Specifications
Vendor
SiMa.ai
AI performance
50 TOPS multimodal
Power envelope
Sub-10 W
Precision support
BF16 / INT8 / INT16
Hardware interfaces
MIPI, USB, PCIe Gen5 RC, SPI
Supported workloads
CNN, ViT, LLM and LMM (vision-language models)
Security
Arm TrustZone, on-chip encryption, eKeys
Architecture
Heterogeneous compute, full pipeline on one chip
Modalix SoM
Production SoM, from $449
SoM compatibility
Pin-compatible with NVIDIA Jetson Orin NX and Orin Nano carriers
Modalix HHHL card
PCIe Gen5 x8, passive fanless cooling, 10 Gigabit Ethernet
Modalix chip-down
50 TOPS BGA package for custom PCB integration
Modalix DevKit
$1,499 including SoM, PSU and 500 GB NVMe M.2
Software
Palette SDK with Palette Neat agentic environment
Supply commitment
Industrial grade, extended temperature, 10+ year assured supply
Award
Edge AI + Vision Alliance 2026 Product of the Year - Best Edge AI Board
Overview
The SiMa.ai Modalix MLSoC is a purpose-built machine-learning system-on-chip for physical AI: robotics, drones, automotive, industrial automation and smart vision. Its defining claim is architectural rather than numeric - a single 50 TOPS device at under 10 W runs the entire application pipeline, from sensor capture and multimodal preprocessing through the vision encoder and the language model to application logic and text-to-speech, without glue chips or external accelerators.
That matters for integrators because the usual LMM edge design needs several parts: a vision SoC, an NPU for the encoder, and separate silicon for the transformer stack. Modalix collapses that into one device that speaks MIPI, USB, PCIe Gen5 and SPI natively. The SoM variant is pin-compatible with the NVIDIA Jetson Orin NX and Orin Nano module footprints, so an existing carrier board can be upgraded in place rather than redesigned - a rare property in edge AI where board respins routinely consume a full design cycle.
SiMa.ai ships three hardware forms around the same silicon: the production SoM, a half-height half-length PCIe Gen5 x8 card with passive cooling and 10 Gigabit Ethernet for standard industrial PCs, and a chip-down BGA package for ODMs wanting the tightest SWaP profile. Software is the Palette SDK, whose Palette Neat environment abstracts the heterogeneous compute so models move from assets to working applications in days.
Key Benefits
50 TOPS at sub-10 W removes the need for GPU-class power and thermal design at the edge. Pin-compatible with Jetson Orin NX/Nano carriers means a drop-in upgrade path with no board redesign. Full-pipeline single-chip execution eliminates encoder and transformer glue silicon. 10+ year assured supply and extended-temperature industrial grades suit long-lifecycle deployments.
Applications
Humanoid and collaborative robotics, autonomous drones, automotive perception and in-vehicle infotainment, industrial inspection, multi-channel intelligent video (16 and 80 channel deployments), retail analytics and healthcare imaging at the edge.
Request a Quote — SIMA.AI MODALIX MLSOC — 50 TOPS SUB-10W EDGE PHYSICAL AI PLATFORM
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →