96% of Enterprises Say Their Networks Are Slowing Down AI

Flexential’s survey of more than 350 senior IT decision-makers found that 96% experienced at least one network-related issue affecting AI workloads in the past 12 months. Latency was cited by 71%, up from 32% in 2024, while fiber constraints increased costs and delayed rollouts for 54% of respondents each. Flexential said it upgraded its private network backbone to 400 Gbps across several major U.S. metro markets; the survey highlights infrastructure bottlenecks but is not evidence of company financial performance.

Analysis

The investable signal is a possible shift in AI infrastructure value from securing power and land toward controlling connectivity between compute, data, and users. If workloads remain split across clouds, colocation sites, and enterprise data environments, interconnection-rich locations and fiber capacity could capture more of the next dollar of infrastructure spend; power availability alone may be less differentiating at already-networked sites. Potential beneficiaries include network-dense colocation and interconnection operators such as Equinix and Digital Realty, plus fiber and optical-network suppliers—but only where bookings, pricing, and utilization confirm the demand is monetizable.

This is not yet evidence of sector-wide earnings acceleration. The report is vendor-sponsored, self-reported, and does not establish network spending per workload or Flexential’s share of the opportunity. Latency is also an architectural problem: caching, workload placement, and software-defined networking could reduce traffic or defer physical buildout. Fiber permitting, construction costs, and carrier access may limit how quickly reported pain converts into revenue.

Near term, the survey alone is unlikely to justify a broad trade. Over 1–3 months, watch colocation/interconnection bookings, fiber lead times and pricing, and hyperscaler or enterprise capex commentary. Over 6–18 months, sustained distributed inference would support a structural premium for connected capacity. The thesis weakens if AI deployment slows, workload consolidation reduces cross-site traffic, or network-related investment fails to show up in operator revenue and returns. The contrarian point: power remains necessary; the market may be underpricing connectivity, but this survey does not prove power-oriented assets are obsolete.

Key Decisions for Investors

• Treat this as a watch signal, not a standalone catalyst trade. Track interconnection bookings, utilization, pricing, and capex disclosures from colocation operators before adding exposure.
• If those indicators strengthen, consider a relative-value long in network-dense colocation/interconnection exposure, including Equinix or Digital Realty, versus a basket selected mainly for power availability. Avoid a naked short on power-focused developers: power remains a prerequisite, and the article does not establish that connectivity displaces it.
• Monitor fiber and optical-network suppliers as a second-order beneficiary group, but require evidence of order growth, backlog conversion, and acceptable returns; reported shortages can raise costs and delay deployments without improving supplier profits.
• Falsify the thesis if operator commentary shows flat interconnection demand or pricing, if AI capex plans are cut, or if software and workload-placement changes reduce cross-environment traffic. Reassess after the next relevant earnings and capex updates.