96% of Enterprises Say Their Networks Are Slowing Down AI

The findings, captured in Flexential's "2026 State of AI Infrastructure Report: AI-Ready Isn't Infrastructure-Ready," establishes networking as the most widespread operational challenge in enterprise AI, ahead of the power and space constraints that have dominated industry discussion.

GPU deployment in the public cloud rose from 30% in 2024 to 54% this year, while the share of organizations housing AI data there fell from 59% to 26%. As compute and data move into different environments, more traffic has to cross the network to connect them.

At the same time, latency has moved to the top of enterprises' network concerns. Cited by 71% of respondents, up from 32% in 2024 when it ranked last among four performance issues, it now outpaces bandwidth shortages (56%) and unreliable connections (45%). No other metric has moved as much in the survey's three-year history. The shift reflects how AI inference is evolving: as it becomes more geographically distributed, the distance it can sit from users is growing, increasing the importance of regional, network-dense data centers.

Meanwhile, the constraint dominating public debate ranks near the bottom among the teams running these deployments. Difficulty scaling data center space and power affected 25% of respondents, and just 4% identified it as their only issue.

"Power determines where AI can run, and the network determines whether it performs once it's operational," said Ryan Mallory, CEO of Flexential. "Organizations are discovering that a site with megawatts available can still fail them if the fiber, the carriers, and the cloud on-ramps aren't in place. Our customers count on us to get that right, and it comes down to planning, due diligence, and execution to build the right ecosystem inside every data center."

Flexential has been building for that reality. The company recently upgraded its private network backbone to 400 Gbps across several major U.S. metro markets, giving customers more capacity to move AI traffic between compute and data environments. Its interconnection ecosystem directory and marketplace simplifies how customers find partners and connect directly with carriers, cloud on-ramps, and service providers across the Flexential platform.

Now in its third year, the report draws on a survey of more than 350 IT decision-makers at the director level or above at companies with more than $100 million in annual revenue, including respondents from companies exceeding $2 billion. All had knowledge of their organization's AI implementation and related infrastructure buildouts.

The survey also found that:

• 42% of respondents consider access to high-capacity fiber and multiple carrier options as important as power and space when selecting locations for AI workloads, and another 50% count it among several major considerations.
  
  
• Fiber availability, carrier diversity, interconnection options, and low-latency connectivity have been a limiting factor even at sites with favorable power availability for 91% of organizations, with 35% saying this happens very often.
  
  
• Over the past 12 months, fiber availability increased costs for 54% of respondents, delayed AI rollouts for 54%, and changed deployment locations for 50%.
  
  
• Network availability has affected deployment timing or sequencing for 59% of organizations and directly limited where another 29% can run workloads.
  
  
• Software-defined networking adoption reached 58%, 20 percentage points higher than in 2024, while over the past year, third-party colocation for edge processing rose from 37% to 46% and on-premises AI hardware deployment fell from 32% to 24%.

Enterprises are moving AI infrastructure out of their own facilities and into environments they manage through software rather than own. But the workloads they're developing, including customer-facing assistants and autonomous decision engines, will test the network harder than anything running on it today.

"Customers used to ask how much space and power we could give them. Now they want to know what their traffic will do once it leaves the building," said Patrick Doherty, Chief Revenue Officer at Flexential. "They're running GPUs in one environment and keeping data in another, and they need a partner whose network can close that gap. Interconnection has moved to the center of every infrastructure conversation we're having."

Flexential's full report, including additional findings on power, compute, tariffs, and ROI, can be accessed here.

About Flexential
Flexential empowers the IT journey of the most complex businesses by offering customizable hybrid IT solutions designed for today's demanding high-density computing requirements. With colocation, cloud, interconnection, data protection, and professional services, the FlexAnywhere^® platform anchors our services in over 40 data centers across 18 highly connected markets on a scalable 100+ Gbps private network backbone. Flexential solutions are strategically engineered to meet the most stringent challenges in security, compliance, and resiliency. Experience the power of IT flexibility and how we enable digital transformation at www.flexential.com.

Media Contact
Alison Brooker
[email protected]

Christian Rizzo
Gregory for Flexential
[email protected]

SOURCE Flexential