Images and reporting cited by TrendForce indicate that Intel´s Jaguar Shores has surfaced as a rack-scale Artificial Intelligence test vehicle. Sources referenced in the article, including Wccftech and an editor at HardwareLUXX, suggest the design leverages Intel´s 18A process node and next-generation HBM4 memory. The visible package footprint measures roughly 92.5 mm by 92.5 mm, a size the report links to a high-performance computing orientation with a quad-tile layout and octal HBM memory subsystems. Wccftech further reports the design will use SK hynix HBM4 and integrate multiple domains and intellectual property blocks to support large-scale training workloads.
The appearance of Jaguar Shores follows a difficult period for Intel´s Artificial Intelligence hardware efforts. The article notes that Gaudi 3 underperformed against its 2024 sales target, and Michelle Johnston Holthaus confirmed in early 2025 that Falcon Shores was scaled back to an internal engineering project and would no longer pursue the datacenter GPU market. Despite those setbacks, Jaguar Shores is presented as a restart for Intel´s flagship Artificial Intelligence chip ambitions and is positioned to compete with offerings from NVIDIA and AMD. The report also highlights that Intel´s 2024-launched Gaudi 3 used eight HBM2E devices, implying Jaguar Shores would represent a substantial memory performance leap.
Beyond the chip itself, the article describes a broader strategic shift at Intel. After appointing Sachin Katti as head of Artificial Intelligence in April, Intel´s AI unit is said to be focusing on workload-specific solutions spanning low-power edge inference to rack-scale data center training, according to Tom´s Hardware. TrendForce´s coverage states Intel is simultaneously exploring custom silicon partnerships and new architectures while leveraging its strengths in x86 processors, IPUs, multi-chiplet designs, advanced packaging, and silicon photonics. Details on Jaguar Shores remain limited, but the report frames it as central to Intel´s renewed push into rack-scale Artificial Intelligence hardware.
