Intel used its Computex 2026 keynote in Taipei to make a case that the shift from AI model training to AI inference, and the rise of agentic workloads is putting the CPU back where it once stood: at the centre of the data centre.
CEO Lip-Bu Tan led the announcements, framing them as Intel’s response to a structural change in how AI compute is consumed. “With the rise of inference, agentic, and physical AI, Intel is poised to bring the world new innovations from the chip to systems level,” Tan said.
The infrastructure bet
The announcement was a rackscale AI infrastructure built in partnership with SambaNova and Foxconn. The system combines Intel Xeon processors with SambaNova SN-50 Reconfigurable Dataflow Units, designed for high-performance inference at data centre scale. Foxconn will handle system integration and plans to manufacture a CPU-dense variant for workloads that do not require additional acceleration, covering cost-optimised inference, data processing, and hybrid AI.
The strategic logic behind this is grounded in a shift in how AI deployments are structured. During the training era, data centre configurations leaned heavily toward GPUs. As inference and agentic AI workloads scale, that balance is shifting. According to Creative Strategies analyst Ben Bajarin, the training-era ratio of roughly one CPU to four GPUs is moving toward approximately one CPU to one GPU — or less — in agentic inference environments. Intel is positioning itself to capture that shift at the systems level, not just the chip level.
Also announced was Vector Core Compute, a new enterprise inference cloud backed by Vista Equity Partners and Cambium Capital running a fully disaggregated inference system using Intel Xeon 6 processors, SambaNova RDUs, and NVIDIA Blackwell GPUs simultaneously. Together.ai is its first commercial customer, and the system delivered what the companies claimed was the fastest enterprise inference on the MiniMax 2.5 model of any architecture to date.
Xeon 6+: The chip built for agentic density
At the chip level, Intel announced Xeon 6+ processors — its next-generation data centre CPU built on Intel 18A, the same process node underpinning its consumer Series 3 lineup. The processor is engineered for the orchestration, concurrency, and data movement demands of agentic AI, with a single liquid-cooled rack capable of delivering 36,864 cores in 32U of compute space. Intel described this as the highest agent density currently available at approximately 100-kilowatt rack power.
From Aerospace to Brain-Computer interfaces
Beyond infrastructure, Intel announced a set of vertical industry partnerships that signal where it sees the next layer of AI value being built — in domain-specific silicon and solutions rather than general-purpose compute.
Foxconn and Siemens extended existing collaborations into chip design and manufacturing. Siemens, specifically, is working with Intel across the full value chain from design to fabrication, exploring purpose-built silicon for edge devices, high-performance computing, and robotics. Hitachi is working with Intel on foundry tools and quantum computing. Two more niche but forward-looking partnerships round out the list — Echo Neurotechnologies on neuromorphic computing and brain-computer interfaces, and Greenstone Biosciences on AI-accelerated drug development using stem cells and genomics.
For CIOs evaluating where Intel sits in their AI infrastructure decisions, the breadth of these partnerships carries depicts that Intel is trying to build vertical relevance in the industry rather than just horizontal compute scale.