Krutrim's road to a competitive Indian LLM runs through Ola's compute spend

Krutrim AI's development trajectory since its unicorn valuation in early 2024 has been shaped by a combination of genuine technical ambition and the financial constraints of building a frontier-model company within the broader Ola group structure. Bhavish Aggarwal's AI venture has been investing in a Bengaluru-housed GPU cluster and has published multiple training runs under the Krutrim brand, positioning itself as the most visible Indian attempt at building a proprietary large-language model from a commercially-oriented private-sector starting point rather than a government-sponsored research institution. The compute infrastructure build has been the most tangible and verifiable dimension of Krutrim's work. The company has announced investments in Nvidia hardware running into thousands of GPUs at its Bengaluru facility, a scale that is meaningful for a startup but still far below the hundreds of thousands of GPUs that frontier labs like OpenAI, Anthropic and Google's DeepMind are running for their latest model generations. This compute gap is the central constraint on Krutrim's ability to train models that compete on general English-language benchmarks with the frontier, though the company has argued - reasonably - that Indic-language capability is where the competitive gap is most worth closing for Indian users. Krutrim's public benchmark results have shown a model that outperforms English-first alternatives on several Indic-language tasks while remaining competitive but not leading on English general-knowledge evaluations. This is exactly the positioning one would expect from a lab that has prioritised Indian-language training data and fine-tuning. The commercial product - a Krutrim assistant available through web and mobile interfaces - has been used primarily by early adopters with a specific interest in Indian-language AI applications rather than by a mass consumer market. The connection to the broader Ola group creates both an advantage and a complication. The advantage is access to Ola's data - ride, food delivery and EV usage data that could theoretically be used to fine-tune models for Indian consumer intent - and the financial backing of a group that has committed publicly to AI as a strategic priority. The complication is that the Ola Electric service and operational difficulties through 2024-2025 have created a perception overhang on Aggarwal-led ventures that makes it harder for Krutrim to attract the best AI talent independently. What to watch: whether Krutrim releases a significantly improved model version that closes the benchmark gap with Sarvam and internationally released Indic-capable alternatives, how the compute investment scales relative to the commercial revenue generated from enterprise API sales, and whether Krutrim pursues independent external capital or remains wholly within the Ola group funding structure.

Original source: Moneycontrol