The company celebrates 40 years in Taiwan with the launch of optimized graphics and AI solutions for workstations, developers, and enterprise data centers.
Intel has taken advantage of the Computex 2025 framework to announce a series of key innovations in its artificial intelligence and high-performance computing strategy. Among the most notable announcements are the new Intel Arc Pro B60 and B50 GPU series, aimed at professionals and AI developers, and the deployment of the Intel Gaudi 3 accelerators, now available in both PCIe cards and rack-scale configurations. Additionally, the company has announced the public availability of its Intel AI Assistant Builder tool for creating optimized local AI agents for Intel platforms.
New Intel Arc Pro B-Series GPUs: AI and Professional Performance
Based on the Xe2 architecture, the new Intel Arc Pro B60 (24 GB) and B50 (16 GB) are designed for demanding workloads, such as AI model inference and professional tasks in workstations across sectors like architecture, engineering, and construction (AEC). Both models feature Intel XMX cores for AI and advanced ray tracing units.
The cards are compatible with consumer and professional drivers on Windows, and offer containerized Linux support for easy integration in development and inference environments. Among their advantages are:
- Multi-GPU support, scaling up to 192 GB of combined VRAM in configurations with 8 GPUs (Xeon platform “Project Battlematrix”).
- Certifications with ISVs for professional applications.
- Optimization for AI models of up to 150 billion parameters.
- Availability:
- Arc Pro B60: sampling begins in June with partners like ASRock, Gunnir, Maxsun, or Sparkle.
- Arc Pro B50: available starting July 2025 from authorized distributors.
Intel Gaudi 3: AI Inference at Scale with Open Flexibility
The third generation of Intel Gaudi 3 accelerators is now available in two formats:
- Gaudi 3 PCIe cards: designed to integrate into current data center servers, they can run models ranging from Llama 3.1 8B to Llama 4 Maverick, adapting to both startups and large enterprises. Availability: second half of 2025.
- Rack-scale designs: open and modular systems supporting up to 64 Gaudi 3 accelerators per rack, incorporating 8.2 TB of HBM memory and liquid cooling to reduce consumption and total cost of ownership (TCO). These configurations are aimed at real-time inference and deploying foundational models in open data centers compatible with OCP.
With this approach, Intel strengthens its proposal against NVIDIA in the AI field, betting on a more open and flexible ecosystem, ideal for CSPs and customers looking to avoid vendor lock-in.
Intel AI Assistant Builder: Custom, Local, and Cloud-Free Agents
Following its initial presentation at CES 2025, the Intel AI Assistant Builder tool is now available in public beta via GitHub. This lightweight framework allows developers to create custom AI agents that run locally on Intel-powered PCs.
Some of the use cases showcased by Acer and ASUS at Computex include smart assistants embedded in laptops with on-device AI, ensuring:
- Lower latency
- Greater data privacy
- Cost reduction by eliminating the need for cloud services
40 Years of Intel in Taiwan: A Future-Focused Celebration
As part of these announcements, Intel is also celebrating its 40th anniversary of operations in Taiwan, highlighting its historical role in the x86 ecosystem and its new phase focused on AI, open architecture, and industrial collaboration.
“Over the past 40 years, our partnership with the Taiwanese tech ecosystem has driven innovation that has changed the world. We now renew our commitment to create products that excite our customers and lead the new era of computing,” stated Lip-Bu Tan, CEO of Intel.
Intel at Computex 2025
📍 Location: Taipei Nangang Exhibition Center, Hall 1 – Stand J0306
📅 Dates: May 20 to May 23, 2025
🕒 Hours: 9:30 AM to 5:30 PM (Taiwan local time)