SambaNova Systems, an AI start-up targeting the data center market, continues to march forward with its fourth-generation chip. The new SN40L employs TSMC 5 nm technology and offers 688 FP16 Tflop/s. It is available to customers on the SambaNova-hosted cloud with on-premises system shipments scheduled for 2024. SambaNova’s chip architecture is based on its reconfigurable dataflow unit (RDU) originally developed at Stanford. The company integrates the SN40L and its predecessors in systems, which it sells to enterprises and employs for its cloud service. These systems target both training and inference but are better suited for training because they offer no additional acceleration for INT8 and other inference data types. Moreover, SambaNova’s initial customers use its machines for HPC. The Silicon Valley startup raised $1.1 billion in funding from investors such as BlackRock, Google Ventures, Softbank, and Intel Capital. SambaNova’s latest Series D round valued the company at $5.1 billion, a record for a chip company at the time. The company started shipping its first production systems based on its initial AI chip in 2Q 2020 and has two more chips in production. Three years after shipping its first products, the company has disclosed little about its customer success. It has published no AI benchmarks, and its Tflop/s per watt falls short of Nvidia’s. SambaNova’s customers include Lawrence Livermore Laboratory, Argonne National Laboratory, and OTP Bank, Hungary’s largest commercial bank.