Intel expands Arc Pro lineup to take on NVIDIA and AMD in AI workstations

Note: This article was first published on 19 May 2025.

Intel has unveiled a significant expansion of its Arc Pro lineup at Computex 2025, in response to market shifts towards AI and professional visualisation into enterprise-grade inference workstations and a planned push towards edge computing. During a press briefing that HWZ attended, Intel detailed a strategy to establish a stronger presence in the GPU market for professional workloads beginning with the Arc Pro B60 – a multi-GPU solution optimised for high-performance AI tasks and enterprise deployments. With this move, Intel is aiming to position itself as a competitive alternative to NVIDIA and AMD.

Under the Arc Pro B60’s hood is Intel's new Battle Range architecture, which was engineered for both cost efficiency and high performance – thanks too to is 24GB of VRAM. Multi-GPU configurations with the B60 allow for memory pooling across GPUs, achieving 192GB of VRAM – enough to handle AI models with up to 70 billion parameters. This capability is made possible by Intel's xMEX engine, which optimises memory management to prevent bottlenecks during inference tasks. Leveraging PCIe 5.0 and Intel’s latest Xeon platforms, Intel says the B60 outperforms NVIDIA’s RTX A1000 by up to 50% in AI inference workloads and will offer a compelling value proposition for enterprise users.

To support diverse enterprise needs, Intel announced that the B60 would be available in multiple form factors through its board partners. Unlike NVIDIA’s rigid design approach, Intel’s strategy could better enable partners to create purpose-built designs optimised for different deployment scenarios.

AI-Driven debugging and Inference workstations

One of the standout announcements was Intel's AI-driven debugging process. Traditionally, debugging inference workloads has been a time-consuming task, often requiring senior engineers to spend days resolving errors. Intel's new agentic AI structure, equipped with triage, debugging, design, and auto-testing agents, cuts that resolution time down to hours. This leap in efficiency is central to Intel's concept of Inference Workstations, allowing localised, high-performance AI processing without the need for third-party cloud dependencies. This reduces latency, improves data security, and trims operational costs – key considerations for enterprise AI solutions.

Paired with containerised software solutions like IPEX and OneAPI, Intel’s Inference Workstations are set to offer a complete, end-to-end stack optimised for large language models and scalable AI applications.

Roadmap and partnerships

Intel's software roadmap for Arc Pro is set to roll out in phases throughout 2025:

• Q2 2025: Baseline Windows and Linux drivers to support ecosystem testing and early deployments.
• Q3 2025: Workstation certifications and initial containerised VL Stage deployments.
• Q4 2025: Enhanced virtualisation with SRV, coupled with expanded telemetry and manageability features.

The company also announced collaborations with seven major system integrators, all of whom are sampling Arc Pro B50 and B60 GPUs. Full system rollouts are projected for Q3 2025, with pricing estimates ranging from $5,000 to $10,000 depending on configuration.

Going after NVIDIA and AMD in the Enterprise AI space

Intel's strategic expansion amidst growing demand for AI-driven workloads into AI-driven workstations signals a clear attempt not just to compete with NVIDIA and AMD, but to challenge the performance, scalability, and cost efficiency of mainstream inference workstations. By offering scalable multi-GPU solutions that are optimised for localised AI processing, Intel is betting on enterprise users' growing need for on-premises AI capabilities that bypass cloud latency and data privacy concerns.

With an edge-optimised variant slated for later this year, Intel’s ambitions stretch far beyond traditional GPU deployments. If it can deliver on its promises of efficiency, scalability, and cost-effectiveness, Intel’s Arc Pro series may very well compete more directly with NVIDIA's dominance in AI-driven workloads and inference markets and carve out a substantial share in the enterprise AI market.

Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.