AI networking spending grows, but enterprise network readiness lags
Network World ·
Spending for AI buildout has reached shocking scale, though much of the spending comes from hyperscalers and neocloud providers. Enterprises are, however, still in the mix, with networking being one of the core elements of spend. Networking vendors are still seeing growth. Dell’Oro Group reports that worldwide enterprise networkin g revenue grew 15% year over year in the first half of 2026 . In an analyst panel at the ONUG AI Networking Summit, IDC analyst Leslie Rosenberg said networking is the fastest-growing category of enterprise AI infrastructure spending. The panel also included Mauricio Sanchez , senior director at Dell’Oro Group, and was moderated by Woo Jin Ho, senior hardware analyst at Bloomberg Intelligence. The discussion covered spending, budgets, readiness, workload placement, and operations. As organizations ramp up to the spending levels forecast for 2028 through 2030, it’s not entirely clear that enterprises are ready. “What we are primarily seeing is that ambition outpaces readiness,” Rosenberg said. The numbers show growth in pockets The numbers behind the AI buildout reveal some interesting trends. Rosenberg said IDC forecasts $623 billion in enterprise AI infrastructure spending in 2026, up 31% from 2025. She said servers make up the majority of the spending. Networking is the fastest-growing category, primarily in the data center. She said IDC also sees a significant refresh in campus networks. Dell’Oro data shows all five major enterprise networking segments posted double-digit gains in the first half of 2026. The segments are network security, branch routing, wireless LAN, campus switching, and enterprise data center switching. Growth accelerated in the second quarter. Revenue growth does not match equipment demand in every segment. Campus switch revenue grew at a double-digit rate while port shipments were nearly flat. In the second quarter, port shipments fell 5% while average selling prices rose 19%. DDR4 (Double Data Rate 4) memory shortages are a major source of pricing pressure. AI infrastructure buildouts have tightened component supply. Wireless LAN growth followed a different path. Higher access point shipments supported revenue as Wi-Fi 7 adoption accelerated. Enterprises also ordered ahead of expected price increases. Dell’Oro expects overall growth to moderate from the first-half pace. Enterprise data center switching shows a similar split. Growth in front-end switching has been concentrated among large organizations. Many of those upgrades support general-purpose servers rather than dedicated AI deployments. Smaller enterprise customers have continued to pull back. Sanchez said the hyperscaler numbers do not translate one-for-one to the enterprise. He pointed to security tooling as an area of fast growth. He said the market for AI system security was about $200 million in 2025. He expects it to reach $8 billion over the next five years. “So, it’s an explosive growth in pockets, but overall, it’s a very slow rising tide when it comes to the overall infrastructure spend,” Sanchez said. Budgets come from a mix of new money and reallocation As Sanchez pointed out, money doesn’t grow on trees; it has to come from somewhere. When it comes to the AI buildout, both analysts said AI budgets are funded through a combination of new money and reprioritization. Rosenberg said some areas have been trimmed to fund AI initiatives. She said AI and security are the two areas least subject to cuts. Vendors offer programs that help stretch budgets. Take-back and circularity programs let customers return equipment, receive credit, and apply it toward a modernization purchase. Rosenberg also cited zero- or low-percent financing programs from HPE , Juniper , and Cisco . With supply chain constraints, some buyers can move ahead with recertified equipment, sometimes at a lower price. Sanchez described a different approach among customers moving workloads out of the public cloud. Some are not buying servers for an on-premises data center. They are renting hardware in private colocation facilities. He said those customers cited the financial packages on offer and uncertainty about how large AI demand will become. Renting or leasing avoids locking in a capital expense over seven years of amortization. “There are creative ways to help finance your technology initiative,” Rosenberg said. AI readiness is weakest in networking and storage While organizations are spending on AI, whether or not they are actually ready to implement AI successfully is another question. Rosenberg said many organizations have a long list of AI initiatives, though few have well-defined, viable use cases. Fewer still have confirmed that the infrastructure can support the initiatives. She said networking and storage are the two areas with the least maturity. For the data center, Rosenberg named network architecture as a priority and said enterprises are getting ready for an 800 gigabit per second fabric. Her second priority was network security for both north-south and east-west traffic. North-south traffic enters and leaves the data center. East-west traffic moves between systems inside it. She said teams must be able to take remediation actions quickly as vulnerabilities reach the network. Sanchez advised enterprises to build a long-term plan before buying. He said there are tools that provide visibility into existing environments. He cautioned against spending on point products or on a platform without a plan. He said CIOs are rushing to check the AI and security boxes and risk costly zigzags on efforts that don’t go in a straight line. Sanchez also described the purchasing pattern his firm sees in its hardware surveys. Enterprises are in a wave of reinvestment after the pandemic. Buyers cite a need for more capacity and more storage to become AI-ready. He said those requirements are not always based on clear technical criteria. Beyond coding, he said the industry is still looking for the next enterprise use case. “I think the vendor community is doing a very good job from the infrastructure perspective to build this fear that, ‘Hey, if you don’t upgrade now, you’re going to be left behind,” Sanchez said. Automation is the long tail ahead of AI for networking There are various reasons why organizations are not ready for AI. “I just finished a whole series just specifically on networking, and one of their biggest challenges is automation,” Rosenberg said. “There’s going to be a very long tail in terms of automation even before they begin to talk about AI for networking.” Customers are consolidating data centers and need to rationalize and automate them. They also need to transfer knowledge to their teams. “When we’re moving into AI for networking, it’s new workflows, new operational best practices, and then once we get into infrastructure as code or network as code, that’s a whole other skill set that many are not prepared for,” Rosenberg said. “So there has to be a lot of training, use of augmented staffing, either from a third party or the vendor.”