As AI/ML implementations accelerate, the stakes are higher for enterprises to secure reliable, scalable connectivity that enables these applications’ real-time functionality. Enterprises are increasingly diversifying their connectivity choices to meet these requirements, with local providers supporting access use cases and wholesale operators providing core connectivity.
Many operators are maximizing diversity, scalability, physical security and other qualities through continuous network expansion and innovation investments to serve these enterprises. So, amid many choices in the market, how can enterprises leverage backbone connectivity to enable AI at scale and unlock business value?
Inner core vs. access layer
A multi-vendor strategy is integral to supporting emerging applications, with many enterprises choosing a mix of connectivity sources. Local providers enable connection to the access layer, allowing enterprises to connect large amounts of branch offices and other sites. Meanwhile, wholesale operators excel in providing reliable connectivity to the high-capacity, long-haul backbone (or inner core) between data centers, cloud regions and strategic Points-of-Presence (PoPs).
When evaluating a wholesale operator’s core connectivity capabilities, enterprises should explore how that operator provides low and predictable latency, enables 10G/100G/400G Wavelengths and facilitates vendor-neutral interconnection. These qualities are essential as bandwidth-consumptive AI applications rely on the fast transfer of massive data capacities across geographically distributed data centers.
Public internet, dedicated lines or both?
When crafting their multi-vendor strategy, enterprises must carefully evaluate the criticality and volume of their traffic, particularly as AI puts more pressure on their networks. Public Internet solutions may be sufficient for smaller sites with fewer employees who mostly use email and Software-as-a-Service tools.
However, dedicated lines that provide Ethernet, Wavelength and Private IP services are suitable for headquarters or data center sites that must contend with high-bandwidth workloads. Enterprises should also remember that real-time AI inferencing and distributed model training require minimal latency and jitter, which public Internet connections may not consistently deliver at scale.
If an enterprise operates within a highly regulated sector, dedicated lines can also help them meet security compliance and other governmental regulations as data sovereignty needs heighten with the rise of AI. This is where enterprises can mix and match connectivity solutions according to their unique requirements at each site.
A balanced multi-vendor strategy ensures they don’t over-engineer low-risk locations. It also allows them to invest in backbone connectivity that provides high performance, uptime and diversity for mission-critical workloads.
Carrier-level AI integration
Internet carriers have traditionally relied on mostly manual network operations. But, like any other level of the telecom ecosystem, operators are also leveraging AI to enhance the visibility of their systems. By strengthening their operations through AI, wholesale providers can improve automation in fault detection, predict outages and reduce mean time to resolution. However, operators must leverage trustworthy data to maximize these benefits.
Machine learning models help wholesale operators achieve proactive issue detection, allowing them to correlate real-time network data and flag operational risks before they affect service. When applied to historical trend analysis, AI can also allow operators to forecast equipment failures and take preemptive action to minimize network disruptions.
No matter how Internet carriers apply AI, the benefits of increased automation trickle down to enterprises, empowering them with more reliable connectivity, improved resilience and faster recovery for supporting their own AI applications.
Physical security, resiliency and redundancy
Cybersecurity has traditionally dominated industry conversations. Now, with increased incidents of natural disasters, geopolitical sabotage and accidental damage, physical network security has become more important than ever before. Amid these unfortunate realities, network redundancy and diversity improve overall resiliency to combat the effects of outages caused by physical disruptions. For example, many operators are building completely underground routes to ameliorate the degradation issues of many legacy cables.
When evaluating wholesale operators, enterprises should not only consider the size of an operator’s network footprint; they should consider its diversity and redundancy. AI workloads have near-zero tolerance for downtime, requiring end-to-end business continuity from the core to the edge, with many enterprises now asking for third and fourth route options instead of just first and second route options.
Diversity and redundancy are enterprises’ best tools in fighting the many-pronged threat of cyber and physical sabotage, allowing them to rely on operators who can switch over AI traffic in the event of disruption or minimize the possibility of a single incident taking out multiple systems simultaneously.
Diverse services, diverse requirements
Amid accelerated AI/ML implementation, one fact has remained since the Internet’s early days. There are no blanket solutions in networking; different applications have different needs. As these requirements diversify, enterprises must carefully evaluate their multi-vendor strategy. AI applications push network operators to be better, faster and make more calculated decisions.
To capitalize on AI’s full benefits, enterprises should take a similarly measured approach when considering connectivity choices to maximize the industry’s newest wave of innovation.
Mattias Fridström, Chief Evangelist