Vsora Jotunn-8 5nm European inference chip (vsora.com)

Vsora announced the Jotunn‑8, a 5nm European-designed inference chip aimed at datacenter and edge AI deployments, positioned to cut cost-per-inference while boosting throughput and energy efficiency. The company pitches Jotunn‑8 as a fully programmable, algorithm‑ and host‑agnostic platform with on‑chip RISC‑V cores to offload or run AI workloads entirely on the device—targeting real‑time use cases like chatbots, search, fraud detection, and emerging agentic/reasoning+generative stacks. Vsora frames the product as a sovereignty‑friendly alternative for Europe’s AI infrastructure with claims of “close to theory” efficiency and sustainability benefits for large‑scale inference. Technically, Jotunn‑8 is offered in multiple performance tiers (Tyr‑4 and Tyr‑2) and modes, with Vsora claiming up to petaflop‑class throughput: fp8 performance ranges from 800 TFLOPS (Tyr‑2) and 1.6 PFLOPS (Tyr‑4) up to a top‑end claim of 3.2 PFLOPS in fp8; fp16 throughput is quoted at 200–800 TFLOPS across configurations. Integer and mixed‑precision rates are also provided (e.g., up to 100 TFLOPS int8/fp8 and 50 TFLOPS fp16/int16 in higher modes). If validated, these specs imply high-density, low‑latency inference capacity with flexible precision support—potentially lowering operational costs for massive, real‑time AI services and easing integration of multi‑modal reasoning and generative models.