ESD eBook September/October 2026: Rethinking Mil/Aero System Design
Designing SWaP-Optimized Compute for UAVs
Dell’s new Pro 5 16 AMD laptop, powered by the Ryzen AI 400 series, marks a pivotal milestone in AMD’s mobile CPU turnaround by achieving parity treatment with Intel in Dell’s corporate laptop lineup, reflecting a shift in OEM perception from budget filler to a viable business platform. Eight years after AMD’s first mobile Ryzen silicon, the company now enjoys equal support and fleet-management features across Dell’s Pro product lines, where customers can choose between Intel and AMD chips for the same laptop model. The reviewed Dell Pro 5 16 (P516265) features AMD’s top Gorgon Point silicon, the Ryzen AI 9 HX PRO 470 with 4 Zen5 + 8 Zen5c cores, 64GB DDR5-5600 memory, a 1TB PCIe Gen4 SSD, and integrated RDNA 3.5 Radeon 890M graphics. Its 16-inch 2560×1600 120Hz VRR display (500 nits, 100% sRGB) and 70Wh battery target business users needing performance and modularity. Notably, Dell includes two USB-C/Thunderbolt 4 ports via an external controller—rare for an AMD laptop—plus a 1GbE RJ45 Ethernet jack, Wi-Fi 7, and a physical camera shutter. The chassis weighs 1.82 kg and measures 18.95 mm thick, with a full numeric keypad and optional IR camera for Windows Hello. This mainstream business laptop demonstrates AMD’s hard-won credibility in the mobile CPU space, where repeated refinements have made its chips fierce competitors to Intel, especially in the corporate segment where OEMs now treat AMD as a primary platform option rather than a secondary choice.
Infineon at Hot Chips 2026 argued that RISC-V is the ideal ISA for automotive zone controllers to avoid IP lock-in, but stressed that market success hinges on auto-qualified toolchains, SoC architecture, and ecosystem enablement rather than the instruction set alone. The talk addressed the compute demands facing automotive microcontrollers, where chassis, powertrain, and ADAS applications require real-time low latency, power efficiency, security, and functional safety on a single part. Infineon detailed the architectural shift from domain controllers to zone control units and eventually full car computers, consolidating body and comfort functions to reduce wiring harness weight and cost. Real-time requirements vary: fast control loops like E-motor control need deadlines under 10 ms and nanosecond interrupt latencies, while strategic functions tolerate seconds. For zone architecture, Infineon favors Path A—optimized multi-domain zone controllers with local intelligence and power distribution—over Path B’s low-complexity I/O aggregators. This approach shrinks attack surfaces, eases maintenance, and cuts system cost. Zone controllers must handle heterogeneous workloads: real-time control, DSP, AI inference, low-power services, and audio processing. Super-integrated zone controllers add an MPU island to offload non-real-time tasks, enabling faster boot and scaling from L2 to L2+ and L3. Infineon promotes RISC-V for its open, scalable ISA that avoids IP lock-in and enables ecosystem synergies across automotive, IoT, and industrial use. Standardization requires unified profiles (e.g., RV32E with optional supervisor mode, vectors, hypervisor). However, Infineon notes that ISA details matter less than microarchitectural optimization; memory, connectivity, BOM cost, and automotive-qualified toolchains determine success. Long-term RISC-V adoption in zone controllers depends on standardized hardware-software interfaces and toolchain support.
d-Matrix has announced it will integrate its next-generation XPUs with NVIDIA’s NVLink Fusion platform, a move that standardizes scaling for AI inference accelerators while giving NVIDIA additional interconnect and CPU socket wins. The core news, revealed this week, positions d-Matrix’s Raptor 3D-DRAM accelerator—first detailed at Hot Chips 2026—to scale beyond single XPUs into larger clusters using NVLink Fusion technology within MGX racks. The announcement also confirms d-Matrix will leverage NVIDIA Vera CPUs, ConnectX-9, BlueField-4 DPUs, NVLink switch trays, and Spectrum-X switching, replacing its earlier JetStream 400G Ethernet card approach for data-center-scale AI inference. By adopting NVLink Fusion, d-Matrix avoids reinventing networking on FPGAs and gains a rack-scale platform familiar to hyperscalers and neoclouds for deployment and maintenance. For NVIDIA, every d-Matrix customer win means a CPU socket that AMD, Intel, Qualcomm, and the Arm AGI CPU do not capture, along with the accompanying interconnect revenue. The partnership underscores d-Matrix’s growing momentum in the AI accelerator market, offering a standardized path to scale out generative inference workloads while deepening NVIDIA’s ecosystem lock-in.
