In the first half of 2026, three vendors—Kontron, ADLINK, and ARBOR—announced Panther Lake-based platforms that illustrate precisely how this architecture is being applied to different edge AI requirements.

Kontron VX30101: 3U VPX for Tactical Edge and Mission Computing

Kontron announced the VX30101 on May 28, 2026, with market availability expected in Q3 2026 . It is a ruggedized 3U VPX single-board computer purpose-built for AI-enabled mission computing, sensor fusion, rugged embedded servers, and tactical edge applications in defense and aerospace . The SWaP-optimized (size, weight, and power) board combines the Panther Lake CPU's heterogeneous compute resources—P-cores, E-cores, LPE cores, Arc Xe3 GPU, and NPU 5.0—on a single card that meets the extended temperature, shock, and vibration requirements of military environments .

Kontron is among the first suppliers to bring Intel's Core Ultra Series 3 to the VPX ecosystem . In practice, the architecture allows a single VX30101 to run sensor fusion (radar, LiDAR, EO/IR) on the GPU, AI threat classification on the NPU, and secure command-and-control processing on the CPU cores, without needing separate inference accelerators or GPGPU cards on a backplane .

ADLINK Express-PTL: COM Express Type 6 for Automation and Machine Vision

ADLINK's Express-PTL is a COM Express R3.1 Type 6 Basic-size computer-on-module, announced in January 2026 with industrial-temperature SKUs expected to ship through Q2 2026 . A COM-HPC Mini variant is also planned for later in 2026 .

The module exposes the full Panther Lake heterogeneous architecture:

• CPU: Up to 16 cores in a 4P + 8E + 4LPE hybrid configuration, with a configurable TDP of 15–65W .
• GPU: Intel Arc Xe3 with up to 12 Xe cores, delivering up to 120 TOPS for computer vision, video analytics, and VLA model inference .
• NPU: Intel NPU 5.0, providing up to 50 TOPS of dedicated, power-efficient AI acceleration for always-on workloads such as anomaly detection, predictive maintenance, and voice commands .
• Total platform throughput: Up to 180 TOPS .

Memory support reaches 128 GB of DDR5 SO-DIMM at up to 7200 MHz, with in-band error correction (IBECC) available on select SKUs . Connectivity includes USB4, multiple display interfaces, and TSN-capable 2.5GbE for deterministic networking in factory automation environments .

The COM Express form factor means system integrators can pair the Express-PTL with a custom carrier board, partitioning AI workloads across the three accelerators while maintaining real-time control loops over TSN-enabled Ethernet. The industrial-temperature SKUs are explicitly targeted at fanless, ruggedized deployments in robotics, machine vision, and general embedded systems .

ARBOR IEC-6700: A 30W, Palm-Sized Box PC for Physical AI

ARBOR Technology previewed the IEC-6700 at COMPUTEX 2026 (June 2–5) and followed up with live demonstrations at Automate 2026 (June 15–16) . It is an ultra-compact, fanless Edge AI Box PC that uses the Intel Panther Lake-H processor to deliver 180 TOPS at just 30W—a significant departure from the large power budgets historically associated with inference at that throughput level .

What differentiates the IEC-6700 is its ability to merge Vision-Language-Action (VLA) processing with deterministic motion control on a single low-power platform . In practice, the NPU handles lightweight continuous inference tasks like object presence detection, the GPU accelerates heavier VLA model passes for scene understanding and grasp planning, and the CPU manages the deterministic motion control loop—all within the same compact chassis. This makes the system suitable for smart manufacturing, industrial robotics, smart cities, and what ARBOR and Intel call "Physical AI": robots that perceive, reason, and act in real time without relying on pre-programmed responses .

How the Integrated CPU + GPU + NPU Delivers Real-Time Edge AI

Across all three platforms, the fundamental architectural advantage is the same: Panther Lake's heterogeneous compute allows deterministic control, high-throughput AI inference, and low-power always-on processing to run concurrently on a single chip without requiring discrete PCIe accelerators.

• Hybrid CPU (P + E + LPE cores): The P-cores handle latency-sensitive control tasks and OS orchestration. The E-cores can be reserved for compute-bound workloads, while the LPE cores provide always-on, low-power background processing for tasks such as data logging and communication .
• Intel Arc Xe3 GPU: The GPU's 120 TOPS capacity is intended for parallel, high-throughput workloads: multi-stream video analytics, machine vision pipelines, VLA model inference, and radar/LiDAR sensor fusion. It can run multiple inference streams without blocking the CPU .
• NPU 5.0: The dedicated NPU is optimized for sustained, low-power AI inference—anomaly detection, acoustic event recognition, predictive maintenance, voice-command processing—tasks that benefit from being offloaded from the CPU and GPU entirely to reduce total system power .

The workload-partitioning model is not theoretical. The three platforms demonstrate it at different scales: ARBOR's IEC-6700 shows VLA and motion control at 30W, ADLINK's Express-PTL makes the partitioning flexible via a modular COM Express carrier-design approach, and Kontron's VX30101 puts the same model inside a MIL-grade ruggedized VPX card for defense. In each case, the outcome is a single-board system that replaces what previously would have been a multi-card, higher-power architecture, lowering latency and simplifying integration .

Intel's decision to certify these processors for embedded and industrial edge conditions—including extended temperature ranges, deterministic performance, and 24/7 reliability —from the moment of launch accelerates their adoption in environments where client-grade silicon was previously rare. As 2026 progresses, the pattern appears to be set: more edge AI tasks that once required discrete accelerators are migrating to a single SoC.

