The Moving Bottleneck: Networking, Power, Memory, and the Race to Win AI

Identifying and solving critical bottlenecks in emerging technology waves defines the entrepreneurial career of Rajiv Khemani. From early innovations in SSL encryption at Cavium to purpose-built cloud switching at Inovium and current AI infrastructure efforts at Upscale AI, success stems from targeting niche constraints that incumbents overlook. The current AI landscape demands massive capital for leading-edge semiconductor development, shifting the focus toward heterogeneous compute, memory-intensive inference, and standardized scale-up networking. While AI’s insatiable energy appetite creates competition for resources, the industry is moving toward open standards like Ethernet to connect diverse accelerators. As the infrastructure build-out continues, the next generation of value will emerge from solving the interwoven challenges of power delivery, memory bandwidth, and high-speed connectivity, rather than relying on a single dominant hardware provider.