Lightspeed India Partner Reflects on Early Investment in SarvamAI, Company Had Only Two Founders, One Idea and Belief
Lightspeed India partner Hemant Mohapatra stated that when the first funding was injected into SarvamAI in mid-2023, the company had only two founders, one idea, and a belief, with the core argument being "token production."
He pointed out that India is about to become the world's largest smart consumer country, yet produces almost no tokens, with all consumption coming from external pricing and sources.
Mohapatra mentioned that he attended the Sarvam Epoch conference, calling it the most cutting-edge conference in India to date, and expressed excitement about the development speed, product quality, and mission realization, with some demonstrations leaving him shocked.
He previously wrote the "Sarvam Manifesto," and this is a sequel, predicting that within five years, any serious AI workload in India will rely on the Sarvam layer.
SarvamAI was founded by Pratyush Kumar and Vivek Raghavan in 2023, focusing on Indian sovereign AI foundational models and local language optimization, having received seed and Series A funding led by Lightspeed, with a subsequent valuation reaching $1.5 billion.
The company has released 30B and 105B parameter models trained from scratch, supporting multiple Indian languages, and launched new products and infrastructure such as voice, vision, and agents at the Epoch conference.
On the market side, the sovereign AI narrative attracts local and strategic capital influx, with funding flowing to Indian foundational model companies that have local data and computing power advantages, benefiting platforms that can scale services for government and enterprises, while pure application-layer entrepreneurs relying on imported large models face pressure.
Source: Public Information
ABAB AI Insight
When Hemant Mohapatra bet on SarvamAI in its early years, it was at the onset of the global large model craze, with Lightspeed simultaneously investing in Anthropic, xAI, Mistral, etc., yet uniquely positioned "token production sovereignty" as India's differentiated path, ahead of the IndiaAI Mission and the implementation of sovereign AI policies.
In terms of capital pathways, Lightspeed locked in equity through early leading investments, later bringing in strategic partners like HCLTech to complete the $1.5 billion valuation round, with funds used for pre-training large models, voice, and multimodal stacks on Indian computing power from scratch, and translating Aadhaar experience into population-scale infrastructure thinking.
Similar cases can be seen with France's Mistral rising with a European sovereignty narrative, and UAE's G42 supporting models; currently, India is in a transition phase from importing intelligence to local production, with Sarvam entering through a full-stack rather than purely model layer.
This essentially belongs to the reconstruction of the industrial chain: when intelligence becomes new infrastructure, the origin of tokens determines pricing power and data sovereignty, with import dependence being replaced by local production, and regulation and capital simultaneously driving this shift.
ABAB News · Cognitive Laws
Consuming intelligence without producing tokens is akin to importing oil.
Sovereign models are the ticket to a population-scale market.
Early beliefs are more determinative of path dependence than later valuations.