Welcome to The Stack’s One To Watch. Every week we pick out a notable startup and go deep on what they’re building, why they’re building it, and the trials behind it all.
There's no doubt NVIDIA has dominated the AI chips conversation and market this decade, with its GPUs powering most of the AI workloads active right now. But in the startup world, SambaNova has attracted plenty of attention with its Reconfigurable Dataflow Units (RDU) architecture.
Set up by Kunle Olukotun, Chris Ré, and Rodrigo Liang in 2017, the chip maker has raised billions in VC funding since then and was most recently valued at $11 billion after closing a $1 billion Series F in early July, months after a $350 million Series E.
Its newest rack hardware, the SambaRack SN50, is built with 16 of its SN50 RDU chips and claims to deliver 5x more compute per accelerator than its predecessor using an average of 10kW, positioning it as an energy-efficient alternative to GPUs.
Chief Product and Strategy Officer Abhi Ingle spoke to The Stack about why the company is focused on energy efficiency, servicing premium inference workloads, and deploying its chips both alone and with existing infrastructure.
