEDIUM | Subsidy or mandate shifts to other builders | Commercial revenue independent of subsidy |
| Open-source monetisation | Flagship weights are free | MEDIUM | Others monetise Sarvam's models better | Strong API and agent revenue |
| Strategic dependency | HCLTech ~64% of the first close | LOW-MED | Partnership underdelivers | Multiple enterprise channels |
| Compute supply | Reported GPU delivery delays for some builders | LOW-MED | Training timelines slip | NVIDIA relationship (extension) and own compute |
22 SCENARIO ANALYSIS
| Scenario | What happens | Implication |
|---|---|---|
| Bull | HCLTech deploys Sarvam models across clients; government services adopt at scale; extension closes with NVIDIA; next models close the gap with the frontier on Indian tasks | Valuation re-rates on revenue; IPO candidate later in the decade |
| Base | Steady government and enterprise adoption; revenue builds slowly; valuation held by strategic interest | Flat-to-modest next round |
| Bear | Global models dominate Indian-language tasks; subsidies shift; open models commoditise | Down-round pressure; strategic acquisition becomes the exit |
23 INVESTOR CLASS ANALYSIS
For HNIs
Direct Sarvam shares are not generally available. The listed route is HCLTech, which owns ~10.46%, though Sarvam is a small part of HCLTech's value. Unlisted secondaries in AI companies are thin and priced off strategic rounds.
For family offices
Access is through the VC funds on the cap table (Peak XV, Khosla, Bessemer, Lightspeed) or co-investment if Sarvam raises again. Family offices should weigh the sovereign-AI thesis against the absence of public revenue data.
For institutions
Strategic investors (IT services, telecoms, cloud providers) are the natural next investors, as HCLTech and NVIDIA show. Financial institutions will want to see revenue compound from the ₹45.1 crore disclosed for FY26.
For NRIs
NRIs can hold HCLTech shares through portfolio investment routes. Direct Sarvam exposure follows FEMA rules for unlisted Indian shares.
24 COMFORT FIT: THE DECISION PAGE
| If you want… | Best fit | Why | Caution |
|---|---|---|---|
| Indian sovereign-AI exposure (listed) | HCLTECH (~10.46% owner) | Daily liquidity; strategic partner | Sarvam is a small part of HCLTech |
| Direct Sarvam exposure | VC funds on the cap table | Peak XV, Khosla, Bessemer, Lightspeed | Long hold; fund-level diversification |
| Space plus AI | PIXXEL (Report 03) | Orbital data-centre partner | Different business |
| Proven AI revenue | Not yet Sarvam | Revenue not in public announcements | Wait for disclosures |
25 THE EARLY SIGNAL: EVENTS TO WATCH
| When | Event | Why it matters |
|---|---|---|
| 2026 | Series B completion (~US$300m) and NVIDIA-led extension | Final round size and strategic line-up |
| Q4 2026 | Pathfinder orbital data-centre satellite (with Pixxel) | Proof of the sovereign-stack idea |
| 2026–27 | HCLTech client deployments of Sarvam models | First evidence of enterprise revenue |
| 2027 | Next model release vs global frontier | Technology gap |
| Ongoing | IndiaAI compute and subsidy policy | Cost advantage |
26 RUPEIA FINAL INDEPENDENT VIEW
Sarvam is the clearest bet on Indian sovereign AI. In under three years it has the government's first foundation-model mandate, open 30B and 105B models trained from scratch, a consumer app, a UIDAI deployment, an orbital AI experiment with Pixxel, and a 10.46% strategic shareholder in HCLTech at a US$1.5 billion valuation. Few Indian startups have assembled that much strategic position so fast.
What it has not yet shown publicly is the business: revenue, customers and margins. The valuation is a bet that government adoption and HCLTech's channel turn into revenue before global models close the Indian-language gap.
| THE RUPEIA NUMBER 10.46%: the stake HCLTech bought in Sarvam for about US$150 million in cash, an Indian IT-services leader's largest strategic bet on an Indian foundation-model company. |
|---|
| RUPEIA VERDICT Strategically outstanding and commercially early. For most investors the accessible route is indirect: listed HCLTech or the VC funds that hold Sarvam. Treat Sarvam as a long-dated strategic position, and watch the HCLTech deployments for the first real revenue signal. |
|---|
Frequently asked questions
What is Sarvam AI's valuation?
US$1.5 billion, after a US$234 million first close of its Series B in June 2026.
How much did HCLTech invest in Sarvam AI?
