Archive · 7 Aug 2026
Sarvam AI's India moment: trillion-parameter model, Nvidia-backed funding
Sarvam AI is a company in Bengaluru, India, that builds artificial intelligence — computer programs that can read, write and understand language.
The company wants to build one of the biggest AI models in the world, with more than a trillion parts.
To help pay for it, Sarvam recently got $75 million from investors, led by a big computer chip company called Nvidia.
Months earlier, Sarvam had already received $234 million, so it now has a lot of money to grow.
The company also built its own special internet service in India where its AI models and other AI models can run.
It opened a new office in San Francisco to hire smart scientists, including a researcher who used to work at xAI.
Sarvam's AI models are especially good at understanding Indian languages, which makes them very useful for people in India.
The company says its models are built from scratch, which is one of the hardest things to do in computing.
Sarvam wants India to make its own AI instead of just using AI made in other countries.
Sarvam AI plans a trillion-plus parameter model and has launched an India-hosted inference platform serving its own models as well as GLM 5.2 and Gemma 4.
The Bengaluru-based startup has reportedly raised another $75 million in a funding round led by Nvidia Corp., months after securing $234 million.
Sarvam's models, Sarvam 105B and Sarvam 30B, are built from scratch and outperform larger models from companies including OpenAI, Anthropic and Google on Indian language benchmarks.
The company opened a San Francisco research office and hired former xAI researcher Devendra Chaplot.
Sarvam, valued at about $1.5 billion, aims to make India a producer of foundational AI technology, using the funds for AI agents, software development and cybersecurity.
- Who
- Sarvam AI, a Bengaluru-based Indian AI startup, and its investor Nvidia Corp.
- What
- Sarvam AI unveiled plans for a trillion-plus parameter model, launched an India-hosted inference platform, opened a San Francisco research office and reportedly raised $75 million.
- When
- Specific dates are not stated; the funding round reportedly came just months after Sarvam secured $234 million.
- Where
- Bengaluru, India (headquarters) and San Francisco, United States (new research office).
- Why
- To fund next-generation AI models for applications such as AI agents, software development and cybersecurity, and to expand the computing capacity needed to deploy them across industries.
Frontier competitor
India's pragmatic AI stack
Business strategy
Frontier competitor
Sarvam should pursue its trillion-plus parameter ambition and build a foundation model from scratch, joining a very small global club and making India a producer of frontier AI technology.
India's pragmatic AI stack
Sarvam should emphasise its India-hosted inference platform, which also serves other models such as GLM 5.2 and Gemma 4 — a potentially smarter business model than trying to outsmart OpenAI's frontier models.
Scale versus relevance
Frontier competitor
With a valuation of about $1.5 billion versus the tens of billions that OpenAI and Anthropic can invest, Sarvam cannot realistically compete with the world's frontier labs.
India's pragmatic AI stack
Measuring Sarvam against frontier labs misses the point; success should be measured by whether Indian banks, hospitals, manufacturers and government departments choose its stack.
National champion versus ecosystem
Frontier competitor
India should back Sarvam as its leading AI lab, comparable to OpenAI in the US or Mistral in France.
India's pragmatic AI stack
India should resist crowning a single national champion and instead nurture a diverse ecosystem of startups, academic initiatives and public infrastructure, as China did with Qwen, GLM, Kimi and DeepSeek.
- Company
- Sarvam AI
- Headquarters
- Bengaluru, India
- Latest funding
- Reportedly $75 million, led by Nvidia Corp.
- Earlier funding
- $234 million, secured months earlier
- Valuation
- About $1.5 billion
- Core models
- Sarvam 105B and Sarvam 30B, built from scratch
- New initiatives
- India-hosted inference platform (serving GLM 5.2 and Gemma 4) and a San Francisco research office
- Google DeepMind reshuffle
- Demis Hassabis named Alphabet chief scientist and Google DeepMind chair; Koray Kavukcuoglu promoted to SVP reporting to Sundar Pichai; Jeff Dean and Sanjay Ghemawat left to launch Discovery Loop