ASUS Pro WS W890E SAGE SE motherboard for Intel Xeon 600 (Granite Rapids) workstation platform launches as the premium DIY option, featuring extensive PCIe Gen5, 8-channel DDR5 memory, and overclocking support. The ASUS Pro WS W890E SAGE SE is the flagship motherboard for Intel’s new Xeon 600 workstation processors, which use the LGA4710-2 socket and W890 chipset. It is one of only two ASUS boards for this platform and the only Xeon 600 motherboard readily available in the North American retail market, making it the de facto choice for DIY workstation builders.
The board uses the full EEB (E-ATX) form factor (12x13 inches) to accommodate a massive 16-phase VRM rated for CPUs up to 420W TDP, plus an optional AngleBoost fan kit to cool VRMs and all eight DDR5 RDIMM slots. It offers seven PCIe x16 slots: six running Gen5 x16 electrically and one Gen5 x8, with CXL 2.0 support on slots 1, 3, 5, and 7. Storage includes four M.2 Gen5 x4 slots (two 22110, two 2280), one MCIO connector (Gen5 x4/x4), two SlimSAS Gen4 x4 ports, and four SATA 6Gbps ports. Networking features dual 10GbE via Intel E610-XAT2, and a BMC (ASPEED AST2600) for remote management. Rear I/O includes two 40Gbps USB-C and six 10Gbps USB-A ports.
ASUS has added robust overclocking features to leverage the unlocked X-series Xeon SKUs, positioning the W890E-SAGE SE as a high-end workstation board that balances enterprise functionality with enthusiast flexibility.
The 69th IT Press Tour in Ljubljana, Slovenia, early September showcased six companies—Cybee, DataCore, ExpectedIT, HYCU, StratoFoundry, and Veridion—addressing IT infrastructure, cloud, cybersecurity, data management, and AI. Cybee, a French startup founded in December 2022, unveiled a cloud-native backup and cyber-resilience platform based on the open-source Restic format, offering immutable, encrypted backups with post-quantum resistance and restore speeds up to 4TB/hour per disk. Targeting regulated sectors (healthcare, finance, transport), it avoids vendor lock-in and complies with NIS2, DORA, and ISO 27001. Pricing starts at €5/VM/month, with managed DRaaS via sister company Nuabee in H2 2026 and Orange Business expansion in H1 2027.
DataCore presented its software-defined storage strategy, unifying its portfolio (acquired Caringo, MayaData, StarWind, etc.) and introducing agentic AI for IT operations. The platform integrates block (SANsymphony), object (Swarm), edge HCI (StarWind), parallel file (Nexus), and Kubernetes-native (Puls8) storage, aiming to autonomize infrastructure with AI-driven capacity forecasting, performance optimization, and cyber resilience validation.
ExpectedIT, a German deep-tech startup, introduced LPool, a chip architecture pooling CPU, memory, storage, and networking across a server rack, targeting agentic AI bottlenecks. Vendor-neutral (Intel, AMD, ARM), it aims for 10-socket tapeout in 2027, scaling to 96 sockets with PCIe Gen 7. Potential applications include KV-cache head nodes with up to 7× throughput. The company projects €398M revenue by 2032, with a $150B TAM, and has seed funding of €6.5M.
HYCU positioned itself as an AI resilience company, addressing data protection risks from AI agents—drift, misinstruction, and standing credentials—that operate at machine speed inside perimeters. The company argues traditional SaaS retention mechanisms are inadequate for automated agent-scale modifications.
