Intel targets inference market with cost-focused 'Crescent Island' AI chip

Intel has announced plans to release its 'Crescent Island' artificial intelligence graphics processing unit by the end of 2026, a move designed to challenge the dominance of rivals Nvidia and AMD in the semiconductor market. The chip is specifically engineered for inference tasks, which involve processing user requests after a model has been trained, rather than the resource-intensive training phase where Nvidia currently holds a commanding lead. Kevork Kechichian, head of Intel’s data centre group, stated that the company is “starting with the basics” to rebuild its capabilities in the AI sector, acknowledging past struggles with its 'Gaudi' training GPU, whose successor was cancelled last year.

The 'Crescent Island' architecture differentiates itself by avoiding the expensive high-bandwidth memory and liquid-cooling infrastructure required by competitors. Instead, the chip utilises LPDDR5 memory and an air-cooling design, offering a lower-cost alternative for data centres. This approach addresses two significant constraints faced by users of Nvidia’s Blackwell chips and AMD’s offerings. The development process took 18 months, and initial shipments will begin in limited quantities. Kechichian, who joined Intel from chip designer Arm last year, noted that the company is not currently aiming for the training market based on previous experience, but is focusing on establishing a foothold in inference.

This product launch represents Intel’s first major push into the AI infrastructure market under chief executive Lip-Bu Tan, who assumed leadership last year following the ousting of Pat Gelsinger. Tan’s strategy has focused on cost-cutting and rein in spending, a shift that investors have welcomed. Intel’s shares have risen more than 200 percent since the start of 2026, contributing to a broader rally in semiconductor stocks driven by enthusiasm for artificial intelligence. The company previously unveiled the GPU in October as part of a broader effort to revive its product line-up and regain ground in a market where Nvidia has dominated the training segment for models such as OpenAI’s ChatGPT.

Intel intends to manufacture the 'Crescent Island' chip in-house at its own foundries, a strategic shift from its historical reliance on Taiwan Semiconductor Manufacturing Company. Kechichian confirmed that the company is moving aggressively into its own foundry for data centre products to reduce costs. This aligns with recent developments where Intel launched advanced PC and server chips built in its own factories this year. The move to in-house manufacturing is expected to further lower production costs compared to rivals who rely on external foundries, reinforcing the chip’s value proposition as a cost-effective solution.

The company is also assessing potential sales in China, contingent upon US export controls. Kechichian indicated that there are tiers of the chip that might be permissible for sale in the Asian nation, where trade tensions have blocked Nvidia and AMD’s AI chip sales. “Clearly there is demand for that particular price point in that particular market,” Kechichian said, suggesting that Intel could capture market share in China if regulatory frameworks allow. The US government is set to take a 10 percent stake in Intel over time, a measure announced in August to deter the chipmaker from selling its foundry business, further solidifying the company’s role in the national semiconductor supply chain.