Article Complete
AIM
Sarvam Takes Aim at Every Layer of the AI Stack
Sarvam Takes Aim at Every Layer of the AI Stack

Sarvam Takes Aim at Every Layer of the AI Stack

Alongside new AI models and products, Sarvam said it has built one of India’s largest Blackwell GPU deployments.

Published Updated

Key takeaways

  • Sarvam AI aims to establish itself as a comprehensive AI provider, expanding beyond large language models.
  • The company is developing a one-trillion-parameter model, expected to be completed in six months.
  • Recent upgrades include enhancements to reasoning, speech recognition, and document understanding models.
  • Sarvam launched a managed model training service for enterprises to customise foundation models easily.
  • New application-layer software includes an AI coding assistant and a platform for workflow automation.
5 min

Following the first close of its $300 million Series B round led by HCLTech, Sarvam AI used its inaugural Epoch conference in Bengaluru to unveil a sweeping expansion of its AI stack, introducing new models, developer tools, enterprise products and infrastructure as it sharpened its positioning as a full-stack AI company rather than just a large language model provider. 

The startup also reaffirmed that its previously announced one-trillion-parameter model remains under development and is expected within the next six months.

The company introduced upgrades to its Sarvam-105B reasoning model, Bulbul v4 for expressive multilingual text-to-speech, improved Saaras speech recognition models, and the latest version of Sarvam Vision for document understanding and OCR. 

It also launched a managed model training service that enables enterprises to fine-tune foundation models on proprietary datasets with minimal code.

Beyond the foundation models, Sarvam also expanded into application-layer software. It introduced an AI coding assistant designed as a lower-cost alternative for software development. The company also launched Indus Work Agents, a platform for enterprise productivity and workflow automation. 

Its Voice Agents platform, which was previously available only to enterprise customers, is now open to all developers. Sarvam further unveiled Kivi, a desktop voice dictation application, along with an AI dubbing system that can translate content across languages while preserving the original speaker’s voice.

Sarvam also unveiled an infrastructure service offering low-cost inference for large open models, allowing enterprises to run them with data residency in India.

To support that vision, Sarvam said it now operates around 2,000 NVIDIA Blackwell GPUs, which it described as one of the country’s largest Blackwell deployments, and plans to scale that to 10,000 GPUs. 

The company said the infrastructure would allow Indian enterprises to deploy large open models while keeping customer data within the country.

The startup also revealed its enterprise momentum. 

It said its voice agent platform has already processed over 325 million minutes of calls and announced pricing aimed at undercutting international AI providers. 

Sarvam said more than 500 enterprises, startups and organisations are using its products today, with most of its revenue currently coming from the private sector.

Alongside commercial products, Sarvam demonstrated AI glasses ‘Kaze’ for visually impaired users, highlighted its work with the Odisha government to digitise land records using vision AI, and outlined plans for startup and university programmes to expand adoption of its platform.

Membership

Stay ahead with AI insights.

Unlimited access to premium articles, exclusive analysis, and the stories shaping India's AI economy.

Subscribe now

AIM Americas

1190 Miraloma Way Suite P, Sunnyvale, CA 94085

AIM India

1st Floor, Sakti Statesman, Bellandur, Bengaluru – 560103

© 2026 AIM. All rights reserved.