- Cisco warns that "AI infrastructure debt" leaves networks unready for modern workloads
- A few prepared "Pacesetters" see profits, while other pilots stall due in large part to legacy tech
- Cisco's gear can help open the bottleneck — but so can other vendors'
Organizations deploying AI are facing an "AI infrastructure debt" that prevents realizing full value from AI, according to Cisco.
AI requires bandwidth, security, speed and energy that legacy infrastructure can't keep up with. Just 28% of organizations believe their infrastructure can handle AI workloads, according to the Cisco AI Readiness Index 2025, released in October, based on a survey conducted in August.
On the other hand, companies that have infrastructure and business processes in place to properly exploit AI reap rewards. "The same survey showed that among AI 'pacesetters' — the 13 percent of companies that are fully prepared for AI — 91% are already increasing profitability," Cisco said in a blog post on Thursday. The most AI-ready organizations are four times more likely to move pilots into production and 50 percent more likely to see measurable value.
Some 71% of Pacesetters say their networks are flexible and can scale instantly for AI, compared with 15% of overall survey respondents. And 77% are investing in new data center capacity within the next 12 months, compared with 43% overall, Cisco said.
Pacesetter preparedness
Pacesetters' preparedness goes beyond infrastructure. They take a disciplined, system-level approach. Nearly all (99%) have a defined AI roadmap, compared with 58% overall, with 79% making AI the top investment priority, compared with 24% overall, according to the study.
Cisco touts its own recently introduced products as helping to bridge the infrastructure gap for hyperscalers, neoclouds, service providers and enterprises.
These include the new Nexus 9300 series smart switch, optimized for improved steering of data traffic; as well as Secure AI Factory, developed in partnership with Nvidia; an expanded Cisco AI POD, comprising compute, networking, AI software, security and observability; Nexus Dashboard to centralize managing data center fabrics; Cisco optics, and integration of Splunk and ThousandEyes for observability and digital resilience.
Additionally, last month, Cisco introduced the Cisco 8223 router, which it claims is the first 51.2 terabits per second (Tbps) Ethernet fixed router for connecting AI workloads between data centers in a "scale-out" architecture, connecting data centers that could be hundreds of kilometers apart and making them act as one logical compute unit.
Other components of Cisco's AI infrastructure include products such as AI Defense for end-to-end security and the Secure AI Factory, in partnership with Nvidia, designed to increase security and resilence and trustworthiness.
The legacy infrastructure gap
According to Hyperframe Research analyst Ron Westfall, AI infrastructure debt is a real problem, and Cisco's products can help — though Cisco is not the only vendor with infrastructure chops.
Legacy systems lack "the necessary bandwidth, speed, and energy capacity to handle the exponentially growing demands of modern AI workloads, especially those involving large models and extensive data sets," Westfall said.
He added, "This shortfall means that most organizations cannot efficiently process or secure AI data and deployments, which prevents them from moving beyond small-scale proofs of concept to realizing the promised cross-domain value and automation."
The infrastructure gap is a contributor to AI projects often stalling in the earliest stages, with 70-90% failing to move beyond the pilot phase, he said. Other factors leading to AI failures include skill shortages and poorly conceived plans and strategies.
Cisco's portfolio aligns with helping to bring infrastructure to AI readiness. But Cisco is not alone. He cited Hewlett-Packard Enterprise, Extreme Networks, Arista, Aviz Networks, Meter and Nile Secure as other providers who can contribtue to closing the infrastructure gap.
Telcos and other communications service providers, like other enterprises, can be handicapped in AI deployments by legacy, siloed architectures, Westfall said. "Moreover, the massive, unpredictable data flows generated by training and deploying large AI models can overwhelm existing network and compute capacity, limiting the ability of many telcos to scale AI applications from pilot to full production," he said. The infrastructure gap limits the ability of CSPs to optimize their own network operations for AI workloads and monetize new revenue streams by offering enterprises high-performance AI-as-a-Service platforms, Westfall said.