A global survey of 857 networking professionals finds less than half (49%) believe the existing networks their organizations rely on will be able to meet the low-latency requirements of artificial intelligence (AI) applications.
Conducted by Dimensional Research on behalf of Broadcom, the survey identifies network congestion (46%), followed by insufficient visibility (39%), congested traffic flows (38%), and latency (37%), as the top networking challenges impacting AI applications. The top three specific shortcomings identified are real-time flow monitoring (47%), real-time infrastructure monitoring (46%), and increased network security (45%)
Additionally, the survey finds that many organizations have yet to apply AI to manage their networks, with only 23% having deployed AI-enabled observability solutions. A full 71% said they don’t fully trust AI to make network operations decisions. In total, 92% of companies are planning to use AI to improve network visibility and resiliency, with just under a quarter (23%) having already deployed some type of AI tool or platform.
Specifically, survey respondents are hopeful AI will lighten their overall stress by improving cloud-to-cloud monitoring (50%), global internet performance visibility (46%), and public cloud network performance visibility (45%), the survey finds.
In effect, there is a paradox building as more organizations build and deploy AI applications, says Jeremy Rossbach, chief technical evangelist for NetOps at Broadcom. Organizations are clearly keen on building and deploying AI applications and agents, but many have yet to assess the networking implications, he added. As a result, the more AI applications an organization deploys the more likely it becomes they will encounter networking issues, said Rossbach.
Many of those issues might have been avoided if networking teams had more aggressively embraced automation, but only 37% have mature automation practices. “Automation is scary,” notes Rossbach.
Ultimately, many of those organizations would be well-advised to encourage networking teams to adopt best DevOps practices that would enable them to manage networking as code, said Rossbach. Unfortunately, the percentage of networking administrators that have coding skills remains fairly limited, while the number of DevOps teams that have added networking expertise is even less, he added.
In the meantime, the survey finds nearly all respondents have some type of cloud strategy in place and are adopting AI. However, 87% also report the Internet and cloud computing environments create network blind spots. A full 95% said they lack visibility into key network segments. More than three quarters (76%) are relying more on third parties for network operations, mainly because they need support for global operations (56%). Other factors for relying more on third parties include outsourcing (46%), lack of expertise (43%), and staff
shortages (40%). The more organizations rely on external networking services, the less visibility they tend to have, noted Rossbach.
Overall, security (49%), followed by reliance on other teams (46%), managing ISPs (43%), budget allocations (41%), lack of expertise (37%), reducing unused capacity (33%), and inadequate solutions (29%) top of the list of networking challenges.
As AI workloads are increasingly distributed it’s only a matter of time before existing networking weaknesses are further exacerbated. The issue now becomes determining how much to allocate to resolve them, hopefully, before they inevitably manifest themselves in ways that no one involved is going to greatly appreciate.
