Official Raspberry Pi AI HAT+ 26 TOPS

Specifications:

• NPU Accelerator: Hailo-8 Neural Processing Unit
• AI Performance: 26 TOPS (Tera-Operations Per Second)
• Operating Voltage: 5V DC (supplied via host board)
• Host Interface: PCIe Gen 3.0 x1
• Connector: 16-pin FPC ribbon cable
• Form Factor: Standard Raspberry Pi HAT+
• Dimensions: 65mm × 56.5mm
• Weight: 0.035 kg [Estimated]
• Operating Temperature: 0°C to 50°C
• Compatible Platforms: Raspberry Pi 5

Unlock New Capabilities

• Past Projects If you've already experimented with basic object tracking using the Raspberry Pi Camera Module 3 or run lightweight OpenCV scripts that completely bogged down your CPU, you already understand the limitations of software-only processing. This means you're perfectly ready to hand those heavy workloads over to a dedicated neural processing unit.
• Feature Upgrade This Official Raspberry Pi AI HAT+ unlocks real-time pose estimation, simultaneous multi-stream video analysis, and high-framerate object detection directly on the edge. These are advanced, zero-latency machine learning applications you simply cannot achieve smoothly on the Pi 5's native processor alone.
• Effort Level & Minimum Hardware
• Effort Level: Advanced

Included in the Box:

• Official Raspberry Pi AI HAT+ 26 TOPS

Frequently Asked Questions

• Can I use this AI HAT+ at the same time as an NVMe M.2 HAT on the Pi 5?
  Officially, the Raspberry Pi 5 only has a single PCIe port, so you cannot stack both this HAT and an NVMe HAT directly out of the box without using a third-party PCIe switch board.
• Will this work on a Raspberry Pi 4 using a standard USB adapter?
  No, this specific HAT is engineered exclusively for the Raspberry Pi 5's dedicated 16-pin PCIe interface and cannot be adapted for older Pi models.
• Does the software stack run on standard Ubuntu, or do I need Raspberry Pi OS?
  While advanced users can compile the drivers for Ubuntu, official support and the easiest setup path is strictly on the latest 64-bit Raspberry Pi OS (Bookworm).
• Can I power the setup with my older 15W Raspberry Pi 4 power supply?
  No, a 27W Official Power Supply is mandatory; the Hailo-8 chip draws significant current under heavy 26 TOPS loads, and a 15W supply will cause immediate under-voltage throttling or system crashes.
• Is the 5V power delivered purely through the GPIO header or the PCIe ribbon?
  The HAT pulls its primary power through the GPIO stacking header, which is why seating the 40-pin connector securely is just as critical as the PCIe ribbon.
• Which direction should the 16-pin PCIe ribbon cable face during installation?
  The exposed metal contacts on the ribbon cable must face inward toward the center of the board on both the Pi 5 and the HAT; inserting it backwards will prevent the OS from recognizing the NPU.
• Does mounting this HAT block access to the standard 40-pin GPIO header?
  No, the kit includes a tall GPIO stacking header that passes all 40 pins through, allowing you to connect other sensors or pHATs on top.
• Can I just run my existing TensorFlow Lite (.tflite) files directly on the NPU?
  No, you must first compile your models into Hailo's proprietary Executable Format (HEF) using the Hailo Dataflow Compiler toolchain before they can run on the hardware.