Intel will ship a limited run of its AI inference-focused GPU, Crescent Island, within this year, the Financial Times (FT) reported on Sunday.

Keivork Ketchichian (케보르크 케치치안), who leads Intel's data centre group, described Crescent Island as a product that came after 18 months of development and is focused on the inference stage rather than model training, the report said. That reflects a situation in which the training GPU Gaudi delivered weak results and development of a Gaudi successor was also cancelled, it said.

Ketchichian said, "We do not specifically target the training market."

Intel is emphasising costs as a point of differentiation from Nvidia and AMD for Crescent Island. It adopted much cheaper LPDDR5 memory instead of high-bandwidth memory used in Nvidia's Blackwell, among others. It also lowered infrastructure buildout costs by using air cooling instead of liquid cooling. Intel plans to produce the chip at its own foundry, and says it can further cut costs compared with rivals that rely on TSMC.

Intel is also reviewing whether to sell in China within the scope of complying with U.S. government export restrictions to China.

Intel shares have risen more than 200 percent since the start of the year after Lip-Bu Tan (립부 탄) became CEO. FT said the GPU launch is part of a product lineup overhaul being led by Tan.