Indian startup Sarvam AI is building a trillion-parameter foundation model and has opened its first San Francisco office. Following a $1.5 billion valuation, the company is positioning itself to compete globally with low-cost AI models and developer tools.
Sarvam AI has announced a significant expansion in its AI capabilities, targeting the development of a trillion-plus parameter foundation model within the next six months. This move marks a major shift for the Indian sovereign AI firm as it seeks to scale its infrastructure and compete directly with major global artificial intelligence developers. The company also confirmed the opening of its first international office in San Francisco, a strategic move intended to tap into global engineering talent while facilitating a knowledge exchange that supports its Indian operations.
Scaling Models and Infrastructure
Co-founder Pratyush Kumar stated that the company is building these high-capacity models from the ground up to address complex tasks in areas such as coding, cybersecurity, and scientific simulation. The initiative is backed by the company's recent Series B funding success. In June 2026, Sarvam AI raised $234 million as part of a $300 million round, reaching a post-money valuation of $1.5 billion. This capital is being directed toward securing the necessary compute infrastructure and accelerating research into agent-based AI systems that can perform complex workflows autonomously.
Competitive Pricing Strategy
To gain market share among developers and enterprises, Sarvam AI is emphasizing a aggressive pricing model. Its 105-billion-parameter model is currently priced at $0.80 per unit of usage, a structure designed to be substantially more cost-effective than international alternatives such as the GPT-5.4 Mini or Gemini 3.5 Flash. By hosting these models within India, the company aims to offer data security advantages alongside its price-focused strategy, specifically targeting local businesses that require domestic data residency.
New Product Ecosystem
Beyond its core foundation model development, the company has expanded its software toolkit. New releases include Sarvas Code, a coding assistant, alongside updated versions of its multilingual speech and text models, Saaras V4 and Bulbul V4. The company has also increased the accessibility of its Voice Agents platform for third-party developers. This broad range of tools is designed to create a full-stack platform, encouraging wider integration of its AI technology across various Indian industry verticals. The key monitorable for investors and stakeholders over the coming months will be the successful deployment of these models and the ability of the company to maintain its competitive pricing advantage while managing the high compute costs associated with training trillion-parameter-scale systems.