Intel announced on Tuesday that it will launch Crescent Island, a new AI chip for data centers, in 2025 as part of its effort to reignite its artificial intelligence business and compete with market leaders Nvidia and AMD.
Revealed at the Open Compute Summit by CTO Sachin Katti, the chip will be optimized for inference workloads and designed to deliver strong performance per dollar with enhanced energy efficiency. “Instead of trying to build for every workload, our focus is increasingly going to be on inference,” Katti said.
Crescent Island will include 160GB of memory, albeit slower than the high-bandwidth memory (HBM) used in rival chips, and will be built on Intel’s consumer GPU architecture. The company did not disclose details of the manufacturing process but said it plans to launch a new AI chip every year, aligning its roadmap with that of AMD, Nvidia, and major cloud providers.
The move follows a series of AI setbacks for Intel, which shelved projects like Gaudi and Falcon Shores. CEO Lip-Bu Tan has vowed to rebuild Intel’s AI division and ensure its CPUs remain central to AI systems. The effort is bolstered by Nvidia’s recent $5 billion investment, giving it a 4% stake in Intel and forming a partnership to co-develop future chips.
Intel’s open, modular approach will allow customers to integrate processors from multiple vendors, a strategy the company hopes will differentiate it as AI demand reshapes the semiconductor industry.