In the first half of 2026, three vendors—Kontron, ADLINK, and ARBOR—announced Panther Lake-based platforms that illustrate precisely how this architecture is being applied to different edge AI requirements.

Kontron VX30101: 3U VPX for Tactical Edge and Mission Computing

Kontron announced the VX30101 on May 28, 2026, with market availability expected in Q3 2026 . It is a ruggedized 3U VPX single-board computer purpose-built for AI-enabled mission computing, sensor fusion, rugged embedded servers, and tactical edge applications in defense and aerospace . The SWaP-optimized (size, weight, and power) board combines the Panther Lake CPU's heterogeneous compute resources—P-cores, E-cores, LPE cores, Arc Xe3 GPU, and NPU 5.0—on a single card that meets the extended temperature, shock, and vibration requirements of military environments .

Kontron is among the first suppliers to bring Intel's Core Ultra Series 3 to the VPX ecosystem . In practice, the architecture allows a single VX30101 to run sensor fusion (radar, LiDAR, EO/IR) on the GPU, AI threat classification on the NPU, and secure command-and-control processing on the CPU cores, without needing separate inference accelerators or GPGPU cards on a backplane .

ADLINK Express-PTL: COM Express Type 6 for Automation and Machine Vision

ADLINK's Express-PTL is a COM Express R3.1 Type 6 Basic-size computer-on-module, announced in January 2026 with industrial-temperature SKUs expected to ship through Q2 2026 . A COM-HPC Mini variant is also planned for later in 2026 .

The module exposes the full Panther Lake heterogeneous architecture:

• CPU: Up to 16 cores in a 4P + 8E + 4LPE hybrid configuration, with a configurable TDP of 15–65W .
• GPU: Intel Arc Xe3 with up to 12 Xe cores, delivering up to 120 TOPS for computer vision, video analytics, and VLA model inference .
• NPU: Intel NPU 5.0, providing up to 50 TOPS of dedicated, power-efficient AI acceleration for always-on workloads such as anomaly detection, predictive maintenance, and voice commands .
• Total platform throughput: Up to 180 TOPS .

Memory support reaches 128 GB of DDR5 SO-DIMM at up to 7200 MHz, with in-band error correction (IBECC) available on select SKUs . Connectivity includes USB4, multiple display interfaces, and TSN-capable 2.5GbE for deterministic networking in factory automation environments .

The COM Express form factor means system integrators can pair the Express-PTL with a custom carrier board, partitioning AI workloads across the three accelerators while maintaining real-time control loops over TSN-enabled Ethernet. The industrial-temperature SKUs are explicitly targeted at fanless, ruggedized deployments in robotics, machine vision, and general embedded systems .

ARBOR IEC-6700: A 30W, Palm-Sized Box PC for Physical AI

ARBOR Technology previewed the IEC-6700 at COMPUTEX 2026 (June 2–5) and followed up with live demonstrations at Automate 2026 (June 15–16) . It is an ultra-compact, fanless Edge AI Box PC that uses the Intel Panther Lake-H processor to deliver 180 TOPS at just 30W—a significant departure from the large power budgets historically associated with inference at that throughput level .

What differentiates the IEC-6700 is its ability to merge Vision-Language-Action (VLA) processing with deterministic motion control on a single low-power platform . In practice, the NPU handles lightweight continuous inference tasks like object presence detection, the GPU accelerates heavier VLA model passes for scene understanding and grasp planning, and the CPU manages the deterministic motion control loop—all within the same compact chassis. This makes the system suitable for smart manufacturing, industrial robotics, smart cities, and what ARBOR and Intel call "Physical AI": robots that perceive, reason, and act in real time without relying on pre-programmed responses .

How the Integrated CPU + GPU + NPU Delivers Real-Time Edge AI

Across all three platforms, the fundamental architectural advantage is the same: Panther Lake's heterogeneous compute allows deterministic control, high-throughput AI inference, and low-power always-on processing to run concurrently on a single chip without requiring discrete PCIe accelerators.

• Hybrid CPU (P + E + LPE cores): The P-cores handle latency-sensitive control tasks and OS orchestration. The E-cores can be reserved for compute-bound workloads, while the LPE cores provide always-on, low-power background processing for tasks such as data logging and communication .
• Intel Arc Xe3 GPU: The GPU's 120 TOPS capacity is intended for parallel, high-throughput workloads: multi-stream video analytics, machine vision pipelines, VLA model inference, and radar/LiDAR sensor fusion. It can run multiple inference streams without blocking the CPU .
• NPU 5.0: The dedicated NPU is optimized for sustained, low-power AI inference—anomaly detection, acoustic event recognition, predictive maintenance, voice-command processing—tasks that benefit from being offloaded from the CPU and GPU entirely to reduce total system power .

The workload-partitioning model is not theoretical. The three platforms demonstrate it at different scales: ARBOR's IEC-6700 shows VLA and motion control at 30W, ADLINK's Express-PTL makes the partitioning flexible via a modular COM Express carrier-design approach, and Kontron's VX30101 puts the same model inside a MIL-grade ruggedized VPX card for defense. In each case, the outcome is a single-board system that replaces what previously would have been a multi-card, higher-power architecture, lowering latency and simplifying integration .

Intel's decision to certify these processors for embedded and industrial edge conditions—including extended temperature ranges, deterministic performance, and 24/7 reliability —from the moment of launch accelerates their adoption in environments where client-grade silicon was previously rare. As 2026 progresses, the pattern appears to be set: more edge AI tasks that once required discrete accelerators are migrating to a single SoC.