The Rise of Disaggregated Architectures: Compute & Connectivity evolution for AI and other Industry-Specific Applications

Introduction: Evolution of Compute and Connectivity Capabilities

Over the decades, compute requirements across various segments—such as desktops, enterprise, gaming, and embedded systems—were primarily addressed using DSPs, microcontrollers, CPUs, ASICs, FPGAs, IPs for Accelerators, and GPUs. These components catered to diverse verticals, including enterprise IT, communications, defence, healthcare, and transportation.

However, recent advancements in autonomous systems, AI, robotics, industrial IoT, and telecommunications require lower latency, higher bandwidth, and improved reliability. This has given rise to disaggregated architectures—a paradigm shift that separates hardware and software components into modular, independently manageable units. This approach is highly scalable, cost-optimized, agile, and vendor-flexible, often supporting open systems.

Disaggregated solutions are enabled by modular compute blocks connected through wired (Fiber, Ethernet) or wireless technologies, tailored to specific performance metrics like latency, bandwidth, and security.

Topologies and Segment Applicability

Some of the key product segments where compute and connectivity form the backbone of solutions, and are best enabled by desegregated solutions are as follows

Segments

• Smart Cities (SCT): Distributed edge nodes enable real-time traffic management, surveillance, and environmental monitoring.

• Healthcare (HC): Federated learning ensures privacy-compliant AI models for diagnostics and patient monitoring.

• Extended Reality (XR): Low-latency edge compute supports immersive AR/VR experiences.

• Transportation: Autonomous systems leverage disaggregated architectures for V2X communication and real-time decision-making.

• Industrial Automation, Logistics and Supply Chain, Defence and Aerospace, Retail & eCommerce, Financial Services, Energy and Automation - are other verticals that benefit from distributed edge architecture.

Topologies

The following topologies cater to most segment requirements. Both the distributed and hybrid forms are examples of disaggregation. 

• Centralized: Data center-focused architectures for heavy compute reusing the Hyperscalers and private data centres capabilities

• Distributed: Edge-first architectures addressing latency and data privacy sensitive use cases.

• Hybrid: A combination of client, edge and cloud for cost and performance balance

• Client: A few consumer segments can depend on a Client only centric architecture. Most forms of sensing enabled via sensor hubs, and AI inferencing in addition to compute and connectivity capabilities. 

Solutions

• Investments in heterogeneous compute platforms

• Scalable, energy-efficient connectivity options (fiber, Ethernet, wireless)

• Open standards for interoperability and modularity

• Hybrid cloud-edge architectures

• Collaboration between OEMs, hyperscalers, and startups

• AI model optimization for domain-specific tasks

Disaggregation - Benefits and Challenges

Listed below are some of the benefits, and challenges seen with disaggregation.

Benefits:

• Real-time processing requirements, scalability, and cost-efficiency

• Increasing diversity of AI workloads and user demands

• Energy efficiency with workload-specific hardware

• Enhanced flexibility and vendor diversity

• Locally controlled data privacy 

• Benefits to the 'Mittelstand' economics, less depedance on the Hyperscalers

Challenges:

• Interoperability between systems and vendors.

• Power, cost, and thermal constraints in high-performance systems.

Evolution of Compute

• Transition from single-purpose microprocessors and CPUs to heterogeneous architectures, including GPUs, FPGAs, NPUs, IP Accelerators, and custom ASICs

• AI-focused chips with high TOPS, energy efficiency, and edge-specific optimisations

• Adoption of distributed and federated computing for scalability

• On the semiconductor front, the age of monolithic SOCs are being slowly supplanted by Chiplets. The electrical interconnects having an alternative optical interfaces, essential as we are seeing high data transfers with massive AI Models used.

• also we are seeing increasing adoption of advanced nodes 

Evolution of Connectivity

• Transition from copper (cable and ethernet based) connectivity to high capacity fiber links in the connected space, and towards advanced cellular and non cellular technologies. Satellite communication systems are also getting more ubiquitous. 

Fiber

• Backbone for high-capacity networks like telecom, data centers, and smart cities.

• Key use cases:Backhaul and midhaul in telecom (e.g., 400G, 800G optical fiber links).Data Center Interconnect (DCI) enabling cloud-native architectures.Converged Networks combining fixed-line, wireless, and cable services.

• In addition to Silicon Photonics, direct bandgap materials viz Indium Phosphide, and Galium Arsenide are finding greater traction to enable optical transmission based on the use case. Shorter distance catered by GaAs, and larger distances (telecom and datacom) by Indium Phosphide. 

Ethernet

• Critical for modular data centers and disaggregated RAN (e.g., 100G/400G Ethernet links).

• Supports AI edge-cloud integration and intra-data center communication.

Wireless (5G/6G)

• Enables ultra-reliable low-latency communication (URLLC) for applications like autonomous vehicles and remote healthcare.

• Supports IoT deployments with massive machine-type communication (mMTC).

• Example: 5G Fixed Wireless Access (FWA) provides connectivity in remote locations.

The Future: Next Steps

• Accelerate adoption of disaggregated systems by aligning with open standards.

• Enhance collaboration between hardware and software players.

• Drive innovation in AI-focused semiconductors and connectivity technologies.

• Focus on industry-specific challenges

The rise of disaggregated architectures marks a transformative shift in how industries leverage compute and connectivity to meet the demands of modern workloads. Disaggregation at many levels from SOC, to complete topologies for solutions are seeing increased adoption. By breaking down monolithic systems into modular, scalable components, organizations can achieve greater flexibility, cost-efficiency, and performance optimization.

 From smart cities to autonomous systems, healthcare to industrial automation, disaggregation is enabling customised solutions for diverse segments while addressing challenges like latency, bandwidth, and interoperability. As we move toward a more interconnected world with advancements in 5G, edge computing, Optical Interconects and connectivity, and AI, disaggregated architectures are not just a technical evolution—they are a foundational change, paving the way for more sustainable, agile, and future-ready systems across industries.