Weekly Intelligence Brief on Silicon - Week of September 15, 2025
Key Takeaways
- Fujitsu develops Artificial Intelligence (AI) reconstruction technology for efficiency.
- Wowza demonstrates real-time video analysis on NVIDIA hardware.
- AI infrastructure demand drives 44% growth in the server market.
- ARM CPUs gain market share, projecting further growth.
- Future plans include trials and accessible AI technologies.
Fujitsu announced the development of new reconstruction technology aimed at enhancing its Takane Large Language Model (LLM). This innovation allows for the creation of lightweight, energy-efficient AI models to improve operational effectiveness. The technology employs quantization techniques alongside specialized AI distillation, which aims to reduce memory consumption significantly while maintaining accuracy.
The proprietary 1-bit quantization method can reduce memory consumption by 94% and retains an 89% accuracy rate compared to unquantized models. This advancement triples inference speed and enables large Generative AI (GenAI) models to operate on a single low-end Graphics Processing Unit (GPU). This efficiency is particularly relevant for deploying AI applications across diverse environments.
Fujitsu plans to initiate trial environments for the Takane model with the new technology in the latter half of fiscal year 2025. As part of its strategy, it will also release models from Cohere's quantized research via Hugging Face, aimed at addressing various societal challenges through advanced AI capabilities.
In a separate development, Wowza Streaming Platform showcased its AI-powered object detection technology running on NVIDIA's Jetson Orin Nano at IBC 2025. The demonstrations revealed capabilities for analyzing multiple HD video feeds in real-time, illustrating the potential for scalable data center workloads. This technology allows for automated alerts and visual insights while supporting deployment in private networks.
Additionally, the latest version of Wowza’s Streaming Engine is optimized for ARM architecture, reportedly enhancing performance and reducing infrastructure costs by up to 30 percent. The flexibility offered by this technology enables organizations to effectively deploy streaming solutions suited to various infrastructures, underlining its potential for diverse applications.
Lastly, a report by Dell’Oro Group indicated a surge in demand for AI infrastructure components, resulting in a 44% increase in the server market during Q2 2025. The report highlights how the growing need for AI workloads contributes to notable market trends, including the increasing market share of ARM CPUs, projected to reach 46% growth in server-related markets by year-end 2025.
- Fujitsu develops GenAI reconstruction technology for optimized and energy-efficient AI models based on Takane LLM
Fujitsu developed a reconstruction technology for GenAI to enhance its Takane LLM with power-efficient AI models. - Fujitsu develops reconstruction technology for GenAI aiming at power-efficient AI models based on Takane LLM
Fujitsu developed a reconstruction technology for GenAI to enhance its Takane LLM with power-efficient AI models. - Wowza demonstrates edge-powered streaming on NVIDIA Jetson Orin Nano at IBC 2025
Real-time video analysis on the NVIDIA Jetson Orin Nano displayed scalability for complex data center workloads. - AI Infrastructure Demand Boosts Server Market by 44% in Q2 2025, Dell’Oro Group Reports
ARM CPUs gain market share in servers, contributing to a projected 46% market growth in 2025 as per Dell’Oro Group. - Wowza showcases AI-powered streaming on NVIDIA Jetson Orin Nano at IBC 2025
Demonstrations at IBC 2025 revealed Wowza's technology for instant captioning and translation without cloud dependency.