Intel, in collaboration with Foxconn and SambaNova, unveiled rack-level AI infrastructure at Computex 2026 and simultaneously launched its first Xeon 6+ processor built on the Intel 18A process node. As AI shifts from training to inference, the CPU-to-GPU ratio is evolving from 1:4 toward 1:1. Intel is leveraging this trend to position itself as a central player in data center compute power during the AI inference era.
$Intel (INTC.US)$ The announcement of a strategic partnership with Foxconn at Computex 2026 to jointly develop AI infrastructure marks the formal implementation of strategic alignment between these two tech giants in the AI inference era. This collaboration spans the full technology stack—from chips to rack-level systems—and aims to capture market opportunities arising from the surge in AI inference demand for data centers.
At Computex 2026 in Taipei, Intel CEO Lip-Bu Tan and Foxconn Chief Product Officer Jerry Hsiao appeared together on stage. They announced that Intel, Foxconn, and SambaNova will jointly develop rack-level AI infrastructure tailored for deployment in data centers, hyperscale environments, and AI computing centers, with the underlying compute platform built on Intel Xeon processors.
The production-ready rack introduced through this collaboration integrates Intel Xeon processors with SambaNova’s SN-50 RDU, targeting high-performance AI inference with improved cost-efficiency and energy efficiency. Foxconn will serve as the system integrator and plans to separately manufacture CPU-intensive racks for workloads that do not require additional accelerators, covering use cases such as cost-optimized inference, data processing, and hybrid AI. Market sentiment toward Intel’s positioning in AI infrastructure was notably bolstered following the announcement.
Inference Demand Reshapes Data Center Landscape, Elevating CPU Importance
The AI industry is undergoing a structural shift from training-centric to inference-centric workloads—a trend that is redefining the logic behind compute resource allocation in data centers.
Ben Bajarin, CEO and Chief Analyst at Creative Strategies, noted that during the training era, the CPU-to-GPU ratio in AI deployments was approximately 1:4; however, with the rise of Agentic AI, this ratio is now moving toward 1:1 or even lower. This indicates a significant resurgence in the strategic value of CPUs within data centers.
Intel views this as a critical market entry window. In his keynote address, Lip-Bu Tan stated that with the emergence of AI inference, Agentic AI, and embodied AI, Intel is well-positioned to deliver end-to-end innovations—from chips to full systems—to the market.
Xeon 6+ Processor Launches, First to Use Intel 18A Process Technology
In parallel with the rack-level collaboration announcement, Intel officially launched the Xeon 6+ processor—the first data center CPU product built on Intel’s 18A process technology.
Xeon 6+ is specifically designed for cloud-native, agent-based AI, and network-intensive workloads, emphasizing sustained performance under real-world power constraints. According to Intel’s published specifications, a single liquid-cooled rack can deliver 36,864 cores within a 32U compute footprint, achieving the highest agent density at approximately 100 kilowatts of rack power consumption. The processor has been specially optimized for performance per watt, throughput per core, and predictable latency, enabling data centers to scale AI workloads without requiring disruptive infrastructure overhauls.
First Public Demonstration of Disaggregated Inference Architecture
At this year’s Computex, Intel also participated in the industry’s first live demonstration of a disaggregated inference system. The demonstration was led by Vector Core Compute, an enterprise inference cloud platform jointly established by Vista Equity Partners and Cambium Capital, and was streamed live from the Computex venue.
The system architecture leverages Intel Xeon 6 processors for orchestration and execution, SambaNova SN40 RDUs for decoding, and NVIDIA Blackwell GPUs for prefilling, all operating from Vector Core Compute’s data center in Los Angeles. Together.ai has become the platform’s first commercial customer, achieving the fastest enterprise-grade inference speed to date on the MiniMax 2.5 model across all existing architectures. Over 90 portfolio companies under Vista Equity Partners have already received early access to the platform, whose service network spans more than 2.5 million enterprise clients and 750 million end users.
Series 3 Platform Expansion Accelerates Edge AI Deployment
In the client computing segment, Intel Core Ultra Series 3 processors (built on the Intel 18A process node) continue to gain market traction and currently support more than 325 consumer and commercial PC designs. Intel also introduced its new Intel Arc G-series processors targeting the handheld gaming market, which will officially launch this month.
Edge deployment is advancing in parallel. Intel stated that the Series 3 platform will, for the first time, be rolled out simultaneously from the PC ecosystem to thousands of global edge customers. To date, over 130 customers have selected Series 3 for edge AI and robotics applications, spanning manufacturing, retail, smart cities, and robotics. Continued improvements in 18A process yield and growing customer demand are accelerating the expansion of the Series 3 product line.
Editor/KOKO