Intel announced at the 2025 OCP Global Summit that it will launch a new data center GPU, codenamed Crescent Island, in 2026 as part of its renewed push to compete in the AI accelerator space.
The Crescent Island GPU is being built for inference workloads in AI systems. It will use Intel’s new Xe3P architecture and is designed for air-cooled enterprise servers with power and cost efficiency in mind.
One standout feature is its memory configuration: the GPU will include 160 GB of LPDDR5X memory, a choice that favors energy efficiency and cost control versus more exotic high-bandwidth memory types.
Intel plans to begin customer sampling in the second half of 2026, giving partners time to test, validate workloads, and optimize with Intel’s software stack.
Strategically, Intel is shifting to a more predictable cadence for GPU releases—planning annual updates to match rivals—and focusing on inference (serving trained models) rather than competing directly in training workloads.
However, Intel faces stiff competition from entrenched players like NVIDIA and AMD. Its success will depend on how well the performance, price, and software ecosystem hold up in real-world deployments.
The Crescent Island GPU is being built for inference workloads in AI systems. It will use Intel’s new Xe3P architecture and is designed for air-cooled enterprise servers with power and cost efficiency in mind.
One standout feature is its memory configuration: the GPU will include 160 GB of LPDDR5X memory, a choice that favors energy efficiency and cost control versus more exotic high-bandwidth memory types.
Intel plans to begin customer sampling in the second half of 2026, giving partners time to test, validate workloads, and optimize with Intel’s software stack.
Strategically, Intel is shifting to a more predictable cadence for GPU releases—planning annual updates to match rivals—and focusing on inference (serving trained models) rather than competing directly in training workloads.
However, Intel faces stiff competition from entrenched players like NVIDIA and AMD. Its success will depend on how well the performance, price, and software ecosystem hold up in real-world deployments.