SambaNova introduces SN50 chip, secures $350m for expansion

California-based SambaNova has introduced its latest AI chip, the SN50, designed for enterprise-scale agentic AI deployments, with shipments expected to statesman aboriginal this year.

The motorboat comes aft a Series E backing circular successful which the institution raised much than $350m to enactment SN50 manufacturing and unreality expansion.

From a method perspective, the SN50 builds connected SambaNova’s Reconfigurable Data Unit (RDU) architecture. SambaNova claims the spot delivers 5 times the compute show per accelerator and 4 times the web bandwidth compared to the erstwhile generation.

The interconnect architecture supports up to 256 accelerators linked by a multi-terabyte-per-second fabric. According to the company, this facilitates deployment of models exceeding 10 trillion parameters and discourse lengths beyond 10 cardinal tokens.

Enterprises tin tally ample connection models and agentic workflows astatine standard with higher throughput, reduced latency, and controlled infrastructure costs, SambaNova said.

SambaNova claims the SN50 achieves 3 times little full outgo of ownership compared to different solutions, portion supporting thousands of concurrent sessions with accordant show metrics. Agentic caching and a three-tier representation strategy are included to optimise hardware utilisation successful environments moving aggregate models simultaneously.

The caller spot is designed for accumulation environments wherever batch size, latency sensitivity, and assets ratio are captious factors.

The SN50 volition beryllium archetypal deployed astatine SoftBank’s next-generation AI information centres successful Japan.

The archetypal rollout volition absorption connected delivering low-latency inference for sovereign and endeavor clients crossed Asia-Pacific. This includes enactment for some open-source and proprietary frontier models that necessitate strict show standards.

SambaNova has an existing narration with SoftBank, which presently operates SambaCloud clusters serving developers successful the region.

SoftBank vice president and exertion portion information level strategy part caput Hironobu Tamba said: “With SN50, we are gathering an AI inference cloth for Japan that tin service our customers and partners with the speed, resiliency and sovereignty they expect from SoftBank.

“By standardising connected SN50, we summation the quality to present world‑class AI services connected our ain presumption — with the show of the champion GPU clusters, but with acold amended economics and control.”

Additionally, SambaNova has entered into a multi-year collaboration with Intel.

The concern aims to supply high-performance, cost-efficient AI inference infrastructure for enterprises, work providers, and authorities organisations by integrating SambaNova’s systems with Intel’s CPUs, accelerators, and networking technology.