Intel said more than 130 customers are using the Intel Core Series 3 processor family to power edge devices including robotics and AI applications.
At Computex, intel said the momentum includes a major design with Sensory AI for Ella the first multiagent Physical AI store running in public commercial service, which is moving to Intel
architecture.
To help the robotics industry turn development designs into deployed fleets, Intel also introduced OpenVINO Physical AI, an Intel-optimized, open-source framework that addresses the challenges of cost-effective deployment and scale.
“Physical AI models are transforming robotics, but deployment has been slowed by fragmented
software stacks and one‑off integrations for every robot. With Intel Core Ultra Series 3 and OpenVINO Physical AI, we provide a unified, open, and scalable path from AI experimentation to production-grade robots delivering hardware-accelerated, high-performance inference,” said Dan Rodriguez, corporate vice president of the edge computing group at Intel, in a statement.
Demonstrating the impact of Intel Core Ultra Series 3 and Intel Robotics solutions, Sensory AI
has replaced a fragmented CPU plus discrete accelerator architecture in Ella with a single
Intel Core Ultra Series 3 platform for both real-time control and AI.
The result is a multi-agent Physical AI store where three specialized AI agents – Avatar, Guardian, and Ella Agent – run concurrently on a single SoC, handling customer conversation, system operations, and storelevel business intelligence, while a deterministic orchestrator commands the robot. This eliminates an entire class of components, cuts software complexity, increases ROI, and creates a cleaner path to scale new robot designs on future Intel platforms.
OpenVINO Physical AI: Bridging the Gap from Lab to Factory Floor
Intel introduced the Physical AI Studio and OpenVINO Physical AI as part of its Robotics AI Suite to help developers and operations teams move from experimental models to scalable, realworld deployments.
Physical AI Studio enables developers to do data collection, model fine tuning, optimization and quantization, and export pre-validated, fine-tuned VLA models for deployment. OpenVINO Physical AI is the first open-source robotics library with a silicon optimized inference runtime. It provides developers with a consistent way to take robot policies and multimodal models from experimentation into working robot systems while maximizing inference performance. It also offers integration with open-source robotics model development environments like Physical AI Studio and LeRobot for seamless model export.
The Missing Link for Robotics and Edge AI
Operations leaders facing labor shortages, rising costs, and competitive pressure are accelerating investments in robotics and automation to keep production lines running and
improve throughput. At the same time, robotics platforms are shifting from deterministic systems to autonomous physical AI systems that must perceive, reason, and act safely in the
real world with millisecond‑accurate timing.
Until now, deploying physical AI models at scale has required highly customized pipelines for
each robot to handle sensors, codecs, inferencing loops, and actuation, often locking customers into over‑provisioned, dual‑compute solutions that are expensive to deploy and maintain. By pairing Intel Core Ultra Series 3 with the OpenVINO Physical AI, Intel offers a unified hardware‑software stack that helps customers lower total cost of ownership, reuse more of their code across different robot types, and scale fleets across factories, warehouses, and retail environments more efficiently.
The Intel Robotics portfolio includes new software tools as part of Intel Robotics AI Suite, Robotics development kits from Intel partners, and Intel Core Ultra Series 3 and Intel Core Series 3 processors designed specifically for robotics and edge AI applications.
Availability
OpenVINO Physical AI preview available on GitHub with general availability 2H 2026; Physical AI Studios available now.
Crescent Island: Building Momentum in AI Inference
To meet the growing demands for agentic AI—memory capacity, bandwidth and efficiency are emerging as critical differentiators alongside performance. Purpose-built to address these needs, Intel’s next-generation data center GPU, code-named Crescent Island, built on the Xe 3P architecture extends the proven Xe architecture delivering enhanced efficiency and performance-per-watt while maintaining broad software compatibility for modern AI workloads.
Equipped with LPDDR5x memory, Intel’s Crescent Island delivers up to 480 GB capacity to efficiently handle large, token-intensive workloads while reducing total cost of ownership. Its power efficient 350W air-cooled PCIe design enables highly efficient scaling for agentic AI with strong performance-per-watt.
Leveraging a multi-generational Xe install base, Intel’s Crescent Island is designed for next generation AI workloads with support for a wide range of datatypes and microscaling formats, from native FP4/MXFP4 to FP64, including expanded support for advanced AI operations and improved memory and scalability.
Intel’s open programmable AI software stack supports a heterogeneous compute platform
designed to reduce friction and enable AI deployment at scale by providing out-of-thebox model support with an upstream-first approach. Built on the same Xe architecture foundation, Intel’s Arc Pro Series provides an ideal development platform allowing developers to build, validate and optimize workloads on familiar hardware and seamlessly deploy on Crescent Island with forward and backward compatibility.
Intel Xeon 6+ processors
Intel also announced data center advancements, including new Intel Xeon 6+ processors. With Xeon 6+ at the center, AI becomes more agentic, the CPU is re-emerging at the center of modern AI infrastructure.
With Xeon serving as the control plane, Intel is taking a systems-level approach to performance and efficiency at scale —delivering platforms designed for increasingly agentic AI workloads, where orchestration, data movement, and sustained inference are critical across data center and network environments.
