> ## Content Index
> Fetch the complete content index at: https://nexi.fund/llms.txt
> Use this file to discover other available public pages before exploring further.

# MountAIn Raises €7.3M to Run Advanced AI on Sub-$10 Chips
- URL: https://nexi.fund/mountain-frugal-ai-2026/
- Published: 2026-11-20T11:30:00.000Z
- Updated: 2026-11-20T11:30:00.000Z
- Description: MountAIn raised €7.3M to run full computer vision applications on microcontrollers costing under $10. Teledyne, Roboflow and Alif are on board; revenue, pricing and shipping products are not disclosed.
- Author: Nexi.fund Labs
- Tags: AI & Infrastructure, #mode-5, #hook-number, #track-E, #layout-chronicle-flow

€7.3 million. Chips under €9.11\. A camera that runs on less than 100 milliwatts. MountAIn, a Grenoble deeptech firm, closed its seed round on 17 November with those three numbers doing most of the talking.

The money came from 360 Capital and Supernova Invest, with Crédit Agricole Alpes Développement and support from Bpifrance. That is about $8.3 million, announced by press release.

For an AI infrastructure round in late 2026, the size is the story. Compute deals get measured in gigawatts and billions. This one gets measured in microcontrollers.

## The ten-dollar ceiling

Computer vision runs in two places today. Cloud GPUs, cheap per model call and expensive per video stream. Or Linux-class edge hardware, which needs more power and a larger bill of materials than a remote camera can justify.

Microcontrollers sit below both. Under $10\. Milliwatts of draw. No operating system to babysit. They also carry kilobytes of memory, and that is where most vision models stop.

MountainCore attacks the integration work rather than the model. It takes a complete vision application — model, pre-processing, post-processing, data movement — and adapts it to the chosen chip. MountAIn claims up to threefold lower memory use without retraining, plus deployment time cut from months to hours. It supports several Arm-based microcontroller families and is built to use whatever AI accelerator sits on the die.

€7.3 million is small money for an AI round in 2026\. That is the most interesting thing about the deal. If the claim holds, inference economics shift from the data-center budget to the bill of materials, and the buyer changes from a hyperscaler to a camera maker shipping 50,000 units.

## A short timeline with one gap

July 2024: a wildfire-smoke pilot starts in the south of France, using IBEX to spot smoke and warn crews before a fire spreads. Roboflow reports the date.

2025: MountAIn is incorporated in Meylan, in the Grenoble area, by Vincent Huard, Olivier Montfort and David Nahmani. Huard is CEO. FrenchWeb dates the incorporation.

October 2026: Roboflow adds direct model export to IBEX. YOLO11n, YOLOv8n and YOLO26n work today; RF-DETR Nano is listed as coming.

16 November 2026: three agreements. Micelia on embedded models, OREUS on sovereign cloud-to-edge deployment, and Modigence Vision on imaging, signed at CES 2026.

17 November: the €7.3 million seed closes. January 2027: a return to CES in Las Vegas to show MountainCore on industrial hardware. End of 2027: a target of 1,000 IBEX units in the field.

The pilot predates the company by about a year, and the announcement does not explain that. A founder project that became a company is common in deeptech. A partner deployment relabelled as a company milestone is less flattering. Either way, the timeline an investor is buying starts in 2025, not 2024.

## What MountainCore actually compiles

Teledyne e2v demonstrated MountainCore on an edge system combining 2D and 3D vision from its Perciva 5D camera at VISION Stuttgart in 2026\. Alif Semiconductor is working with MountAIn on its Ensemble and Balletto processor families. Roboflow brings the developer funnel, by its own count more than two million developers and 80,000 companies.

IBEX shows the stack under constraint. It runs on a solar panel and a small battery, senses in RGB, near-infrared and long-wave infrared, and sends short alerts over satellite instead of streaming frames. In early tests a YOLO nano model hit up to 22 frames per second on the device. A fire-detection deployment is being prepared in Canada.