Qualcomm has disclosed key details of its next-generation premium mobile Snapdragon platform ahead of the Snapdragon Summit, revealing the Oryon CPU, Adreno GPU, and Hexagon NPU designed for the agentic AI era. The Oryon CPU features two Prime cores rated at 5 GHz, marking the first mobile CPU to reach that frequency, with FlexCache dynamically allocating cache across heterogeneous cores to reduce system-memory accesses. The Adreno GPU includes three slices clocked at 1.45 GHz, 18 MB of High Performance Memory (HPM) for local graphics storage, and Matrix Cores for dedicated AI processing, delivering a 12% power improvement over the Snapdragon 8 Elite Gen 5 baseline. Neural Fusion AI rendering achieves 40% power savings in internal demos. The Hexagon NPU adds a new Element Accelerator for transformer operations, supports context lengths up to 32K with KV-cache acceleration, and increases shared memory by 50% to keep model state near accelerators. For INT4 models, Qualcomm reports up to 50% prefill uplift versus the previous generation, and Mixture of Experts models up to 30 billion total parameters run on the NPU. The Sensing Hub routes voice input to the Oryon CPU, which orchestrates task execution across GPU, NPU, or cloud. The platform emphasizes keeping data and computation close to processing resources to reduce reliance on slower system memory.
Neura Robotics and Seco Partner to Scale Physical AI
Bridging the HPC Software Gap for Practical Quantum Computing
Mission-Critical Systems: MCUs vs. FPGAs vs. SoCs
Opti9 Technologies LLC, a North American cloud, data protection, cybersecurity and managed services provider, has acquired Hut 8’s Canadian managed cloud business, expanding its cloud platform and data center presence in Canada. The transaction includes Hut 8’s traditional managed cloud services in Canada, with 10 team members transitioning to Opti9. Hut 8 retains ownership of its Canadian data centers and continues operating its GPU and HPC business. Opti9 and Hut 8 entered a multi-year colocation agreement, allowing Opti9 to operate the acquired cloud infrastructure at Hut 8’s Canadian data centers.
The acquisition builds on Opti9’s 15-year history serving Canadian organizations, strengthening its established cloud platform and managed services footprint. Customers gain access to Opti9’s broader technology platform, spanning private, public and hybrid cloud, data protection, disaster recovery, cybersecurity, compliance and managed services. The expanded footprint provides greater geographic flexibility for channel partners and customers, addressing workload location, data residency, resiliency and disaster recovery within Canada.
Opti9 operates infrastructure across 13 data centers globally, supported by strategic partnerships including being named Veeam’s 2025 Americas Cloud & Service Provider Partner of the Year. CEO TJ Houske stated that M&A remains a key growth strategy, and the acquisition strengthens the platform for channel partners while giving Hut 8 customers access to Opti9’s broader capabilities. Opti9 also provides managed services for security, compliance and emerging AI to help organizations accelerate secure AI adoption.
Microchip, Marelli Target Software-Defined Vehicle Display Systems
Canonical presented the evolution of enterprise open source RISC-V at Hot Chips 2026, detailing how Ubuntu is built and how ISA support is managed to make RISC-V enterprise-ready. Gordan Markuš, Canonical’s Director of Silicon Alliances, argued that an Ubuntu system should behave identically regardless of the processor, applying the same release and maintenance cycle to every architecture. Ubuntu’s Universe archive contains roughly 30,000 packages, while the smaller Main archive has about 3,000 packages with up to 15 years of support via Ubuntu Pro. For RISC-V, the riscv64 baseline moved from RVA20 to RVA23 with Ubuntu 25.10, requiring vector operations, hypervisor support, and bit manipulation for binary portability. Canonical also introduced architecture variants in Ubuntu 25.10, allowing packages built for different CPU features, a pattern planned for future RISC-V profiles. Package coverage on RISC-V has reached about 95% compared with amd64, but test pass rate stands at only 55.1% on an Ubuntu 26.10 snapshot, well below amd64’s 92%, partly due to emulated builds. Canonical is porting Kubernetes and MicroCloud for native RISC-V container orchestration and edge cloud. Open source collaboration through the RISE Project with RISC-V International targets commercial software readiness across mobile, datacenter, and automotive. Native RISC-V runners on Ubuntu 24.04 back a build farm hosted at Scaleway, reporting 24,000 jobs and 700,000 minutes across 116 organizations. With server-class RVA23 hardware expected from multiple vendors in 2026 and 2027, Canonical frames standardized profiles as the key to converging software targeting and achieving enterprise-grade multi-architecture maturity.