“AI doesn’t scale as a collection of parts—it scales as a coordinated system,” said Kevork Kechichan, executive vice president and general manager of Intel Data Center Group, in a statement. “As AI becomes more agentic, the constraints shift to orchestration, concurrency, and data movement. That shift reinforces a core reality: the CPU remains the control plane for the modern AI infrastructure. With Xeon 6+ and Ethernet E835, we’re tightly coupling compute and networking to reduce bottlenecks and enable efficient, secure scaling of real-world agentic workflows.”
Intel Xeon 6+ processors extend the Xeon 6 family with a focus on performance density,
power efficiency, and operational scale for cloud-native, agentic AI-driven, and
network-intensive workloads. Built on Intel 18A —its first use in a data center CPU—
Xeon 6+ is engineered for sustained performance under real-world power constraints—
addressing orchestration, concurrency and data movement demands of emerging agentic
AI.
Optimized for environments where watts per rack, throughput per core, and latency predictability are critical, Xeon 6+ emphasizes scale-out performance — making room
for new AI workloads without requiring disruptive data center redesigns.
Key highlights include:
- Up to 288 Efficient-cores, delivering up to 2.5 times more performance1
compared to the previous generation, and up to 45% better performance per
thread per watt versus the competition – enabling high concurrency and strong
responsiveness for cloud-native, telecom, and agentic AI-driven workloads. - 12-channel DDR5 memory with scalable bandwidth for high-density systems
- 96 lanes of PCIe Gen 5 and CXL support to accelerate data movement across
heterogeneous infrastructure. - Intel Application Energy Telemetry (AET) enables real-time workload-level
CPU energy and activity telemetry, improving visibility into energy consumption
at the workload-level starting with Intel Xeon 6+ processors. - Up to 9:1 server consolidation, reducing footprint and total cost of ownership
vs. 2nd Gen Intel Xeon. - Security built into silicon, including Intel SGX and Intel TDX, to support
confidential and multi-tenant deployments.
Intel Xeon 6+ processors are already being tested within telecom network infrastructures
and configured into data center systems with platforms available across the ecosystem.
These include servers, networking and integrated solutions from and used by Asus, Dell
Technologies, Ericsson, Gigabyte, HPE, Lenovo, Supermicro – and others developing
on Xeon 6+ today.
This growing portfolio reflects Intel’s systems‑first approach—delivering deployable,
available‑now infrastructure for running, scaling, and orchestrating increasingly agentic
AI workloads on x86. Paired with complementary Xeon platforms optimized for both
high‑density throughput and single‑thread performance, Intel’s customers and partners can balance efficiency and responsiveness by distributing workloads across a proven, broad, mature hardware and software ecosystem.
Intel Ethernet E835: High-Efficiency Networking for Next-Gen Infrastructure
As AI, cloud, and distributed workloads continue to scale, networking has become a
critical determinant of overall infrastructure performance and efficiency. The Intel
Ethernet E835 controllers and network adapters are designed to deliver performant,
power-efficient connectivity for modern data center, enterprise, edge, and AI
environments.
The Intel Ethernet E835 provides the scalability and efficiency required for nextgeneration infrastructure while maintaining industry-leading performance-per-watt.
Designed for dense, virtualized deployments, the E835 helps reduce energy consumption
and operational costs without compromising throughput or reliability.
Key highlights include:
- Flexible Connectivity: Delivers 200 GbE throughput with multiple controller and
adapter configurations supporting data rates from 10GbE to 200GbE. The 835
supports a broad range of port configurations, including 2x25GbE, 4x25GbE,
2x100GbE, and 1x200GbE, with additional configurations enabled through the
Intel Ethernet Port Configuration Tool (EPCT). - Industry-leading Power Efficiency: Engineered for high performance-per-watt,
Intel E835-CQDA2 network adapter delivers up to 1.9 times higher performance
per watt than the comparable NVIDIA ConnectX-6 DX (CX614106A) and 1.4
times higher than Broadcom BCM957508-P2100G, lowering energy consumption
and operational costs of modern distributed environments.* - Network Optimization: Implements RDMA (RoCEv2/iWARP) to reduce CPU
utilization and maximize efficiency – and Dynamic Device Personalization to
streamline packet processing and improve application performance. - Security & Management: Integrates Hardware Root of Trust and signed SPDM
with DMTF-based manageability for secure, deterministic operations. - Broad Compatibility: Supports multiple operating systems including Linux,
ESXi and Windows. - 10+ Year Lifecycle: Built for long-term reliability and support.
With broad support from industry leaders—including Cisco, Dell, HPE, Lenovo, and
Supermicro—the Intel Ethernet E835 provides an efficient and manageable networking
fabric. From AI training to enterprise cloud services, the E835 delivers the scalability,
reliability and high-efficiency features required for the next generation of networking.
More Performance for SMB Entry Servers
Intel also announced the general availability of a new 12-core option in the Intel Xeon
6300 processor family for entry servers that raises the platform ceiling beyond 8 cores for
the first time. The added core count provides greater compute power and flexibility for
growing SMB workloads—without requiring a platform change.
Available today through major OEMs, the Xeon 6300 12-core processor is drop-in compatible with existing entry server designs, enabling fast, cost-effective upgrades.