Local ServicesEurope
← Sticker Wall
AGENT · DETAIL
Silo AI: Nordic local AI deployment firm with €20M annual revenue acquired by AMD
Workflow: Input consists of customer industrial data and business pain points, followed by data integration (IoT, ERP, quality ins
AGENT
Key Fields
FIELD STAMPSIndustryLocal Services
RegionEurope（欧洲（芬兰））
ScaleSME
ChannelOnline
🔧 Workflow
Input consists of customer industrial data and business pain points, followed by data integration (IoT, ERP, quality inspection lines). Next, AMD ROCm is used to fine-tune open-source Poro and Viking models on local GPUs. Finally, predictive maintenance, visual quality inspection, and logistics scheduling are deployed via edge or local agents, outputting production-grade AI workflows with continuous monitoring and iteration. The typical path involves data integration, model training and optimization, edge or local deployment, and finally, monitoring and feedback to form a closed loop.
🛠 Setup Requirements
Requires machine learning engineering and MLOps capabilities, familiarity with local GPU deployment (AMD Instinct and Ryzen AI with ROCm), Docker, and Kubernetes, preferably with domain expertise in a vertical such as maritime, pharmaceuticals, or heavy industry. Transitioning from proof-of-concept to production typically takes 6 to 12 months; individuals can start by running a demo with a single-card GPU and open-source models before deepening focus on a single pain point.
🧰 Toolchain
- 🔧 AMD ROCm (local GPU training and inference runtime)
- 🔧 Open-source multilingual models Poro and Viking (SiloGen platform)
- 🔧 Docker and Kubernetes (edge and private cloud orchestration)
- 🔧 Prometheus and Grafana (production monitoring and iteration)
💰 Revenue
Pre-acquisition revenue in 2023 was approximately €21 million (approx. $22 million), with 2024 revenue at about €20 million, representing a 33.4% year-over-year growth, though still in a loss-making expansion phase. In 2024, it was acquired by AMD for $665 million in cash, valuing the company at 20 to 30 times its revenue. Since being integrated into AMD, revenue has not been disclosed separately; it continues to operate as a subsidiary of the AMD AI Group.
💸 Cost
Costs are primarily driven by GPU compute and laboratory personnel. Local Instinct clusters, data labeling, and industrial site implementation are expensive, with individual project delivery costs often reaching the hundreds of thousands of euros. For individual replications, initial investment can be kept under a few thousand euros by using open-source models and second-hand GPUs.
⏱ Time Investment
Full-time consulting delivery teams invest 40 to 60 hours per week (on-site and remote). Individual replicators can start with a 20-hour-per-week part-time pilot and scale up gradually.
🚀 Getting Started
Select a heavy industry pain point (e.g., equipment downtime prediction or logistics arrival time prediction), build a demonstrable proof-of-concept using open-source models and local GPUs, and then find a Nordic SME manufacturer for a low-cost pilot to gain access to real data and case studies. Once a paid project is successful, template the solution and charge across three tiers—consulting, deployment, and MLOps maintenance—to form a reusable delivery pipeline.
🔑 Keys to Success
- ✅ Position data sovereignty and GDPR compliance as the core selling point for localized deployment to appeal to privacy-sensitive heavy industry clients.
- ✅ Deepen expertise in vertical industries such as maritime, pharmaceuticals, and heavy industry to build trust barriers based on real production data.
- ✅ Bind with the AMD hardware ecosystem to lower compute costs and leverage the AMD developer program for resources and endorsements.
- ✅ Utilize Nordic industrial customer alliance networks like Combient to secure pilot opportunities in bulk, allowing one success story to leverage multiple orders within the same industry.
⚠️ 风险
- ⚠️ Enterprise AI projects have long cycles and slow payment terms; reliance on a single or few large clients can lead to cash flow volatility.
- ⚠️ Localized deployments are highly customized and difficult to scale like pure SaaS, with labor costs rising linearly alongside projects.
- ⚠️ The technology stack is highly dependent on AMD; if the ROCm ecosystem lacks maturity or if Instinct hardware supply fluctuates, delivery schedules and client trust may be impacted.
📌 Real Cases
- 📌 AMD's $665 million cash acquisition of Silo AI is the largest private AI company acquisition in Europe since Google's acquisition of DeepMind in 2014; Silo AI now serves as the core of the AMD AI Group.
- 📌 Collaborated with Elomatic to develop a digital platform for ship maintenance, using IoT and AI for predictive maintenance and asset lifecycle optimization; deployed as a pilot in 2023 to reduce downtime losses in maritime and offshore energy scenarios.
- 📌 Partnered with Körber on edge visual quality control AI for pharmaceutical inspection machines, achieving a production line speed of 600 vials per minute for vaccine ampoule inspection.