QNAP launched its new QSW-M2130 series of managed switches at Computex, targeting businesses needing affordable 2.5GbE client connectivity with 10GbE uplinks. The lineup includes two models: the QSW-M2130-2C2S26T (M2130) and the PoE++ variant QSW-M2130P-2C2S26T. Both are 1U switches offering 24 2.5GbE RJ45 ports and six 10GbE ports for backhaul or high-speed client connections. The M2130 features four combo ports (RJ45/SFP+ shared) plus two dedicated SFP+ ports, totaling 30 usable ports with up to 60Gbps bandwidth via link aggregation. It uses a Realtek RTL9313 MIPS controller with 1GB RAM and runs QNAP’s QSS web management software supporting VLANs, ACLs, and QoS. The PoE M2130P delivers 375W total PoE power, with the first 16 2.5GbE ports limited to 45W and the remaining eight 2.5GbE ports plus four 10GbE RJ45 ports supporting 90W (PoE++). It has two combo ports, two dedicated 10GbE RJ45, and two dedicated SFP+ ports. The PoE model adds extra cooling with three fans versus one in the base switch. Both switches share 240Gbps switching capacity and full QSS management, including PoE-specific controls. As 2.5GbE becomes standard on new PCs, QNAP positions the M2130 series as a cost-effective backbone for enterprise network upgrades, offering flexibility between copper and fiber cabling for uplinks.
ServeTheHome (STH) launched a paid Substack in Q3 2024 to provide deeper analysis on data center infrastructure and technology markets, supplementing its free weekly newsletter that recaps top stories and previews upcoming content. The STH Weekly Newsletter, personally written by founder Patrick Kennedy even while traveling, delivers curated highlights every Saturday to subscribers who lack time for daily site visits. The newsletter format includes a recap of the week, the top five stories chosen by the STH team based on industry impact, and a preview of next week’s content plan. The new Axautik Group Substack (pronounced “exotic”) serves as STH’s analyst arm, targeting a lower-volume, more detailed paid tier for financial communities and data center providers. One featured Substack post compares Qualcomm Cloud AI 100 to LPDDR options from AMD, NVIDIA, and Apple. STH emphasizes no pop-up overlays for newsletter signups, using MailChimp for easy subscription management without selling email addresses. The goal is to keep readers quickly updated on semiconductor, AI chip, memory, SSD, GPU, CPU, and storage markets covered by STH reviews and industry analysis. The newsletter is designed as a simple, reader-friendly tool—not a burden—and is sustained by positive feedback from event attendees.
NXP MCX A5 MCUs, Edge AI for Military Drone Autonomy, Modular Computing for Long-Lifecycle Aerospace Platforms: Embedded Week Insights
AMD announced at IFA Berlin the Threadripper Halo Station, a high-end developer workstation for local AI inference and development, pairing a Threadripper Pro CPU with Instinct MI350P accelerators. Unveiled by AMD Senior VP Jack Huynh, the tower system augments the existing Ryzen AI Halo mini-PC by targeting the upper end of the market with far greater performance. The demo unit features a 96-core Threadripper Pro 9995WX (Zen 5, 192 threads), up to 2TB DDR5 memory, and two liquid-cooled Instinct MI350P cards (each with 144GB HBM3E and 2.3 PFLOPS MXFP8), totaling 288GB GPU memory. AMD notes a path to four accelerators for 576GB. The system uses an ASUS Pro WS WRX90E-SAGE SE motherboard and includes an unidentified SFP networking card, addressing the Ryzen AI Halo’s lack of high-speed networking. All major processors are liquid-cooled via closed-loop radiators. The MI350P cards are custom liquid-cooled versions, a first for AMD, as standard cards are air-cooled for PCIe servers. The Threadripper Halo Station competes directly with NVIDIA’s DGX Station, though AMD relies on multiple PCIe GPUs versus NVIDIA’s single B300 accelerator. No pricing or availability has been announced, but the COTS hardware suggests a relatively quick path to market. The system is expected to run a Linux-based stack, with a Windows option possible given Microsoft’s Project Zenith toolkit spotlight during the keynote.
Everspin, Teledyne HiRel Partner on Mission-Critical MRAM
SomDevices, Celus Bring μSMARC Modules to Design Platform