Inside an industrial plant, cameras record every movement, every part passing by. Until recently, that footage remained in a surveillance server or was reviewed only after a machine halt. Berlin-based startup Almetra decided to flip that model: point cameras at production lines and turn each frame into immediately usable data.

With a fresh €16.3 million Series A round, the young company has already proven its value in facilities of giants like Bosch and ABB. The idea is simple yet powerful: existing cameras become smart sensors that analyze production in real time, extracting operational metrics – cycle times, bottlenecks, anomalies – and delivering them for instant decisions.

From video to data: what really changes on the factory floor

The platform stands on computer vision that transforms video streams into structured information. Unlike traditional monitoring systems that require manual setups or dedicated hardware, Almetra leverages cameras already in place and processes data on site. This eliminates cloud latency and keeps footage inside the corporate perimeter – a non-negotiable for manufacturers of critical components or those guarding trade secrets.

Real-time video analysis demands significant compute power, but the startup hasn't disclosed specific hardware details. What's clear is that on-premise deployment is mandatory: only by processing data at the edge or on local servers can sub-second response times and full data sovereignty be guaranteed. The same logic drives companies evaluating whether to bring Large Language Models into their own data centers instead of relying on external APIs.

Why on-premise deployment is a decisive factor here

When assessing AI for manufacturing, two constraints always surface: speed and confidentiality. Assembly line footage contains process details, layouts, and improvements that represent competitive advantage. Sending it to external servers, even over encrypted links, is a risk many plant managers won't accept. Almetra understood this and built an architecture that keeps data within the factory.

For those designing AI infrastructure, this case shows how strong the interest is for everything that can be installed and controlled directly. It's not just about GDPR or compliance; the need to react in real time without depending on an internet connection drives the push toward on-premise solutions. Moreover, when scaling across dozens of lines, data transfer costs to the cloud can soar, making the TCO of local infrastructure more competitive over the long term.

U.S. expansion and the broader picture

After gaining traction in Europe, Almetra now eyes the United States, where manufacturing reshoring and government incentives are fueling heavy investment in automation. The challenge will be adapting the technology to different regulations and heterogeneous factory standards, but the local processing principle remains a cross-border advantage.

In a landscape where generative AI grabs most headlines, companies like Almetra remind us that digital transformation in industry runs on practical computer vision applications, delivering value measured in processed parts and avoided downtime. For any organization evaluating similar technology, the trade-off is stark: the full control offered by on-premise processing more than compensates for upfront hardware and maintenance investments, especially in sectors where a single minute of downtime costs tens of thousands of euros.

It's no coincidence that, as Almetra raises capital to scale, the debate around edge and fog computing architectures grows more intense. Those who follow AI infrastructure evolution know that the game is increasingly played closer to the data source.