Edge Computing and IoT App Development: A Founder's Guide

Edge computing is not a single technology. It is an architecture pattern where you process data closer to its source instead of sending everything to a centralized cloud. For IoT applications, "the edge" means the device itself, a local gateway, or a nearby micro data center.

Why does this matter? Three reasons: latency, bandwidth, and reliability.

A factory floor sensor generating 1,000 readings per second cannot afford a 200ms round trip to AWS us-east-1 for every decision. An autonomous drone needs to detect obstacles in under 50ms. A retail camera analyzing foot traffic should not consume 10TB of bandwidth per month streaming raw video to the cloud.

Edge computing solves all three problems by processing data locally and only sending summaries, alerts, or aggregated results to the cloud. The global edge computing market exceeded $60B in 2026 and is growing at 37% CAGR. Apple, Google, and Qualcomm are all pushing on-device inference, making edge AI accessible to startups for the first time.

This guide covers everything you need to build edge and IoT applications: connectivity protocols, platforms, edge AI, architecture patterns, and realistic cost estimates.

The protocol you choose for device communication affects battery life, range, throughput, and development complexity. Here are the protocols that matter in 2026.

MQTT (Message Queuing Telemetry Transport)

MQTT is the default protocol for most IoT applications. It is a lightweight publish/subscribe messaging protocol designed for constrained devices and low-bandwidth networks. A typical MQTT message is under 2 bytes of overhead. It runs over TCP/IP, supports QoS levels (0, 1, 2) for delivery guarantees, and works well on intermittent connections. Use MQTT for sensor data collection, device commands, and real-time monitoring. Brokers: Mosquitto (open-source), HiveMQ (managed), AWS IoT Core (fully managed).

CoAP (Constrained Application Protocol)

CoAP is a RESTful protocol designed for extremely constrained devices (8-bit microcontrollers with kilobytes of RAM). It runs over UDP instead of TCP, reducing overhead further. Use CoAP for battery-powered sensors that need to sleep most of the time and wake briefly to send readings. Less common than MQTT in 2026, but still the right choice for ultra-low-power devices.

Bluetooth Low Energy (BLE)

BLE is the standard for short-range device communication (under 100 meters). Use it for wearables, beacons, medical devices, and any IoT device that pairs with a smartphone. BLE 5.3 supports mesh networking, allowing devices to relay messages across a building. Development is straightforward with iOS CoreBluetooth and Android BLE APIs.

LoRaWAN

Long Range Wide Area Network for devices that need kilometer-range connectivity with minimal power consumption. A LoRa sensor can run for 5 to 10 years on a single battery. Use for agriculture sensors, smart city infrastructure, and any deployment where WiFi and cellular are impractical. The trade-off: very low data rates (0.3 to 50 kbps).

Cellular IoT (LTE-M and NB-IoT)

For devices that need wide-area coverage without deploying your own network. LTE-M supports voice and higher throughput. NB-IoT is optimized for low-power, stationary devices. Monthly connectivity costs range from $0.50 to $5 per device depending on data volume. Use for fleet tracking, smart meters, and remote monitoring.

You need a platform to deploy, manage, and update code running on edge devices. The big three cloud providers each offer edge platforms, and several specialized options exist.

AWS IoT Greengrass

Greengrass extends AWS services to edge devices. Deploy Lambda functions, ML models, and Docker containers to devices running Linux. Greengrass Core handles local messaging, data caching, and cloud sync. Pricing: $0.16 per active device per month. Best for teams already on AWS who want tight integration with IoT Core, SageMaker, and S3.

Azure IoT Edge

Microsoft's edge runtime deploys containerized workloads to edge devices. Tight integration with Azure IoT Hub for device management and Azure Machine Learning for model deployment. The Azure Percept hardware program provides reference designs for edge AI cameras and audio devices. Pricing: $0.15 per device per month for IoT Hub Standard tier.

