VSORA Jotunn8 Inference Chip

VSORA SAS designs inference-native AI chips and the software to run them, with its controller address in Meudon, France and offices across Europe, the US, the Gulf, India and East Asia including Singapore. It addresses the cost and latency of AI inference for cloud, data center and enterprise operators.

The Jotunn8 chip is listed at 3200 TFLOPS with 288 GB of HBM3e memory and 8 TB/s bandwidth, and offers plug-and-play deployment with developer tools. VSORA announced the tape-out of Jotunn8 on October 22, 2025, and partners named on the site include Scaleway and ZML.

• Jotunn8 chip: 3200 TFLOPS inference accelerator with 288 GB HBM3e.
• High bandwidth: 8 TB/s memory bandwidth for large models.
• Software stack: production-ready software and flexible developer tools.
• Plug-and-play deployment: designed to integrate into existing infrastructure.
• Regional hub: partners Scaleway and ZML in the Ile-de-France compute hub.

A Vietnamese data center or telecom operator can evaluate Jotunn8 for hosting LLM inference services. Research or enterprise AI teams can benchmark it against GPU servers for cost per token.

Large on-package memory of 288 GB HBM3e and 8 TB/s bandwidth per the product listing, aimed at high-throughput inference. Plug-and-play deployment and developer tools reduce integration effort. A Singapore office on the site places support within reach of Southeast Asian buyers.

Demand for inference capacity is growing as enterprises and sovereign cloud operators deploy LLMs. Vietnamese data center and telecom operators building local AI capacity can look at non-GPU inference chips as an alternative supply source.

Controller address in Meudon, France. Raised $46 million (announced April 29, 2025); Jotunn8 tape-out October 2025; Ardian joined shareholders July 2026; Sandra Rivera became Chair in January 2026. Founding year not stated.