IN Brief:
- MountAIn secured €7.3m in seed investment from three investors, with additional Bpifrance support.
- MountainCore adapts complete computer vision applications for microcontrollers costing less than $10.
- Funding will support commercialisation and expansion in Grenoble, Europe, and Silicon Valley.
MountAIn has secured €7.3m in seed financing to commercialise software that enables computer vision applications to run on inexpensive microcontrollers, extending the range of embedded hardware capable of executing artificial intelligence locally.
The Grenoble company received investment from 360 Capital, Supernova Invest, and Crédit Agricole Alpes Développement, with additional support from Bpifrance. It will use the funding to expand its engineering operations and strengthen commercial activities in Europe and the United States.
MountAIn is developing MountainCore, a software platform designed to adapt complete computer vision applications for deployment on processors costing less than $10. Its approach encompasses neural network models and supporting application software rather than concentrating exclusively on the inference model.
Executing computer vision on a microcontroller presents different engineering constraints from running the same application on a workstation or graphics processor. Embedded devices typically have smaller memory capacities, lower processing throughput, and more restrictive power budgets.
A complete vision application must acquire images, prepare the data, execute inference, and transfer the resulting information to the surrounding equipment. Each stage consumes memory, processing time, and energy, with requirements that vary according to image resolution and the complexity of the selected model.
Neural networks also require storage for intermediate results generated during inference. These temporary datasets can consume a substantial proportion of the available memory, particularly where multiple processing layers must retain information for subsequent operations.
Reducing model parameters alone does not necessarily resolve the wider deployment problem. The application must still manage image buffers, peripheral interfaces, software libraries, and the execution of operations across the available hardware resources.
MountainCore addresses these constraints by adapting the model and its supporting software to the target device. MountAIn states that the platform can reduce model memory requirements by up to a factor of three while improving utilisation of integrated AI accelerators.
The manufacturer also claims that deployment work which previously required months of specialised engineering can be reduced to hours in suitable applications. Those figures are company-reported results, and the actual savings will depend on the processor, model architecture, software dependencies, and performance requirements.
Memory optimisation can involve restructuring the execution of operations, improving buffer allocation, and reducing unnecessary movement of intermediate data. The available options depend on the processor architecture and the extent to which its instructions or accelerator hardware support the selected neural network operations.
Some optimisation techniques alter numerical precision or model behaviour. MountAIn states that its approach preserves model accuracy without retraining, although application developers must still assess whether the resulting system meets their own accuracy, latency, and reliability requirements.
The company supports several families of Arm-based microcontrollers and is working with Alif Semiconductor on embedded computer vision using its processor platforms.
Hardware variation remains a practical challenge because microcontrollers differ in available memory, accelerator support, peripheral interfaces, and power management. A deployment tool must account for these differences if the same application is to operate across multiple device families.
MountAIn has also demonstrated its technology with Teledyne e2v. At VISION Stuttgart in 2026, MountainCore was used in a system combining two-dimensional and three-dimensional imaging from the Perciva 5D camera.
The demonstration establishes an industrial application of the technology beyond software-only development. It does not, however, establish identical performance or integration results across every camera and microcontroller configuration.
The company has developed a separate vision sensor, IBEX, to demonstrate how local AI processing can operate within a restricted energy budget. MountAIn specifies power consumption below 100mW for the device.
IBEX is intended for remote monitoring applications, including installations using solar power and satellite communications. Its embedded processing enables image analysis to take place locally instead of requiring every captured frame to be transmitted to a remote computing system.
Local processing can reduce communication requirements when an application needs to report selected events rather than continuous image streams. The resulting energy savings depend on the sensing duty cycle, communication technology, inference workload, and frequency of reporting.
MountAIn has also established an integration with Roboflow, enabling models developed through that platform to be exported for use with IBEX. This provides a route between model preparation and deployment on the company’s hardware demonstration platform.
Applications such as infrastructure monitoring and wildfire detection involve additional system requirements. Equipment must manage changes in lighting, weather, image quality, and environmental conditions while maintaining suitable detection performance.
False detections and missed events can affect the usefulness of an automated monitoring system, so deployment validation remains necessary even where the software optimisation process preserves the underlying model’s reported accuracy.
MountAIn plans to use its funding to recruit engineers in Grenoble and expand its presence in Silicon Valley. Its commercial programme targets semiconductor suppliers, camera manufacturers, and industrial equipment companies.
Further demonstrations are planned for CES in January 2027. The next phase will test whether MountainCore’s deployment approach can be applied consistently across additional processor families and production equipment.
The company’s software addresses a specific obstacle in embedded AI development: the engineering effort required to fit complete applications within limited computing resources. Its commercial progress will depend on the performance, portability, and reliability demonstrated when those applications are deployed on customer hardware.



