Intel will ship its new Crescent Island AI chip by year-end, targeting inference workloads where users make requests rather than model training. The chip uses LPDDR5 memory and air-cooling technology, significantly cheaper than the high-bandwidth memory and liquid-cooling systems required by Nvidia and AMD competitors.

The move represents Intel's strategic pivot after its previous Gaudi GPU failed to gain traction and its successor was cancelled. Kevork Kechichian, who leads Intel's data centre group, said the company is "starting with the basics" to rebuild its AI capabilities. This marks the first major AI infrastructure push under CEO Lip-Bu Tan, who replaced Pat Gelsinger amid investor concerns about the turnaround strategy.

Intel's shares have surged more than 200% this year as investors back Tan's cost-cutting approach. The company plans to manufacture the chip in-house, potentially undercutting rivals who rely on Taiwan Semiconductor Manufacturing Company. This vertical integration strategy could provide a significant cost advantage in the competitive AI chip market.

Intel is evaluating whether stripped-down versions could sell in China while complying with US export controls. With Nvidia and AMD blocked from that market due to trade tensions, Intel sees an opportunity at a price point attractive to Chinese customers.