Building the AI Infrastructure Investment Thesis
Mapped India's competitive position across the AI value chain and identified two structural white spaces to guide sourcing.
Vasu Guptta
Senior Analyst | Lead, Frontier & Strategic Tech at India Accelerator
From their time as
Senior Analyst | Lead, Frontier & Strategic Tech
India Accelerator • 2024
Overview
Vasu built the AI infrastructure thesis independently as part of India Accelerator's broader shift toward frontier technologies. The brief was open-ended: understand where India could compete in the AI stack and identify investment opportunities with durable moats.
The Story
Vasu built the AI infrastructure thesis independently as part of India Accelerator's broader shift toward frontier technologies. The brief was open-ended: understand where India could compete in the AI stack and identify investment opportunities with durable moats.
Mapping the Value Chain
He started by analyzing the global shift from traditional cloud computing to accelerated computing. He studied demand drivers including foundational model development, GPU shortages, inference workloads, and enterprise AI adoption, then mapped the full value chain from semiconductor design and GPU manufacturing through cloud providers, model optimization platforms, vector databases, orchestration tools, and inference infrastructure.
The mapping exercise surfaced a clear picture of where Indian startups could realistically create differentiated value. While India was unlikely to compete in advanced chip fabrication, it had structural strengths in AI software infrastructure, developer tools, optimization layers, and enterprise AI platforms.
Identifying the White Spaces
Two structural gaps emerged from the research. The first was the absence of sovereign AI models: beyond early-stage efforts, there was no credible sovereign model infrastructure in India. The second was the lack of federated learning infrastructure capable of handling sensitive data in legal, government, defense, and healthcare contexts, where private sector data was locked away from AI model training due to privacy and security constraints.
He validated these gaps through primary research with IIT professors and industry experts, and cross-referenced them against capital flow data and global VC investment patterns.
Shaping the Deal Pipeline
The thesis directly shaped sourcing. He shortlisted companies across sovereign AI infrastructure, decentralized federated learning, and fabless semiconductor design, including accelerator chip builders, inferencing chip companies, and photonics-based chip startups. One federated learning company was generating active healthcare revenue at the time of evaluation. The thesis also informed the firm's decision to continue scouting for GPU-adjacent infrastructure companies.