Berlin-Based NeuralBridge Closes €28M Series A to Scale Multilingual AI Infrastructure

From Research Project to Infrastructure Company

NeuralBridge was founded in 2022 by three former researchers from the German Research Center for Artificial Intelligence (DFKI). What began as an academic project investigating inference optimization for morphologically complex languages — a category that includes German, Finnish, Turkish, and many others — became a commercial product when the team realized that most AI infrastructure providers had built their systems around English-first assumptions.

The core insight: tokenization strategies, caching behavior, and latency optimization all perform measurably worse for languages that don't share English's relatively simple morphological structure. NeuralBridge built its infrastructure stack from scratch with these languages as first-class citizens.

What the Funding Will Be Used For

The €28 million Series A, led by Northwave Ventures with participation from two undisclosed strategic investors, will be deployed across three areas:

• Expanding the company's data center footprint with new points of presence in Warsaw, Vienna, and Istanbul, reducing inference latency for Central and Eastern European deployments.
• Growing the engineering team from its current 34 employees to over 80 by end of year, with a focus on compiler engineering and distributed systems.
• Accelerating enterprise sales in the DACH region and the Nordics, where the company already has a pipeline of inbound interest from public sector and financial services clients.

Why Multilingual AI Infrastructure Is Harder Than It Looks

The technical challenge NeuralBridge is addressing is less visible than the product layer above it, but its effects are significant. Applications that perform well in English benchmarks frequently degrade by 20 to 40 percent on equivalent tasks in other languages — not because of model capability, but because of how the underlying infrastructure handles token throughput and memory allocation.

The company's benchmarks show consistent latency improvements of 30 to 60 percent for non-English inference workloads compared to leading cloud providers, though these figures have not been independently verified.