furcate

Hardware

From a $1 microcontroller to an industrial edge cluster.

Furcate runs on consumer-grade single-board computers, industrial-grade edge gateways, and microcontrollers in the sub-$5 BoM. The runtime adapts to whatever silicon is available — TensorRT on Jetson, Pixtral on Hailo-10H, LiteRT on Coral, TFLite Micro on ESP32 / Arduino / RP2350 — and the orchestration plane treats them all the same.

Single-board computers

SBCs that run the full Furcate stack.

If it boots Linux and has a few hundred MB of RAM, Furcate runs. The boards in this list are validated targets — we ship reference firmware images, OTA channels, and supported runtimes for each.

Raspberry Pi 5

$80-120

ARM Cortex-A76 · 4-16 GB · USB-C PD · 8 GB best for Furcate full stack

General-purpose edge node, often paired with Hailo-10H AI HAT+ 2

Raspberry Pi 4

$60-90

ARM Cortex-A72 · 2-8 GB · LPDDR4

Existing fleets · LiteRT + Coral USB Accelerator pairings

Pi Compute Module 4 / 5

$60-150

Pi-on-PCB · industrial form-factor · -40 to 85°C variants

OEM-embedded products · ruggedised gateways

NVIDIA Jetson Orin Nano Super

$249

67 TOPS · 6-core Cortex-A78AE · Ampere GPU · 8 GB LPDDR5 · 7-25 W

VLM / LLM inference at the edge · Qwen3-VL · OpenVLA · TensorRT Edge-LLM

NVIDIA Jetson Orin NX / AGX Orin

$400-2,000

100-275 TOPS · industrial-grade · larger model envelopes

Heavier on-device LLM / VLM workloads · multi-camera vision

NVIDIA Jetson Nano Original

$99

0.5 TFLOPS · legacy Maxwell · 4 GB LPDDR4

Legacy fleets · LiteRT only · upgrade path to Orin Nano Super

BeagleBone AI / BeagleY-AI

$70-120

ARM Cortex-A53 · TI C7x DSP · 4 TOPS

Industrial OEMs preferring TI silicon

Radxa ROCK / Banana Pi

$80-150

Rockchip RK3588 · ARM Cortex-A76+A55 · 6 TOPS NPU

Pi-class alternative with onboard NPU

AI accelerators

NPUs and TPUs that ride the SBC.

Pair an accelerator with an SBC for an order-of-magnitude lift in inference throughput at modest power. Furcate detects the accelerator at boot and routes the right model to the right silicon.

Hailo-10H + Pi AI HAT+ 2

$130

40 TOPS INT4 / 20 INT8 · 2.5 W · M.2 · LPDDR4/4X on-module

On-device LLM (10 tok/s for 2B-param) on a Pi 5

Hailo-8 / Hailo-15

$70-180

26 TOPS · M.2 / mini-PCIe / chip-down · automotive-qualified

Industrial cameras · POS systems · automotive

Google Coral M.2 + USB

$25-60

4 TOPS · 2 W · TFLite-quantised models

Classical vision inference at very low power

Intel Movidius Myriad X

$30-80

1 TOPS · USB / M.2 · OpenVINO native

Smart camera / industrial vision

Kinara Ara-2

$150-300

40 TOPS · M.2 · transformer-tuned

Edge-LLM / VLM serving where Hailo doesn't fit

Microcontrollers · TinyML

Where the model fits in tens of kilobytes.

TensorFlow Lite Micro / LiteRT Micro inferences without an OS, on flash budgets measured in kilobytes. Gesture recognition, voice commands, sensor anomaly detection, sub-mW always-on patterns.

ESP32-P4 (Espressif)

$8-25

RISC-V dual-core 400 MHz · MIPI-CSI camera · USB · TFLM

Camera-equipped microcontroller — TinyML person/animal detection at < $20 BoM

ESP32 / ESP32-S3

$3-12

Xtensa dual-core · Wi-Fi + BLE · TFLM

Wireless sensor pods · TinyML gesture / audio / motion

Raspberry Pi RP2350

$0.60-12

ARM Cortex-M33 / Hazard3 RISC-V dual-core · 520 KB SRAM

TinyML for under $1 in volume — the Pico 2 ecosystem

Arduino Portenta H7 / Nicla Vision

$80-130

Cortex-M7 + M4 · onboard camera · TFLM

Industrial-grade Arduino with a real ML toolchain

Arduino Nano 33 BLE Sense Rev2

$45

Cortex-M4F · 9-DOF IMU + microphone + environmental sensors

TinyML reference board — gesture, voice, anomaly

Nordic nRF52 / nRF54

$8-25

Cortex-M4 / M33 · BLE · sub-mW always-on

Battery-powered edge — TinyML voice keyword spotting

STM32H7 / STM32U5

$5-15

Cortex-M7 / M33 · STM32Cube.AI

Industrial microcontroller TinyML — STM ecosystem

Industrial edge gateways

Ruggedised compute for the floor.

Sealed enclosures, DIN-rail mounts, redundant power, fanless thermals, IEC 62443 zones-and-conduits-aware networking. Used in process plants, gigafactories, mines, ports, ISO-rated environments.

NVIDIA IGX Thor (industrial-grade)

Inquire

Real-time physical-AI edge · functional safety · enterprise reliability

Highest-end sovereign edge inference for medical / industrial / autonomous machines

Advantech / ASUS / Kontron / Avalue

$500-3,000

x86 + ARM rugged industrial PCs · DIN-rail · -40 to 70°C

Bring-your-own-OS-image gateway for plant-side fleets

Crusoe Edge Zones / Azure Local hardware

Inquire

Modular AI datacentre · sovereign-zone reference

Customer-owned modular DCs that run inference at the customer's perimeter

HPE Edgeline / Dell NativeEdge / Lenovo ThinkEdge

Inquire

Enterprise server-class edge

Large-footprint industrial sites · regulated sectors · air-gapped operation

Reference designs

Branded firmware, your label.

For OEMs, Furcate ships reference hardware designs and branded firmware images. Take a Pi CM5 or Jetson Orin NX module, drop it on your carrier, ship a product with our runtime + orchestration + your application stack already integrated. Long-term-support contracts available for customers under regulatory or industrial maintenance regimes.