Intel's data center division is gearing up to ship a new AI accelerator by year‑end, aiming at inference workloads rather than model training. The product, dubbed Crescent Island, follows an 18‑month development cycle and will initially ship in limited quantities. By focusing on the request‑response phase, Intel hopes to carve a niche where Nvidia currently dominates.

The chip distinguishes itself with an air‑cooling design and LPDDR5 memory, sidestepping the costly high‑bandwidth memory and liquid‑cooling systems used in Nvidia's Blackwell and AMD's offerings. This approach reduces both bill‑of‑materials and power draw, potentially lowering total cost of ownership for data centers seeking scalable AI inference solutions.

Intel's push marks the first major AI hardware effort under CEO Lip‑Bu Tan after a turbulent leadership change. The company hopes the streamlined design will revive its AI ambitions after the poorly received Gaudi GPU line. If the limited rollout proves successful, Intel could secure a foothold in the fast‑growing AI infrastructure market.