By Jim Shimabukuro (assisted by Grok)
Editor
Over the past two decades, the general architecture of desktops and laptops has remained strikingly consistent, relying on familiar components like motherboards, CPUs, RAM, GPUs, hard drives, and peripheral interfaces such as USB, Bluetooth, and WiFi, all housed within standard cases and driven by conventional operating systems and applications. While these components have seen incremental improvements in speed and efficiency, the core design—rooted in the von Neumann model of sequential processing and separated compute and memory—has persisted largely unchanged.
This stagnation, however, is being challenged by a wave of innovative efforts worldwide, spurred by the rise of agentic AI, which is both enabling new design methodologies and powering novel computing paradigms. From brain-inspired neuromorphic systems to dataflow architectures and task-based operating systems, organizations in the United States, Saudi Arabia, and Israel are exploring groundbreaking approaches to reimagine personal computing.
These initiatives, driven by a mix of visionary startups, academic collaborations, and industry giants, hold significant promise to redefine the desktop and laptop experience, addressing inefficiencies like the von Neumann bottleneck and paving the way for sustainable, AI-driven devices.
In Saudi Arabia, Humain AI, backed by the formidable Public Investment Fund, is pioneering a bold departure from traditional computing with its Humain Horizon Pro laptop and the forthcoming Horizon S variant. Led by CEO Tareq Amin, the company has developed Humain One, a task-based operating system that eliminates the conventional app-centric interface. Instead, it leverages agentic AI to orchestrate complex workflows through natural-language interactions, enabling users to execute multi-step tasks—like researching stocks or summarizing documents—without launching discrete applications.
This “zero-app” paradigm, which Humain claims is 100 times faster than human thought, integrates services seamlessly, relying on a hybrid of local and cloud processing powered by Qualcomm’s Snapdragon X Elite chip. Launched at the Snapdragon Summit in 2025, the Horizon Pro boasts impressive specifications, including 32GB of RAM, a 1TB SSD, and over 18 hours of battery life.
With 500 student-focused units set for distribution and an enterprise subscription model priced 40% below competitors, Humain is positioning itself as a disruptor, particularly in the global South. Its Arabic-first large language model, ALLAM, further enhances its regional appeal. By making hardware “evergreen” through AI-driven updates, Humain’s approach could reshape the $100 billion PC refresh cycle, though the maturity of its operating system remains a critical hurdle.
Across the Atlantic, Efficient Computer, a Carnegie Mellon spinout based in the United States, is tackling the energy inefficiencies inherent in traditional architectures with its Fabric dataflow system and Electron E1 chip. Founded by Nathan, Graham, and Alex, the company has developed a spatial architecture that reimagines programs as parallel “circuits” executed across simple processor arrays, co-locating compute and memory to eliminate the von Neumann bottleneck, which wastes up to 99% of energy on data movement.
This design, supported by the effcc Compiler, allows compatibility with standard programming languages while achieving 100 times the efficiency of low-power CPUs in benchmarks. With $16 million in seed funding secured in 2024 and first silicon taped out by July 2025, Efficient is initially targeting edge devices like IoT and satellites but holds potential for laptops and desktops through low-power modules.
Partnerships, such as with BrightAI, suggest adaptability for scalable agentic AI deployments at the edge. While consumer applications are not yet on the roadmap, Efficient’s focus on sustainability and extreme efficiency positions it as a dark horse for eco-friendly personal computing, provided it can build a robust software ecosystem.
In parallel, NVIDIA, in collaboration with UC Berkeley’s Wireless Research Center, is revolutionizing the way new architectures are designed by harnessing agentic AI for hardware development. Under the leadership of Mark Haoxing Ren, an IEEE Fellow and NVIDIA director, this initiative focuses on automating register-transfer level (RTL) design for post-von Neumann systems.
Agentic tools, powered by large language models, autonomously optimize power, performance, and area metrics, debug verification processes, and synthesize hardware with unprecedented efficiency. Demonstrated through multiple best-paper awards at design automation conferences in 2024 and 2025, these tools have been integrated into NVIDIA’s GPU design flow and partially open-sourced via Berkeley.
While not a consumer product, this work directly impacts desktops and laptops by enabling the creation of custom neuromorphic or dataflow chips, such as those used in Arm or Intel’s efficient system-on-chip designs for laptops. The promise here lies in accelerating the development cycle—cutting design time by up to 50%—which could lead to hybrid architectures blending traditional and novel paradigms. However, proprietary barriers may limit widespread adoption outside NVIDIA’s ecosystem.
Intel, another American heavyweight, is pushing the boundaries of consumer computing with its Loihi 2 neuromorphic chip, developed by Intel Labs in collaboration with institutions like Sandia National Laboratories. Unlike traditional processors, Loihi 2 employs brain-inspired spiking neural networks, using asynchronous, event-driven processing to mimic neurons and synapses.
