eRacks Launches Dual Intel Arc Pro B70 AI Servers with Turnkey Provisioning

eRacks Open Source Systems, a company founded in 1999 and headquartered in Campbell, California, has launched a new line of dual Intel Arc Pro B70 AI servers with turnkey AI provisioning. These servers are designed to provide a robust solution for artificial intelligence applications, with specific benchmarks that highlight their performance capabilities.

According to the company's own testing, a single Intel Arc Pro B70 card can generate 26 tokens per second on Qwen3.6-27B and 54 tokens per second on Qwen3-14B. These figures underscore the server's performance in handling large language models, which are critical for AI applications. The Arc Pro B70 card itself comes equipped with 32GB of VRAM, the maximum capacity for models fitting on a single card. By configuring a two-card server, users can access 64GB of GPU memory, offering a cost-effective solution compared to a single 96GB flagship datacenter card. This setup is positioned as the "value play of 2026," providing a significant advantage in terms of memory-to-dollar ratio.

The company’s AI provisioning and setup service, priced at $1,495 or $2,495 including a private retrieval-augmented generation (RAG) stack, is designed to ensure that customers receive a fully optimized and benchmarked system. This service includes burn-in, GPU bring-up with all necessary fixes, deployment, and benchmarking of the customer's chosen models on their actual hardware, along with full rebuild documentation.

eRacks offers a range of configurations for these AI servers, starting from a single B70 card at $13,895, to the eRacks/AINSLEY configuration with two B70 cards and 64GB of GPU memory, priced at $21,395. The company also provides a full AI server line starting at $7,695 and an 8-GPU eRacks/HIGHLANDER flagship server priced at $154,995. These configurations are available for order, with detailed pricing and configuration options available on the company’s website.

The servers are built with open-source technology, using llama.cpp's official Intel build in rootless Podman containers, exposing the industry-standard OpenAI-compatible API. The models are resident entirely in GPU memory, ensuring efficient and fast processing. The company’s founder and CTO, Joseph Wolff, emphasized the importance of practical benchmarks, stating, "Benchmarks on a spec sheet are marketing. Benchmarks on your machine are engineering." This approach ensures that customers receive a system that has been thoroughly tested and optimized for their specific needs.

In summary, eRacks Open Source Systems has introduced a new line of dual Intel Arc Pro B70 AI servers, offering a cost-effective and high-performance solution for AI applications. With comprehensive benchmarking and setup services, these servers are poised to meet the demands of businesses looking to integrate advanced AI technologies into their operations.