br.ai.ns Institute

Why This Matters for Europe

While Huawei Ascend has been widely deployed across China, there has been a persistent perception that the technology is "China-only" — untested, unsupported, and incompatible with European workflows. br.ai.ns Institute in Novi Sad, Serbia shattered that perception by becoming the first European organization to build a production AI research environment entirely on Ascend hardware.

This is a landmark reference point for any European enterprise evaluating Ascend: if a research institute in the EU neighborhood is running real workloads, publishing open-source tools, and pushing code to GitHub, then the ecosystem is ready.

The Institute

br.ai.ns (Belgrade Research on AI and Neuro Science) is an interdisciplinary AI research institute affiliated with the University of Novi Sad. Their research spans:

• Natural Language Processing — multilingual models for Serbian, Croatian, Bosnian, and other South Slavic languages
• Computer Vision — medical imaging analysis and agricultural remote sensing
• Reinforcement Learning — autonomous systems research
• Open-source AI tooling — utility libraries for Ascend development, published on GitHub

Infrastructure

The institute operates a cluster of Atlas 800 (3010) training servers, each equipped with multiple Ascend 910 NPUs. This gives them:

• Multi-petaflop training capacity — sufficient for pre-training mid-size language models (7B-30B parameters)
• Full MindSpore + PyTorch compatibility — researchers use both frameworks depending on the project
• On-premise data sovereignty — all research data stays within the institute's physical infrastructure, a requirement for EU-funded research projects

"We chose Ascend because it gave us GPU-class compute at a price point that made large-scale AI research viable for a European academic institution. The cost of an equivalent Nvidia cluster would have been prohibitive."

Technology Stack

Open-Source Contributions

What sets br.ai.ns apart from typical enterprise deployments is their commitment to open-source. The institute maintains public GitHub repositories with:

• Ascend utility tools — helper scripts for environment setup, model conversion, and performance profiling
• Benchmark datasets — curated sets for South Slavic NLP research
• Model fine-tuning recipes — documented procedures for adapting LLMs to run on Ascend NPUs

This is invaluable for the European Ascend community: instead of starting from scratch, new adopters can build on proven, publicly auditable foundations.

Relevance for Central & Eastern Europe

Serbia's geographical and regulatory proximity to the EU (candidate country, data protection laws aligned with GDPR) makes this deployment directly relevant for Czech, Slovak, Polish, and Hungarian enterprises considering Ascend:

• Same regulatory environment — data sovereignty requirements are comparable
• Same talent pool — Eastern European AI engineers familiar with PyTorch can transition to CANN
• Proven supply chain — hardware delivery and support demonstrated in the region
• Academic collaboration potential — br.ai.ns is open to cross-border research partnerships

Key Takeaway

br.ai.ns Institute proves that Ascend is not a China-only story. European institutions can deploy, operate, and contribute to the ecosystem — with full data sovereignty, open-source transparency, and at a cost point that democratizes large-scale AI research. For any European CTO reading this: if an academic institute in Serbia can do it, your enterprise certainly can.

📎 Source: br.ai.ns Institute public GitHub repository, University of Novi Sad partnership announcements