The x86 giant pivots to AI inference with cost-competitive Gaudi 3 accelerators and enterprise-focused Crescent Island GPU roadmap
AI INFRASTRUCTURE
INTEL
The x86 giant pivots to AI inference with cost-competitive Gaudi 3 accelerators and enterprise-focused Crescent Island GPU roadmap
Intel repositioned its AI strategy in 2025 under new CEO Lip-Bu Tan, focusing on AI inference workloads and cost-competitive solutions rather than competing directly in the training-dominated GPU market. The Gaudi 3 accelerator, available in both PCIe and rack-scale configurations, emphasises price-performance advantages with projected 92 % better cost efficiency than Nvidia H200 while using standard Ethernet networking to avoid proprietary infrastructure lockin. IBM Cloud became the first major service provider to deploy Gaudi 3 at scale in May 2025, followed by Dell AI Factory integration. At OCP Global Summit 2025, Intel unveiled Crescent Island, a next-generation data centre GPU with 160GB memory based on Xe3P microarchitecture, targeting customer sampling in H2 2026. The company’ s strategic collaboration with Nvidia, announced in September 2025 with US $ 5bn Nvidia investment, positions Intel to manufacture custom x86 CPUs with NVLink integration, creating hybrid architectures that leverage both companies’ strengths while Intel expands its foundry services business model.
“Intel was once widely seen as the world’ s most innovative company. There’ s no reason we can’ t get back there, so long as we drive the changes needed to improve”