Sarvam AI co-founder and CEO Pratyush Kumar said the Bengaluru startup’s total compute spend across three model builds was "definitely lesser than what DeepSeek took by significant margin," with the whole compute being "about 20 million, or a little bit less." Kumar clarified the figure is not the cost of a single training run but the cumulative spend across all exploratory and data creation work, stating, "It’s a single-digit million dollar model." When asked if a startup running on single-digit million dollar funding could build a competitive model, Kumar replied, "Absolutely," adding, "You need to have the talent." The company recently raised $234 million at a $1.5 billion valuation in what is called the largest Series B round in Indian startup history, with funds earmarked for training its next frontier model and buying additional compute. Sarvam released Sarvam 105B, a Mixture-of-Experts model with 105 billion total parameters pre-trained entirely from scratch in India, pushing the country to sixth place globally on the Artificial Analysis Intelligence Index. Much of that training relied on subsidized compute; Sarvam was the first company selected under India’s sovereign AI mission, with the government providing access to 4,096 Nvidia H100 GPUs for six months in exchange for an equity stake.