Blog post · October 28, 2026 · delloro.com
Dell'Oro Group says AI networking is unlikely to converge on one NIC
- About Networking
- Also names center
- Data
- Dell'Oro Group
- Open Compute Project (OCP)
- specialized
- success
- Read the original
Summary
Dell'Oro Group projects the Ethernet Adapter and Smart NIC market to reach $53.6 billion by 2030. The analyst says network adapter evaluation increasingly depends on performance, programmability, and power efficiency, not port speed alone.
Market Overview
Dell'Oro Group projects the Ethernet Adapter and Smart NIC market to reach $53.6 billion by 2030. The analyst said the more important question is how evolving networking architectures will influence demand and competitive positioning across adapter segments.
The analyst said general-purpose NICs, DPUs and IPUs, AI-optimized NICs, and emerging scale-up transport occupy different positions on a performance-programmability spectrum. These are conceptual distinctions, and actual capabilities overlap and depend on implementation.
Segment Analysis
General-purpose NICs
General-purpose NICs serve enterprise, cloud, and data center workloads that require reliable connectivity, compatibility, and cost efficiency. Dell'Oro Group said Ethernet adapters operating at 100 Gbps and below are projected to account for 86% of port shipments in 2030 across front-end networks of enterprises, non-AI cloud providers, and telecom operators.
DPUs and IPUs
Data Processing Units and Infrastructure Processing Units are programmable processors designed to offload networking, security, storage, and other infrastructure functions from host CPUs. Dell'Oro Group said DPUs and IPUs are categorized within the Smart NIC segment, which is projected to account for 83% of NIC port shipments in hyperscalers' front-end networks by 2030.
The analyst said agentic AI may create new opportunities to offload selected networking, security, and infrastructure tasks. The key market question is whether this drives broader adoption beyond established cloud deployments.
AI-optimized NICs
AI-optimized NICs target distributed AI training and accelerator-intensive workloads that require high effective bandwidth, efficient transport, congestion management, and predictable latency. Specialized adapters may accelerate RDMA, supported transport protocols, and selected collective-communication functions.
Indicative power profiles range from approximately 15–30 W for some 400G fixed-function adapters to 40–75 W for some 800G-and-higher designs, and actual consumption varies by implementation. The analyst said the relevant metric is effective networking performance per watt rather than bandwidth alone.
Scale-Up Networking
Scale-out networking connects servers through a fabric, while scale-up networking extends high-performance communication across accelerators, potentially spanning multiple racks. Open Compute Project initiatives, including Ethernet for Scale-Up Networking (ESUN) and Scale-Up Ethernet Transport (SUE-T), reflect this direction.
If these architectures reach deployment at volume, networking functions could move closer to accelerator silicon or into tightly integrated packages. Dell'Oro Group said on-die scale-up networking is not yet covered in its Ethernet Adapter and Smart NIC research.
Key Findings
Dell'Oro Group said higher port speeds alone do not guarantee better workload performance, and improvements in accelerator utilization, job completion time, and total system cost provide stronger evidence of differentiation. The analyst said DPUs and IPUs can consume substantially more power than fixed-function adapters, so the right comparison is net system-level power and performance.
The analyst said standards progress and product announcements matter, but interoperability, customer qualification, and repeatable deployments are stronger indicators of commercial adoption. Product leaders should also assess whether emerging architectures improve power efficiency and thermal density at the system level.
Strategic Outlook
The analyst said the AI networking market is unlikely to converge on a single adapter architecture. General-purpose NICs, DPUs and IPUs, and AI-optimized adapters address different requirements, while emerging on-die scale-up designs may change where connectivity and transport functions reside.
The analyst said the deeper shift involves redistributing networking functions across the computing system. Success will depend on identifying which functions warrant dedicated hardware, which require programmability, and which may integrate into the accelerator or fabric.
Dell'Oro Group's analysis indicates that adapter selection increasingly depends on performance, programmability, and power efficiency at the system level, with demand shaped by where networking functions reside. For Decision Insights readers, these factors bear on infrastructure planning and vendor evaluation. This Analyst Signals brief reflects a neutral, fact-based summary of the original research note.
Our summary. The original keeps its own byline.