Kandou AI, an artificial intelligence chip company founded by former Goldman Sachs managing director Srujan Linga, has raised $225 million in a strategic funding round led by SoftBank Group. The investment values the company at approximately $1.8 billion, according to Bloomberg.
Kandou AI designs custom AI inference chips optimized for edge deployment — data centers, autonomous vehicles, and industrial automation systems where latency and power efficiency matter more than peak training performance. The company’s architecture focuses on reducing the energy cost per inference operation, targeting the growing market for AI chips that run trained models rather than train new ones.
SoftBank’s investment continues its aggressive positioning in AI infrastructure. The firm has committed over $550 billion to AI-related investments in the US alone, including the Ohio data center project announced in March 2026. Kandou AI fits the portfolio thesis that AI’s economic impact will be driven by inference at the edge rather than centralized training — a bet that the volume of AI chips deployed in production will far exceed those used for model development.
The AI chip market has attracted significant competition. NVIDIA dominates training and high-end inference, while startups like Groq (LPU-based inference), Cerebras (wafer-scale), and now Kandou AI target specific segments where NVIDIA’s general-purpose GPUs are either too expensive or too power-hungry. Kandou AI’s edge inference focus differentiates it from competitors focused on data center deployment.
Linga’s Goldman Sachs background is notable in a field dominated by semiconductor engineers. The company’s pitch to investors centers on the commercial opportunity in inference — which McKinsey estimates will account for over 80 percent of AI compute spending by 2028 — rather than purely technical differentiation. Whether finance-informed chip design produces competitive silicon against engineering-first competitors remains the core question for Kandou AI’s valuation.