Google Cloud IoT (via Chronicle)

Google sunset its original Cloud IoT Core in 2023, but rebuilt edge capabilities through Chronicle and Coral hardware. The Coral Edge TPU ($25 for a USB accelerator, $150 for a dev board) runs TensorFlow Lite models at 4 TOPS. Best for computer vision at the edge where Google's ML ecosystem is a strength.

balenaCloud

A container-based platform specifically for managing fleets of Linux devices (Raspberry Pi, Jetson, Intel NUC). OTA updates, device monitoring, and fleet management for $1.49 per device per month. Simpler than the big three cloud options if you do not need deep cloud integration. Popular with hardware startups deploying hundreds to thousands of devices.

EdgeX Foundry

Open-source, vendor-neutral edge computing framework from the Linux Foundation. Runs on any hardware. No per-device fees. Requires more engineering effort to deploy and maintain but avoids cloud vendor lock-in entirely. Good for enterprises deploying to controlled environments (factories, hospitals) where self-managed infrastructure is acceptable.

Edge AI is the most exciting development in IoT. Instead of sending raw data to the cloud for inference, you run machine learning models directly on the device. This enables real-time decisions without network latency or bandwidth costs.

Hardware for Edge AI

• NVIDIA Jetson Orin Nano ($199): 40 TOPS of AI performance. Runs PyTorch and TensorFlow natively. The go-to for computer vision, robotics, and industrial inspection.
• Google Coral Edge TPU ($25-$150): 4 TOPS, optimized for TensorFlow Lite models. Best for simple classification and object detection at ultra-low power.
• Qualcomm QCS6490 ($50-$100): Built into smart cameras and gateways. Supports ONNX Runtime and Qualcomm AI Engine. Strong for consumer IoT devices.
• Raspberry Pi 5 with AI HAT ($80 total): Budget option for prototyping. The AI HAT adds a 13 TOPS NPU. Not production-grade but excellent for proof of concept.

Model Optimization for Edge

Cloud models do not run on edge devices without optimization. A full YOLOv8 model is 50MB and needs a powerful GPU. For edge deployment, you need:

• Quantization: Convert 32-bit floating point weights to 8-bit integers. Reduces model size by 4x with minimal accuracy loss (typically under 1%).
• Pruning: Remove neurons that contribute little to predictions. Can reduce model size by 50 to 90%.
• Knowledge distillation: Train a small "student" model to mimic a large "teacher" model. The student runs on edge hardware at a fraction of the cost.
• Framework conversion: Convert PyTorch/TensorFlow models to TensorRT (NVIDIA), TFLite (Google), or ONNX Runtime for hardware-specific acceleration.

For a deeper look at the trade-offs between processing on-device versus in the cloud, read our guide on on-device AI vs cloud AI.

Production IoT systems rarely use a single tier. Most follow an edge-fog-cloud pattern where different processing happens at different levels.

The Three-Tier Model

Edge tier (devices and sensors): Collects raw data. Performs immediate filtering (discard noise, detect anomalies). Runs latency-critical AI inference (collision detection, quality inspection). Stores data locally during connectivity outages.

Fog tier (gateways and local servers): Aggregates data from multiple edge devices. Runs more complex analytics across device clusters. Manages local device orchestration and firmware updates. Caches data for batch upload to the cloud. A fog node might be a Jetson Orin in a factory or a mini server in a retail store.

Cloud tier (AWS, GCP, Azure): Long-term data storage and historical analytics. Model training and retraining on aggregated data. Fleet management and remote monitoring dashboards. Business intelligence and reporting. Integration with enterprise systems (ERP, CRM, supply chain).

Data Flow Design

Design your data flow to minimize what goes to the cloud. A camera generating 30 frames per second produces roughly 1.5GB per hour of raw video. Streaming that to the cloud is expensive and slow. Instead, process video at the edge, extract events (person detected, anomaly found, count updated), and send only the events to the cloud. This reduces data transfer by 99%+ while preserving all the information you need for analytics.