About US$150 million (₹1,427 crore) in cash for roughly 10.46%, as the lead strategic investor in the Series B.
Who founded Sarvam AI?
Vivek Raghavan, who spent more than a decade at UIDAI (Aadhaar), and Pratyush Kumar, both previously with AI4Bharat at IIT Madras. The company was founded in mid-2023 (Inc42 called it five months old in December 2023).
Who are Sarvam AI's investors?
HCLTech, Bessemer Venture Partners, Khosla Ventures, Peak XV Partners and Lightspeed. NVIDIA was reported to be leading a US$74 million extension.
What models has Sarvam released?
Sarvam-1 (2024), Sarvam-M (24B, 2025), and Sarvam-30B and Sarvam-105B (2026). The 30B and 105B models are mixture-of-experts, trained from scratch and open-sourced under Apache 2.0.
What is Sarvam's role in the IndiaAI Mission?
In April 2025 MeitY selected Sarvam as the first startup to build India's sovereign foundation model, with access to subsidised GPU compute.
What is Indus?
Sarvam's consumer AI assistant app, launched in February 2026 on iOS, Android and the web, running on Sarvam-105B.
Can I invest in Sarvam AI?
Sarvam is private. Indirect routes include HCLTech's listed shares (HCLTech owns ~10.46%) and VC funds that hold Sarvam.
What revenue does Sarvam AI make?
HCLTech's June 2026 stock-exchange filing put Sarvam's turnover at ₹45.1 crore in FY26 (unaudited), up from ₹1.5 crore in FY25. Profit figures were not disclosed.
How does Sarvam compare with Krutrim?
Krutrim became India's first AI unicorn in January 2024 at a US$1 billion valuation. In 2026 it pivoted to AI cloud services and reported about ₹300 crore of FY26 revenue and its first net profit. Sarvam is valued higher, at US$1.5 billion, on a smaller revenue base, because it is building sovereign foundation models.
Source log
TechCrunch: Sarvam becomes India's newest AI unicorn with $234M round led by HCLTech
DealStreetAsia: Sarvam raises $234m in first close of Series B at $1.5b valuation
IIFL: HCLTech acquires 10.5% stake in Sarvam AI in $150.7 million investment
Business Standard: HCL Tech's gains from Sarvam AI bet hinge on execution and scale-up
MediaNama: Sarvam raises $234 million, becomes AI unicorn
DealStreetAsia: NVIDIA to lead $74m Series B extension in Sarvam AI (report)
Yahoo Finance (BeInCrypto): Sarvam AI hits record Series B valuation
TechCrunch: Sarvam scores $41M funding (2023)
Wikipedia: Sarvam AI
Sarvam AI x Pixxel: orbital data-centre satellite
Sarvam AI on X: selection under the IndiaAI Mission
Analytics India Magazine: IndiaAI, where are the GPUs?
Inc42: Govt to offer 100% compute subsidy to Sarvam, others for foundational models
IndiaAI: Sarvam AI launches open-source foundational models
PIB: Cabinet approves IndiaAI Mission (7 Mar 2024)
Inc42: Cabinet approves IndiaAI Mission, allocates ₹10,372 Cr
Built In: Mistral raises Series C at $13.7B valuation (Sep 2025)
Reuters via Tiger Brokers: ASML becomes Mistral AI's top shareholder
HCLTech: FY26 revenue up 3.9%, led by Advanced AI demand
Entrepreneur India: HCLTech signs US$1.14bn deal with Fortune Global 50 firm
StreetInsider: HCLTech stock rises as Morgan Stanley notes Sarvam AI stake
Inc42: Sarvam enters unicorn club with US$234m round led by HCLTech (15 Jun 2026)
Muthoot Securities: HCL Technologies company news on the Sarvam investment (16 Jun 2026)
Outlook Business: Krutrim pivots to AI cloud, reports first net profit in FY26
YourStory: Krutrim announces strategic pivot to AI cloud services
Business Standard: Krutrim cuts fundraising target (Apr 2025)
DT Next: Krutrim becomes India's first AI unicorn (Jan 2024)
| DISCLAIMER This document is published strictly for informational, educational and academic record-keeping purposes. The publisher is not a SEBI Registered Investment Adviser (RIA) or Research Analyst (RA). It does not constitute personal investment advice, a financial recommendation, or an offer or solicitation to buy, sell or subscribe to any securities. Investments in the securities market are subject to market risks. Read all related documents carefully before investing. |
|---|
