ai offers a server-class compute platform that combines a 64-bit RISCV CPU with DDR5 memory, PCIe Gen5 I/O, and programmable XPU acceleration for high‑concurrency agent workloads.
Funding
Funding not disclosed
Founders
Product
Problem
Enterprises deploying large‑scale AI agents face challenges in orchestrating many concurrent sandboxed workloads, providing sufficient high‑speed memory for persistent agent state, and delivering low‑latency model inference without relying on heterogeneous, fragmented hardware stacks.
Solution
zett.ai offers a unified server‑class compute platform that combines a 64‑bit RISC‑V CPU for orchestration and virtualization with programmable XPU accelerators for model inference, vector, and matrix operations. The base system includes up to four DDR5 subsystems (6400 MT/s, up to 2 TB each) and 64 PCIe Gen5 lanes for high‑bandwidth I/O. An optional CXL/LPU card adds external DDR4 memory, M.2 SSD storage, and QSFP‑DD networking, enabling massive agent state, retrieval indexes, and off‑loaded data paths. Built‑in hardware security, RAS, and IOMMU‑backed DMA provide isolated sandboxes and reliable production operation. Together, these components deliver a scalable, secure infrastructure for high‑concurrency agent workloads.
Target Audience
Primary customers are cloud providers, data‑center operators, and enterprise AI teams that need to run large numbers of concurrent, secure AI agents and require high‑performance memory and accelerator integration.
Features
- 64‑bit server‑class RISC‑V CPU with 64 KB L1 I/D caches, 1 MB L2, and 32 MB L3, supporting virtualization, interrupt virtualization, and IOMMU‑backed DMA for isolated sandboxes
- Programmable XPU acceleration (GPU/NPU/RPU class) for LLM inference, embeddings, reranking, multimodal generation, and vector/matrix compute
- Four DDR5 memory subsystems (up to 6400 MT/s, 2 TB each) with ECC and RDIMM/UDIMM/SODIMM options
- Up to 64 PCIe Gen5 lanes (32 GT/s per lane) for high‑bandwidth attachment of XPUs, storage, and networking
- Optional CXL/LPU expansion card providing PCIe Gen5 x16, CXL, DDR4 RDIMM memory, M.2 SSD, and QSFP‑DD networking for large agent state and I/O offload
- Integrated hardware security features: trusted boot, cryptographic engines, RAS error handling, and BMC monitoring for production readiness