Use MQTT for device-to-fog communication and HTTPS or gRPC for fog-to-cloud. Store time-series data locally in SQLite or InfluxDB on the fog tier, and sync to TimescaleDB or InfluxDB Cloud in the cloud tier on a configurable schedule.

Edge computing and IoT are not abstract concepts. Here are the verticals where we see the most startup activity and clearest ROI.

Smart Manufacturing

Predictive maintenance using vibration sensors and acoustic analysis. Quality inspection with computer vision detecting defects at production line speed. Energy optimization by monitoring and controlling equipment in real time. ROI: manufacturers report 20 to 40% reduction in unplanned downtime and 10 to 25% reduction in energy costs.

Fleet and Logistics

GPS tracking with real-time route optimization. Driver behavior monitoring (harsh braking, speeding) using accelerometers. Cold chain monitoring for temperature-sensitive cargo. Load optimization using weight sensors. Companies building supply chain applications increasingly rely on edge devices for real-time visibility.

Retail Analytics

In-store foot traffic counting and heatmaps using ceiling cameras with edge AI. Shelf inventory monitoring detecting stockouts. Queue length estimation for staffing optimization. Customer demographics analysis (age range, gender) for marketing. All processing happens on-device for privacy compliance.

Agriculture

Soil moisture and nutrient sensors driving precision irrigation. Drone-based crop health monitoring with NDVI analysis. Weather station networks providing hyperlocal forecasts. Pest detection using computer vision on field cameras. LoRaWAN enables sensors across thousands of acres with minimal infrastructure.

Smart Buildings

HVAC optimization using occupancy sensors and weather data. Lighting automation. Air quality monitoring. Predictive maintenance for elevators and mechanical systems. Water leak detection. A typical smart building deployment uses 500 to 5,000 sensors with 5 to 20 edge gateways.

IoT security is not optional, and the attack surface is larger than traditional web applications.

Security Essentials

• Device identity: Every device needs a unique certificate or key for authentication. Use X.509 certificates managed through AWS IoT Core or your own PKI. Never use shared credentials across devices.
• Encrypted communication: TLS 1.3 for all device-to-cloud communication. DTLS for CoAP. Encrypt data at rest on edge devices in case of physical theft.
• OTA updates: Secure firmware update mechanisms with code signing. If you cannot update a device remotely, a single vulnerability compromises your entire fleet permanently.
• Network segmentation: IoT devices should never share a network with general IT infrastructure. VLANs or dedicated IoT networks prevent lateral movement after a breach.
• Minimal attack surface: Disable unused ports, services, and interfaces. Run minimal operating systems (Yocto, Buildroot) rather than full Linux distributions.

Cost Breakdown

• IoT MVP (3 to 5 months, $60K to $120K): 1 to 3 sensor types, edge gateway, cloud dashboard, basic alerting. Supports 100 devices.
• Production platform (6 to 10 months, $150K to $300K): Multiple sensor types, edge AI inference, fleet management, OTA updates, historical analytics, API access.
• Enterprise IoT (10 to 18 months, $300K to $600K+): Multi-site deployment, custom hardware design, advanced edge AI, ERP/MES integration, SOC 2 compliance.

Ongoing Costs

Per-device cloud costs: $0.15 to $0.50/month. Cellular connectivity: $0.50 to $5/month per device. Edge hardware: $25 to $500 per node depending on AI requirements. OTA update infrastructure: $500 to $2,000/month at fleet scale.

Getting Started

Start with a proof of concept on Raspberry Pi or Jetson hardware. Validate that your sensors produce useful data and that edge processing delivers the latency and cost benefits you expect. Then design for production hardware. Jumping straight to custom PCB design before validating the concept wastes $50K to $100K if the approach needs to change.

Ready to explore edge computing for your product? Book a free strategy call to discuss your IoT architecture and get a detailed project estimate.