This in-memory compute approach bypasses the von Neumann bottleneck, achieving ultra-low power consumption in the milliwatt range. Already integrated into the Hala Point system, which simulates 1.15 billion neurons, Loihi 2 is being piloted in wearables and IoT, with plans to embed it in laptops and desktops as part of neural processing unit hybrids in successors to Intel’s Meteor Lake processors.
By enabling real-time, on-device learning and adaptation without cloud reliance, Intel’s approach supports agentic AI applications like predictive user interfaces and always-on assistants, targeting 15% of the edge AI market by 2025. The potential for 1,000-fold efficiency gains is transformative, particularly for sustainable AI-driven PCs, though challenges remain in developing a mature software ecosystem to fully leverage neuromorphic capabilities.
In Israel, NextSilicon is advancing a reconfigurable dataflow architecture with its Maverick-2 chip, led by founders Zibi Klein and Guy Gulik. Backed by $303 million in funding, this system maps algorithms directly to arithmetic logic units, combining RISC-V cores with a dataflow engine to eliminate the overhead of instruction fetch and decode, dedicating 98% of silicon to computation.
Shipping in October 2025 on TSMC’s 5nm process, Maverick-2 targets high-performance computing but includes a companion chip, Arbel, that matches the performance of Intel’s Xeon or AMD’s Zen5 processors. While primarily aimed at data centers, its modularity suggests applicability to high-end desktops and workstations.
NextSilicon’s architecture delivers 2 to 5 times the efficiency of traditional systems for AI and graph-based workloads, making it a candidate for future agentic AI orchestration in personal computing. However, its high power consumption, with a 400W thermal design power, currently limits its suitability for mainstream laptops or desktops.
These global efforts signal a turning point for personal computing, driven by the convergence of agentic AI, energy efficiency demands, and innovative architectural paradigms. Humain AI’s task-based system could redefine user interaction, making apps obsolete and hardware perpetually upgradable. Efficient Computer’s Fabric architecture promises unparalleled efficiency, potentially powering the next generation of eco-conscious devices. NVIDIA and Berkeley’s agentic design tools are accelerating the creation of custom chips, indirectly shaping consumer hardware.
Intel’s neuromorphic advancements offer a path to intelligent, low-power PCs, while NextSilicon’s dataflow approach could elevate high-end desktop performance. Together, these initiatives address critical challenges—energy waste, scalability, and user experience—while leveraging agentic AI to both design and operate these systems.
However, hurdles remain, including ecosystem compatibility, software development, and scaling prototypes to mass production. With the AI PC market projected to reach 114 million units in 2025 and global investment in alternative architectures surging, the next few years could see these innovations coalesce into a new era of desktops and laptops, breaking free from the von Neumann mold and delivering sustainable, intelligent, and user-centric computing.
Additional Context on Potential and Promise
- Overall Trends: These efforts converge on in-situ computing (processing near data) and parallelism to slash energy use by 50-1,000x, vital as AI demands explode (Gartner: 114M AI PCs shipped in 2025). Agentic AI accelerates this by automating design (e.g., NVIDIA’s 25% faster RTL cycles) and enabling intuitive use (Humain’s “conversational PC”). US leads in research/funding ($10B+ VC in 2025 for alt-arch chips), Saudi Arabia in bold consumer bets via PIF ($500B sovereign fund), Israel in dataflow innovation.
- Challenges & Risks: Compatibility with x86/Arm ecosystems; software porting (e.g., Efficient’s compiler); and scaling from prototypes. Successes like Humain’s launch show market hunger, but full “mold-breaking” (e.g., no discrete GPU/RAM) may take 2-5 years.
- Why Now?: Agentic AI lowers barriers—e.g., Berkeley’s tools cut design time 50%—while climate regs push efficiency. Watch for hybrids: Intel’s neuromorphic + NVIDIA agents could yield “thinking” laptops by 2027.
__________
Prompt: Today’s desktops and laptops don’t seem to have progressed much in the last 20 years in terms of general architecture. They still have the same components: motherboard with CPU, RAM, GPU, hard drives, fans, USB ports, Bluetooth, WiFi, modem, soundboard, PSU, case, keyboard, mouse, monitor as well as operating systems and apps. Components have upgraded, but they still remain in the same architecture. With the aid of agentic AI, is anyone, in the US or the world, exploring or developing a new architecture that breaks out of this mold? If yes, please identify them, their names, efforts, successes, organizations, countries, and whatever else you consider important to understand their potential and promise.
[End}
Filed under: Uncategorized |
Educational Technology and Change Journal 
Leave a comment