Sarvam AI reduces Vision API price by 67% after reaching 35 million documents digitized.

Sarvam AI, an Indian sovereign AI startup focused on foundational models for India's linguistic landscape, has cut the price of its Sarvam Vision API by 67%, from 1.5 to 0.5 per page. The reduction follows the platform's processing of more than 35 million pages since its February 2026 launch, driven by enterprise demand in finance, healthcare, education, and government. The company achieved the cost reduction through optimizations to its serving stack, including inference kernel improvements for state-space architecture models, page-level batching, and better hardware utilization on its sovereign cloud infrastructure.

Why it matters: Sarvam AI is executing a textbook capital-compression arc within the document intelligence segment, converting early scale into structural cost advantages that act as a moat. By passing infrastructure efficiency gains directly to customers at a 67% price cut, Sarvam is positioning its Vision API as the default low-cost option for Indian-language OCR at national scale, effectively commoditizing the price floor for multilingual document processing in India. The move mirrors the hyperscaler-distribution pattern where volume drives down unit costs faster than competitors can match, while simultaneously building a defensible dataset moat from 35 million pages of unique Indian language documents. This update reinforces the sovereign AI thesis that domestic infrastructure and locally-optimized models can undercut global API pricing for regionally-specific workloads.

Grounded expert take: Sarvam AI's price cut is not merely a promotional move; it signals that the company has reached an inflection point in serving efficiency that typically separates infrastructure leaders from feature-based competitors. The re-engineering of inference kernels for state-space architectures rather than standard transformers indicates a deliberate architectural moat, one that gets harder to replicate as document volume scales. For enterprises across Indian financial services, healthcare, and government—all cited as target verticals—this price point makes AI-native document digitization an operating expense rather than a capital project, potentially accelerating adoption rates that will strain competitors who lack both the language coverage and the domestic compute footprint.