Particle.news
Download on the App Store

Intel Shifts to Lower-Cost AI Inference Chips as NVIDIA Escalates PC Push

The move narrows Intel’s AI focus to inference, accelerates in-house fabrication, opens the door to export‑compliant China variants.

Overview

  • Intel unveiled Crescent Island, a new GPU aimed at inference workloads, and said it plans limited shipments by the end of 2026.
  • The chip is designed to cut cost by using LPDDR5 memory and standard air cooling instead of high‑bandwidth memory and liquid cooling used in top-tier training GPUs.
  • Intel’s data‑center chief said the company stepped back from chasing large‑model training after problems with its Gaudi program and is concentrating on the lower‑risk inference market.
  • Under CEO Lip‑Bu Tan, Intel is pushing more production into its own foundries to lower costs and gain control over manufacturing for data‑center products.
  • NVIDIA’s same‑day RTX Spark Superchip announcement intensified competitive pressure, investors knocked Intel’s stock about 5% in premarket trading on Monday, and analysts remain broadly cautious with a consensus Hold and an average target near $77.65.