Jun 5, 2026 - eRacks AI Servers Now Ship with Intel Arc Pro B70 32GB GPUs - 128GB VRAM Workstations from $12,000, 256GB Flagship Under $40,000

Jun 5, 2026 - eRacks AI Servers Now Ship with Intel Arc Pro B70 32GB GPUs - 128GB VRAM Workstations from $12,000, 256GB Flagship Under $40,000

Fremont, CA - Jun 5, 2026

FOR IMMEDIATE RELEASE

eRacks AI Servers Now Ship with Intel Arc Pro B70 32GB GPUs - 128GB VRAM Workstations from $12,000, 256GB Flagship Under $40,000

Single-card 32GB VRAM at $1,000 cost cuts price-per-GB-VRAM by 40-80% versus NVIDIA pro silicon. 4U flagship scales to 256GB unified VRAM for 70B-class language model inference.

eRacks Systems today added Intel's recently launched Arc Pro B70 32GB workstation GPU to its AI server line, repositioning the entire AI catalog around high-VRAM inference economics. The Arc Pro B70, based on Intel's Battlemage Xe2 architecture, delivers 32 gigabytes of GDDR6 memory and 608 GB/s of memory bandwidth at $949 MSRP - roughly half the price per gigabyte of VRAM versus comparable NVIDIA professional cards.

The eRacks/AIDAN entry AI workstation, starting at $11,082, now ships with a single Arc Pro B70 32GB as the default GPU, providing a full 32 GB of inference VRAM in a 2U rackmount chassis. The eRacks/AINSLEY mid-tier ($19,675) ships with four Arc Pro B70 cards for 128 GB of unified inference VRAM across four PCIe Gen 5 slots. The eRacks/AISHA flagship ($29,995) also ships with four cards by default in a Supermicro 4U barebone chassis that supports expansion to eight cards - 256 GB of total VRAM in a single 4U server.

For language model inference workloads, the constraint is usually memory capacity rather than raw compute. A 70-billion-parameter model in 4-bit quantization fits comfortably in 128 GB of VRAM; the same model in 8-bit quantization fits in 256 GB. Until now, those configurations required NVIDIA RTX 6000 Ada or NVIDIA H100 silicon at four to ten times the per-card cost. The Arc Pro B70 changes that math.

All four eRacks AI products run Linux and open-source AI stacks: PyTorch and TensorFlow with Intel oneAPI backends, llama.cpp with Vulkan or SYCL, vLLM, Ollama, and the Hugging Face Transformers library. There is no proprietary AI software lock-in, and customers own the hardware outright with no per-seat or per-token licensing.

The new defaults are available now in the eRacks configurator at https://eracks.com/products/ai-rackmount-servers/ . eRacks Systems has built custom open-source servers since 1999. Visit https://eracks.com for the full product catalog.

About eRacks Open Source Systems

eRacks Open Source Systems is an open-source server and storage specialist founded in 1999 and headquartered in Fremont, CA. The company builds rackmount servers, NAS, HPC clusters, and AI inference servers configured to customer requirements, running Linux and open-source software. eRacks serves businesses, research institutions, healthcare providers, and government agencies worldwide.

Media Contact

Joseph Wolff eRacks Open Source Systems joe@eracks.com https://eracks.com