The deployment of edge AI models is often not a one-time delivery. When many customers inquire, the most common question is not about model performance itself, but how subsequent deployment, upgrade, and maintenance are charged. This is a practical question because edge scenarios differ from the cloud: devices are scattered, environments are complex, and service costs are indeed harder to estimate. Our current charging methods can be roughly divided into three types, corresponding to different levels of cooperation depth.
The first is project-based deployment fees. It suits single-point projects with fixed scenarios and clear requirements, such as quality inspection equipment in a factory or foot traffic statistics in a store. We provide a package price based on the number of devices, computing platforms (industrial PCs, Raspberry Pi, edge boxes, etc.), and model complexity. This price covers the entire process from environment adaptation, model compression, to on-site debugging. Customers only need to receive a running system, without worrying about the trivial details in between.
The second is annual subscription, which is also the current norm for most long-term customers. Edge AI has a characteristic: after a model has been running for half a year or a year, its accuracy often quietly declines as new data accumulates. At this point, it is necessary to regularly perform incremental training with new data and push updates to edge devices via OTA. We charge a subscription fee annually, including a fixed number of model iterations, remote inspections, and basic troubleshooting. If usage exceeds the agreed number, additional charges are applied per occurrence. The advantage of this approach is that costs are predictable, without sudden large bills.
The third is on-demand response. Some customers already have their own technical teams and only want expert support when encountering tough problems, such as model performance not meeting standards on specific hardware, or compatibility issues after system upgrades. We charge expert service fees by hour or by case, with response times and resolution standards agreed in advance. This model is flexible and suitable for teams that already have in-house capabilities but need external backup.
It should be noted that regardless of the method, we clearly define service boundaries in the contract, such as what counts as optimization and what counts as new feature development, to avoid disputes later. The value of edge AI lies in long-term stable operation, and the charging model is simply to make this sustainable. Customers choose as needed, we deliver as agreed, and only then can cooperation last long.