> The next great leap in AI won't happen only in data centers. It will happen mostly across the billions of devices operating in the physical world.— François Tison, Partner, 360 Capital

The sensor is a reference design with commercial intent. Cameras, industrial gear and infrastructure monitoring are the market. The compiler is the product that scales across those customers.

## The design win is the product

Three memoranda in forty-eight hours looks like momentum. Memoranda are cheap. Design wins are not. MountAIn has not disclosed revenue, price per deployed unit, or how many products ship with MountainCore inside.

FrenchWeb makes the same point and adds the commercial test: converting demos and evaluations into serial production. None of those figures appear in the funding announcement.

Dealroom places the round in the 95th percentile of all-time seed deals for French AI startups. The same note names the competitive frame. Infineon and Nordic are expanding microcontroller support for AI workloads. MountAIn's edge is that its software fits chips already in market instead of asking customers to adopt a new accelerator.

That makes it a software business with a hardware distribution problem. Every microcontroller family is a port. Every port is engineering. Cutting memory needs by threefold lowers the cost of each port. It does not remove the work.

The unit economics stay unproven until one product ships in volume. Consider the shape of the model rather than a number: a camera maker that moves from a Linux module to a microcontroller saves on silicon, power and enclosure, and pays MountAIn for the software that makes the swap possible. If the fee is per unit, revenue scales with the customer's production line. If it is per project, MountAIn stays a consultancy with a compiler attached. The announcement supports neither reading.

Our read: the durable asset is the library of proven adaptations inside the compiler, not IBEX. Sensors buy field data and credibility. The compiler is what gets paid again on the next product, and the one after that.

## What the next two quarters decide

Watch three numbers. Named products in serial production, not partnerships announced. Independent benchmarks of the memory and latency claims on a matched use case. And any move to publish a per-unit price, which would confirm a royalty model rather than bespoke project work.

Watch the incumbents too. If Infineon or Nordic bundles a comparable toolchain with silicon, a standalone compiler loses pricing power quickly. That risk is larger than the standard "can the startup execute" question, and it is the one the round does not insure against.

⚠️

**The bottleneck is integration, not inference power.**  
MountAIn's €7.3M buys a bet that the next AI market is the installed base of cheap chips already controlling cameras, pumps and pipelines — enormous, unglamorous, and slow to upgrade. 

The company has a working demo, credible industrial partners and a claim that would matter if it survives independent testing. It does not yet have the thing that decides this category: a design win that ships in volume. That is the number to wait for.

[ MountAIn Secures €7.3 Million in Seed Funding to Run Advanced AI on Chips Costing Less Than $10 Primary announcement: round size, investors, MountainCore claims, and the Teledyne e2v, Roboflow and Alif partnerships. GlobeNewswire / MountAIn ](https://www.globenewswire.com/news-release/2026/11/17/3404553/0/en/mountain-secures-7-3-million-in-seed-funding-to-run-advanced-ai-on-chips-costing-less-than-10.html?ref=nexi.fund) 

The only source for the exact round terms and the company's own performance claims.

[ MountAIn Brings Always-On Vision AI to Solar-Powered Cameras with Roboflow Deployment detail: the south-of-France wildfire pilot since July 2024, 22 fps on IBEX, the Canada rollout and the 1,000-unit target. Roboflow ](https://roboflow.com/blog/mountain-brings-always-on-vision-ai-to-solar-powered-cameras-with-roboflow?ref=nexi.fund) 

Where the operational numbers sit — and where the pilot date contradicts the company's founding year.

[ MOUNTAIN lève 7,3 M€ : le test industriel de son IA embarquée Independent French coverage that separates the demo from the factory test and lists the commercial figures the round does not disclose. FrenchWeb ](https://www.frenchweb.fr/mountain-de-la-demonstration-a-l-usine-le-vrai-test-de-l-ia-embarquee/900000053?ref=nexi.fund) 

The sceptical read: partnerships validate the technology, serial production validates the business.