Abaco Systems has opened early access for the SBC3618, a new single-board computer that packs Intel Core Ultra Series 3 processors with integrated neural processing, 100-gigabit Ethernet, and alignment to SOSA compute-intensive profiles. The board won’t enter full production until the first quarter of 2027, but developers and defense integrators can get their hands on it now.

The Hardware: What’s Inside the SBC3618

The SBC3618 is built around Intel’s Core Ultra Series 3, a processor family that combines CPU, GPU, and NPU (neural processing unit) on a single die. For edge workloads—especially inference on sensor data—that on-chip AI accelerator is the headline feature. It means power-efficient machine learning without a discrete GPU or external FPGA, lowering SWaP (size, weight, and power) in the kinds of rugged chassis where board real estate is tight.

Abaco aligned the board with the Sensor Open Systems Architecture (SOSA) compute-intensive profile. SOSA is a modular, plug-and-play framework backed by the U.S. Department of Defense, and the compute-intensive profile is designed for high-throughput signal processing, radar, and electronic warfare applications. Practically, that means the SBC3618 fits into existing SOSA backplanes with defined slot profiles, using open-standard connectors and pinouts.

Networking is handled by a 100-gigabit Ethernet interface, which is fast becoming the baseline for tactical data links and multi-sensor fusion. Compared to the 10G or 40G ports still common on many embedded boards, 100G removes the bandwidth bottleneck when pulling in high-resolution video, lidar, or spectral data from multiple sensors simultaneously.

Early Access: What the Program Means for Developers

According to Abaco’s announcement, early access is open now. That typically includes evaluation hardware, a board support package (BSP), and access to engineering support for integration. For teams building long-lead defense programs—think ground vehicle modernization, shipboard processing, or airborne pods—starting evaluation in 2025 gives them roughly two years to certify the hardware before production units ship in Q1 2027.

For Windows-focused shops, the board likely targets Windows 11 IoT Enterprise or Windows Server IoT. Intel’s Core Ultra drivers already support Windows, and the NPU acceleration is accessible through DirectML, ONNX Runtime, and Intel’s OpenVINO toolkit. That means developers can use familiar Windows ML frameworks without porting to a proprietary RTOS. Linux and VxWorks are almost certainly part of the BSP as well, as they remain dominant in defense embedded systems.

Why This Board Matters Now

The SBC3618 arrives at a moment when three trends are colliding. First, the Pentagon’s push for modular open systems is accelerating, and SOSA is the de facto standard. Every new major program—from the Army’s ground combat vehicles to the Navy’s surface combatants—requires SOSA-aligned hardware in RFPs. Second, edge AI is moving from lab demonstrations to operational reality. Troops need real-time object detection, automatic target recognition, and RF signal classification, all at the tactical edge with intermittent connectivity. An NPU that can run these models at low power, inside a rugged SBC, is a direct answer to that need.

Third, Intel’s Core Ultra launch in 2024 brought AI acceleration to consumer and commercial PCs, but defense applications lag because of lengthy qualification cycles. Abaco, as a major supplier of rugged board-level products for the DoD, bridges that gap by delivering Core Ultra in a SOSA form factor now, so programs can bake it in early.

How We Got Here: SOSA and the Shift to Open Standards

The Sensor Open Systems Architecture was conceived to break the cycle of proprietary, single-vendor hardware lock-in that plagued military systems for decades. Before SOSA, upgrading a radar processor often meant redesigning an entire subsystem because connectors, cooling, and software were all custom. SOSA defines hardware, software, and electrical interfaces within a modular chassis, so a new SBC from any vendor can slide in and work—provided it meets the profile.

The compute-intensive profile is one of several defined under SOSA, targeting workloads that demand high floating-point performance and memory bandwidth. Previous SOSA boards often used Intel Xeon D or Xeon E, sometimes with FPGA accelerators. The SBC3618 represents a shift toward a more integrated architecture, where the CPU, GPU, and AI accelerator share a single thermal and power envelope. That’s easier to cool in conduction-cooled formats and simplifies software management because there’s one processor vendor to support.

What to Do Now if You’re Evaluating Edge AI Hardware

If you’re a systems integrator or engineering lead for a defense program that will field in the late 2020s, request an early access kit now. Here’s a checklist:

  • Contact Abaco: Reach out through their sales channel to get on the early access list. Expect to sign a non-disclosure agreement.
  • Assess software compatibility: Confirm that your chosen ML framework (TensorFlow, PyTorch, ONNX) can leverage the NPU via DirectML or OpenVINO on Windows, or via the Intel NPU driver stack on Linux.
  • Review thermal and power envelopes: The SBC3618’s exact TDP isn’t disclosed, but Intel Core Ultra Series 3 typically ranges from 15W to 28W in commercial SKUs. In a conduction-cooled setting, that’s manageable, but verify with your chassis supplier.
  • Plan for 100G infrastructure: To fully utilize the network interface, your backplane and switches must support 100G. Check your existing SOSA chassis or plan for upgrades.
  • Start qualification documentation early: Capture environmental test results, BSP maturity, and security certifications now, so you’re ready for production in 2027.

For Windows administrators less familiar with the defense supply chain, this board might not directly land in your data center. But the underlying technology—Intel Core Ultra with NPU—is already in commercial laptops and coming to industrial PCs. Understanding its capabilities helps you prepare for next-gen edge servers that run Windows Server IoT and do AI inference on camera streams or IoT sensor data in a factory or warehouse.

Outlook: The Road to 2027 and Beyond

Between now and Q1 2027, expect Abaco to iterate on firmware and possibly release follow-on variants with different I/O mixes. The defense community will be watching early access feedback closely, as the SBC3618’s success could determine whether Intel’s AI PC silicon becomes a staple of tactical edge computing or remains a niche. Competitors like Mercury Systems and Kontron will likely announce their own SOSA boards with similar processors. For now, Abaco is first to market with a Core Ultra SOSA board, and that two-year evaluation window is a strategic asset for any program office willing to commit.