What’s next for Ethernet, AI networks, and other key technologies in 2025

By Steve Douglas / Spirent Communications

Predicting the future is always a dicey proposition. Unless that is, you work with the people actively building and testing that future, giving you unique visibility. That’s the position we find ourselves in at Spirent. We’re as excited as everyone else to see what the coming year holds for data center networks, artificial intelligence (AI), and other evolving industry trends.

We’re working with equipment manufacturers, service providers, chipset makers, and other industry stakeholders to prepare new networking innovations.

Based on hundreds of testing engagements worldwide, here are the most significant emerging trends we’ll be following in the coming months:

Ethernet has started taking the lead in AI networking
Network efficiency dictates profitability for companies building AI data centers, connecting billions of dollars worth of specialized processing units (CPUs) to support ever-larger AI models. Given the unique requirements of AI workloads, packet loss of just 1% can translate to 33% lower performance—and a huge hit to return on investment (ROI). This is why data center operators have historically built xPU networking fabrics using lossless InfiniBand. Many operators, however, would prefer standardized, widely available Ethernet if it could meet AI networking requirements. Thanks to protocols like remote direct memory access (RDMA) over Converged Ethernet version 2 (RoCEv2) and emerging industry specifications like Ultra-Ethernet Transport, now it can.

This year, the market will start formally reflecting this shift. Dell’Oro Group recently revised its high-speed Ethernet (HSE) forecast upwards by more than 30%, noting that demand for AI networks is accelerating the transition from 400G port speeds to 800G and beyond. The analyst predicts that HSE port shipments will increase by 50% compound annual growth rate through 2028 when Ethernet becomes the dominant AI network technology.

Enterprises will shift AI investment away from the public cloud
Another key trend at the intersection of networking and AI is that this year, many enterprises will start repatriating AI infrastructure out of the public cloud in favor of private and hybrid deployments. This change comes in response to mounting regulations governing AI data sovereignty, security, and privacy. By adopting private and sovereign deployment models, enterprises and government agencies can maintain tighter control over their data and ongoing infrastructure costs.   

This trend will open new opportunities for telecom, cloud, and networking companies to provide scalable, secure network infrastructure-as-a-service (NIaaS) solutions to these customers. Telcos can combine physical presence in customer locations with extensive experience delivering multitenant networking and hosting services. They’re in a prime position to help simplify enterprise compliance and governance by offering localized AI networking services on demand.

Networks will grow increasingly automated and powered by agentic AI
The previous two trends focused on the evolving ways that networks can enable AI. But what about AI-enabling networks? This year, we will also see significant growth in this area.

Traditional service providers and large network operators have recently invested in network automation, seeking to reduce manual effort and errors, improve quality, and lower costs. In 2025, AI will become a “force multiplier” for these efforts, making network automation frameworks more intelligent and adaptive. New agentic AI technologies will be able to handle more complex autonomous and semi-autonomous network operations. Relatedly, we expect operators to increase investment in automated testing technologies to assist with training AI agents and validating and governing their interventions.

Operational resiliency will matter more than ever
Several major network outages have occurred in recent years, such as the CrowdStrike outage, which knocked out millions of computer systems worldwide and caused billions of dollars in losses. Recognizing the central role of IT networks in mission-critical industries like telecom, transportation, healthcare, and banking, governments and regulators are now issuing a wide range of new mandates and recommendations to improve operational resilience.

This year, look for organizations across these industries and others to implement more comprehensive network testing to better guard against cyberattacks and faulty updates caused by human error. These initiatives will often be part of broader efforts to expand continuous testing within continuous integration/continuous delivery (CI/CD) software toolchains and enable more frequent and automated testing.  

5G’s second half will kick off in earnest
Believe it or not, this year marks the midpoint of the planned 10-year release cycle for 5G. Through 2024, much industry news focused on operators’ monetization struggles and sluggish 5G network investments. But the 5G story is far from over. Based on our work with telecoms and vendors, we expect a significant uptick in 5G activity in 2025.

There are several reasons why telcos and industry observers are so optimistic. First, less than 10% of operators have launched 5G Standalone (5G SA), meaning that 90% still haven’t tapped into 5G’s more advanced capabilities for driving value and differentiation. Additionally, 5G was always intended to evolve via iterative releases, and many of its most potent capabilities have only recently become available through recent 3GPP updates. The pieces are in place for a significant boost in 5G momentum this year.