Krutrim Retreats From Frontier Models, Pivoting to AI Cloud

Krutrim, the AI venture of Ola founder Bhavish Aggarwal and India's first generative-AI unicorn at a $1B valuation, has retreated from the frontier. In May 2026 it paused its in-house chip-design and frontier-model work, pivoted to an enterprise AI-cloud business, cut more than 200 roles, and withdrew Kruti, its consumer app. Its last released model, Krutrim-2, was a 12B dense model, modest against the 100B-parameter systems now emerging from India's subsidised-compute labs.

A unicorn that ran out of road on foundation models

The pivot is notable because Krutrim was not under-capitalised. It carried unicorn status and a well-resourced founder, and still concluded that building custom silicon, frontier models and a consumer app simultaneously, largely on its own balance sheet, was unsustainable. The economics of full-stack ambition, enormous compute bills, fast-depreciating capability, no near-term monetisation, caught up with it.

The redirected business has a clearer path. Krutrim reported FY26 revenue of roughly ₹3 billion (about $31.5M), three times the prior year, and its first net profit. Enterprise AI cloud, selling compute and managed services rather than racing to train the largest model, has identifiable customers and unit economics. It is a smaller ambition with a defensible floor.

The cautionary signal

The contrast with the sovereign-model cohort is sharp, and Krutrim's exclusion from it is the point: it is not among the twelve organisations the IndiaAI Mission funded. Those labs train on subsidised state compute at under ₹100 per GPU-hour and concentrate on verticalised or voice-first systems. Krutrim attempted the frontier on commercial economics and stepped back; Sarvam reached unicorn status on the frontier using subsidised compute in the same quarter.

That divergence is the lesson for investors. Even a well-funded unicorn found pure foundation-model economics unworkable without state-subsidised compute or a focused vertical to monetise. Krutrim's retreat does not weaken India's AI thesis; it validates that thesis, cheap public compute plus verticalisation, by showing what befalls those who attempt the alternative. The frontier, on a private balance sheet alone, remains a place from which even unicorns withdraw.