The most expensive phrase in the self-service industry is “end of life.” We take a look at the brownfield and greenfield options for Edge AI in computing. From Intel, AMD, Qualcomm, Hailo, Google and Coral and yes, Raspberry PI. Amazon is conspicuously absent we think.
For deployers managing 500+ kiosks in the field, the pressure to add modern features—like computer vision for loss prevention or facial authentication for check-in—usually comes with a terrifying price tag: replacing the entire PC.
If your fleet is running on standard Intel Core i5 or i3 processors from three or four years ago, those chips are perfectly capable of running Windows and your transaction app. They just fail at AI. They don’t have the NPU (Neural Processing Unit) required to process video streams in real-time without crashing the CPU.
The solution isn’t a forklift upgrade. It’s probably just a retrofit.
Enter the Hailo-8 AI Module.
What is the Hailo-8?
Think of it as a graphics card, but for AI, shrunk down to the size of a stick of gum.
The Hailo-8 is an M.2 AI Accelerator module. It fits into the same slot on your motherboard that you would typically use for a WiFi card or an NVMe SSD. Once installed, it acts as a dedicated brain for artificial intelligence tasks.
Performance: It delivers up to 26 TOPS (Trillions of Operations Per Second).
Efficiency: It typically consumes less than 2.5W of power, meaning you don’t need to upgrade your kiosk’s power supply or add massive cooling fans.
The Cost: The industrial-grade module typically retails between $170 and $200.
Scenario A (The Rip and Replace): You buy 500 new industrial PCs with integrated NPUs (like the new Intel Core Ultra).
Cost: ~$800 per PC + labor to swap them out.
Total: $400,000+
Scenario B (The Hailo Retrofit): You open the existing box and slot in a Hailo-8 module.
Cost: ~$180 per module.
Total: $90,000
You save over a quarter-million dollars while unlocking the exact same computer vision capabilities found in brand-new hardware.
Installation & Compatibility
Before you order a box of modules, here is the technical checklist for your engineering team:
The Slot: Your legacy PC needs an available M.2 Key M or Key A+E slot. Most industrial box PCs from the last 5 years have at least one expansion slot open.
The OS: Hailo provides robust drivers for Windows and Linux. This is critical for kiosks, which often run on locked-down Windows 10 IoT Enterprise LTSC versions.
The Thermals: While the chip is efficient, AI generates heat. Ensure your kiosk enclosure has basic airflow. If your PC is a sealed fanless brick, you may need a thermal pad to bridge the module to the chassis for heat dissipation.
What Can You Do With It?
Once installed, your “dumb” kiosk suddenly has 26 TOPS of vision power. This enables:
Retail: Real-time object detection to spot non-scanned items at self-checkout.
Access Control: Face-based authentication for employee check-in (replacing ID badges).
Analytics: Anonymous audience measurement (age/gender/dwell time) for digital signage.
The Verdict
If your CPUs are still healthy, don’t retire them. Retrofit them. The Hailo-8 offers the most cost-effective bridge between the hardware you paid for yesterday and the AI features you need today.
A Warning
The Corporate Micro-PC Warning (Dell / HP / Lenovo)
Many operators try to save money by retrofitting standard 1-liter corporate desktops (like the Dell OptiPlex Micro or Lenovo ThinkCentre Tiny) housed inside their kiosks. While these machines do have M.2 slots, integrating a Hailo-8 into them presents two specific engineering challenges:
The M.2 Slot Trade-off: Inside a commercial micro-PC, the high-bandwidth M.2 Key M slot is almost always occupied by the primary NVMe boot drive. You cannot use it without removing the hard drive.
The Workaround: Hailo manufactures an A+E Key version of their module. You can pull out the PC’s Wi-Fi card and slot the AI module there instead. For mission-critical kiosks, losing Wi-Fi shouldn’t be a dealbreaker (they should be hardwired via Ethernet anyway), but it is a required architectural trade-off.
The Thermal Trap: An industrial fanless Box PC acts as one giant, extruded aluminum heat sink designed to dissipate the heat generated by AI components. A corporate micro-PC is a thin sheet-metal box with a single, small CPU fan. While the Hailo-8 is incredibly efficient, running continuous computer vision models inside a sealed, unventilated metal kiosk enclosure using a consumer-grade PC chassis is a recipe for thermal throttling.
(If you are using larger Small Form Factor (SFF) corporate towers, you bypass the M.2 issue entirely by using a standard $15 PCIe-to-M.2 adapter card, allowing you to keep both your boot drive and your Wi-Fi).
(Note: Some of the higher-end HP EliteDesk 800 G4/G5/G6 models do actually squeeze in a second M.2 storage slot, but you can never guarantee that across a mixed legacy fleet. Furthermore, the thermal warning still applies 100%—those tiny HP CPU fans are not designed to exhaust AI heat loads inside a sealed kiosk enclosure).
About the Author: Craig Keefner has over 40 years of experience in self-service technology. This guide is maintained independently by TIG – The Industry Group to provide fact-based hardware analysis.
3.5mm headphone jack module (mandatory in many EU interpretations)
Tactile keypad (PIN entry + navigation)
Speakers + amplified audio
Microphone array (optional but increasingly expected)
Braille-labeled keypad
Raised tactile navigation buttons
Physical “start accessibility mode” button
Dot Inc.refreshable braille (emerging)
Haptic feedback modules
Audio confirmation systems
Reach ranges (typically ~15”–48”)
Knee/toe clearance
Approach space for wheelchair
Ideal
Multimodal (voice + touchless + tactile)
AI-assisted interaction (guidance, translation)
Dynamic UI adaptation per user
Software layer:
TPGiscreen reader / TTS
Audio navigation prompts
High-contrast UI modes
Larger font rendering (software)
Anti-glare / high-brightness displays
Optional: adjustable height or tilt mechanisms
BOCA Systemsprinters
2D barcode / QR scanners
Cameras (AI vision, identity, telehealth)
Receipt printers, speakers
4G/5G modems & routers
IoT gateways
Remote monitoring modules
LG ElectronicswebOS players
BrightSign
Android media boxes
And Then There is Dragonwing
Where Dragonwing wins vs loses
Wins
Power efficiency
Cost (fewer components)
AI-native workloads (vision, voice, multimodal)
Always-connected devices
Loses
Enterprise IT compatibility (Windows ecosystem)
Long lifecycle predictability (Intel still stronger)
Upgradeability (monolithic design risk)
Dragonwing is not competing with Intel CPUs directly — it is competing with the entire Intel + GPU + AI accelerator stack.
dragonwing compare
Pro Tip -- Sometimes it makes more sense to replace. You can make the argument between Intel and AMD but to date AMD still suffers from "zombie syndrome" when it comes to remote monitoring. For a large enterprise that is unacceptable. Stick with company like Giada and see all the options.
Author: Craig Allen Keefner
Please sign in or register for FREE
If you are a registered user on AVIXA Xchange, please sign in
Please sign in or register for FREE
If you are a registered user on AVIXA Xchange, please sign in